Search

Accurate precipitation intensity forecasting is crucial for effective flood management and early warning systems. This study evaluates the performances of convolutional neural network (CNN) and long short-term memory (LSTM) models in predicting hourly precipitation intensity using data from Sainte Catherine de la Jacques Cartier station near Québec City. The models predict precipitation levels from one to six hours ahead, which are categorized into slight, moderate, heavy, and very heavy precipitation intensities. Our methodology involved gathering hourly precipitation data, defining input combinations for multistep ahead forecasting, and employing CNN and LSTM models. The performances of these models were assessed through qualitative and quantitative evaluations. The key findings reveal that the LSTM model excelled in the short-term (1HA to 2HA) and long-term (3HA to 6HA) forecasting, with higher R2 (up to 0.999) and NSE values (up to 0.999), while the CNN model was more computationally efficient, with lower AICc values (e.g., −16,041.1 for 1HA). The error analysis shows that the CNN demonstrated higher precision in the heavy and very heavy categories, with a lower relative error, whereas the LSTM performed better for the slight and moderate categories. The LSTM outperformed the CNN in minor- and high-intensity events, but the CNN exhibited a better performance for significant precipitation events with shorter lead times. Overall, both models were adequate, with the LSTM providing better accuracy for extended forecasts and the CNN offering efficiency for immediate predictions, highlighting their complementary roles in enhancing early warning systems and flood management strategies. [ABSTRACT FROM AUTHOR]

Inundation maps that show water depths that occur in the event of a flood are essential for protection. Especially information on timings is crucial. Creating a dynamic inundation map with depth data in temporal resolution is a major challenge and is not possible with physical models, as these are too slow for real-time predictions. To provide a dynamic inundation map in real-time, we developed a data-driven multi-step inundation forecast system for fluvial flood events. The forecast system is based on a convolutional neural network (CNN), feature-informed dense layers, and a recursive connection from the predicted inundation at timestep t as a new input for timestep t + 1. The forecast system takes a hydrograph as input, cuts it at desired timesteps (t), and outputs the respective inundation for each timestep, concluding in a dynamic inundation map with a temporal resolution (t). The prediction shows a Critical Success Index (CSI) of over 90%, an average Root Mean Square Error (RMSE) of 0.07, 0.12, and 0.15 for the next 6 h, 12 h, and 24 h, respectively, and an individual RMSE value below 0.3 m, for all test datasets when compared with the results from a physically based model. [ABSTRACT FROM AUTHOR]

Inundation maps that show water depths that occur in the event of a flood are essential for protection. Especially information on timings is crucial. Creating a dynamic inundation map with depth data in temporal resolution is a major challenge and is not possible with physical models, as these are too slow for real-time predictions. To provide a dynamic inundation map in real-time, we developed a data-driven multi-step inundation forecast system for fluvial flood events. The forecast system is based on a convolutional neural network (CNN), feature-informed dense layers, and a recursive connection from the predicted inundation at timestep t as a new input for timestep t + 1. The forecast system takes a hydrograph as input, cuts it at desired timesteps (t), and outputs the respective inundation for each timestep, concluding in a dynamic inundation map with a temporal resolution (t). The prediction shows a Critical Success Index (CSI) of over 90%, an average Root Mean Square Error (RMSE) of 0.07, 0.12, and 0.15 for the next 6 h, 12 h, and 24 h, respectively, and an individual RMSE value below 0.3 m, for all test datasets when compared with the results from a physically based model.

Abstract Background Dynamical mathematical models defined by a system of differential equations are typically not easily accessible to non-experts. However, forecasts based on these types of models can help gain insights into the mechanisms driving the process and may outcompete simpler phenomenological growth models. Here we introduce a friendly toolbox, SpatialWavePredict, to characterize and forecast the spatial wave sub-epidemic model, which captures diverse wave dynamics by aggregating multiple asynchronous growth processes and has outperformed simpler phenomenological growth models in short-term forecasts of various infectious diseases outbreaks including SARS, Ebola, and the early waves of the COVID-19 pandemic in the US. Results This tutorial-based primer introduces and illustrates a user-friendly MATLAB toolbox for fitting and forecasting time-series trajectories using an ensemble spatial wave sub-epidemic model based on ordinary differential equations. Scientists, policymakers, and students can use the toolbox to conduct real-time short-term forecasts. The five-parameter epidemic wave model in the toolbox aggregates linked overlapping sub-epidemics and captures a rich spectrum of epidemic wave dynamics, including oscillatory wave behavior and plateaus. An ensemble strategy aims to improve forecasting performance by combining the resulting top-ranked models. The toolbox provides a tutorial for forecasting time-series trajectories, including the full uncertainty distribution derived through parametric bootstrapping, which is needed to construct prediction intervals and evaluate their accuracy. Functions are available to assess forecasting performance, estimation methods, error structures in the data, and forecasting horizons. The toolbox also includes functions to quantify forecasting performance using metrics that evaluate point and distributional forecasts, including the weighted interval score. Conclusions We have developed the first comprehensive toolbox to characterize and forecast time-series data using an ensemble spatial wave sub-epidemic wave model. As an epidemic situation or contagion occurs, the tools presented in this tutorial can facilitate policymakers to guide the implementation of containment strategies and assess the impact of control interventions. We demonstrate the functionality of the toolbox with examples, including a tutorial video, and is illustrated using daily data on the COVID-19 pandemic in the USA.

Background: Dynamical mathematical models defined by a system of differential equations are typically not easily accessible to non-experts. However, forecasts based on these types of models can help gain insights into the mechanisms driving the process and may outcompete simpler phenomenological growth models. Here we introduce a friendly toolbox, SpatialWavePredict, to characterize and forecast the spatial wave sub-epidemic model, which captures diverse wave dynamics by aggregating multiple asynchronous growth processes and has outperformed simpler phenomenological growth models in short-term forecasts of various infectious diseases outbreaks including SARS, Ebola, and the early waves of the COVID-19 pandemic in the US. Results: This tutorial-based primer introduces and illustrates a user-friendly MATLAB toolbox for fitting and forecasting time-series trajectories using an ensemble spatial wave sub-epidemic model based on ordinary differential equations. Scientists, policymakers, and students can use the toolbox to conduct real-time short-term forecasts. The five-parameter epidemic wave model in the toolbox aggregates linked overlapping sub-epidemics and captures a rich spectrum of epidemic wave dynamics, including oscillatory wave behavior and plateaus. An ensemble strategy aims to improve forecasting performance by combining the resulting top-ranked models. The toolbox provides a tutorial for forecasting time-series trajectories, including the full uncertainty distribution derived through parametric bootstrapping, which is needed to construct prediction intervals and evaluate their accuracy. Functions are available to assess forecasting performance, estimation methods, error structures in the data, and forecasting horizons. The toolbox also includes functions to quantify forecasting performance using metrics that evaluate point and distributional forecasts, including the weighted interval score. Conclusions: We have developed the first comprehensive toolbox to characterize and forecast time-series data using an ensemble spatial wave sub-epidemic wave model. As an epidemic situation or contagion occurs, the tools presented in this tutorial can facilitate policymakers to guide the implementation of containment strategies and assess the impact of control interventions. We demonstrate the functionality of the toolbox with examples, including a tutorial video, and is illustrated using daily data on the COVID-19 pandemic in the USA. [ABSTRACT FROM AUTHOR]

The stock market’s volatility, noise, and information overload necessitate efficient prediction methods. Forecasting index prices in this environment is complex due to the non-linear and non-stationary nature of time series data generated from the stock market. Machine learning and deep learning have emerged as powerful tools for identifying financial data patterns and generating predictions based on historical trends. However, updating these models in real-time is crucial for accurate predictions. Deep learning models require extensive computational resources and careful hyperparameter optimization, while incremental learning models struggle to balance stability and adaptability. This paper proposes a novel hybrid bidirectional-LSTM (H.BLSTM) model that combines incremental learning and deep learning techniques for real-time index price prediction, addressing these scalability and memory challenges. The method utilizes both univariate time series derived from historical index prices and multivariate time series incorporating technical indicators. Implementation within a real-time trading system demonstrates the method’s effectiveness in achieving more accurate price forecasts for major stock indices globally through extensive experimentation. The proposed model achieved an average mean absolute percentage error of 0.001 across nine stock indices, significantly outperforming traditional models. It has an average forecasting delay of 2 s, making it suitable for real-time trading applications.

Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) are revolutionizing hydrology, driving significant advancements in water resource management, modeling, and prediction. This review synthesizes cutting-edge developments, methodologies, and applications of AI-ML-DL across key hydrological processes. By critically evaluating these techniques against traditional models, we highlight their superior ability to capture complex, nonlinear relationships and adapt to diverse environments. We further explore AI applications in precipitation forecasting, evapotranspiration estimation, groundwater dynamics, and extreme event prediction (floods, droughts, and compound events), showcasing their timely potential in addressing critical water-related challenges. A particular emphasis is placed on Explainable AI (XAI) and transfer learning as essential tools for improving model transparency and applicability, enabling broader stakeholder trust and cross-regional adaptability. The review also addresses persistent challenges, including data limitations, computational demands, and model interpretability, proposing solutions that integrate emerging technologies like quantum computing, the Internet of Things (IoT), and interdisciplinary collaboration. This review charts a strategic course for future research and practice by bridging AI advancements with practical hydrological applications. Our findings highlight the importance of embracing AI-driven approaches for next-generation hydrological modeling and provide actionable understandings for researchers, practitioners, and policymakers. As hydrology faces escalating challenges due to human-induced climate change and growing water demands, the continued evolution of AI-integrated models and innovations in data handling and stakeholder engagement will be imperative. In conclusion, the findings emphasize the critical role of AI-driven hydrological modeling in addressing global water challenges, including climate change adaptation, sustainable water resource management, and disaster risk reduction.

As climate change increases the occurrences of extreme weather events, like flooding threaten humans more often. Hydrodynamic models provide spatially distributed water depths as inundation maps, which are essential for flood protection. Such models are not computationally efficient enough to deliver results before or during an event. To ensure real-time prediction, we developed a feature-informed data-driven forecast system (FFS), which interpreted the forecasting process as an image-to-image translation, to predict the maximum water depth for a fluvial flood event. The FFS combines a convolutional neural network (CNN) and feature-informed dense layers to allow the integration of the distance to the river of each cell to be predicted into the FFS. The aim is to ensure training for the whole study area on a standard computer. A hybrid database with pre-simulated scenarios is used to train, validate, and test the FFS. The FFS delivers predictions within seconds making a real-time application possible. The quality of prediction compared with the results of the pre-simulated physically-based model shows an average root mean square error (RMSE) of 0.052 for thirty-five test events, and of 0.074 and 0.141 for two observed events. Thus, the FFS provides an efficient alternative to hydrodynamic models for flood forecasting. [ABSTRACT FROM AUTHOR]

Intraday trading is popular among traders due to its ability to leverage price fluctuations in a short timeframe. For traders, real-time price predictions for the next few minutes can be beneficial for making strategies. Real-time prediction is challenging due to the stock market's non-stationary, complex, noisy, chaotic, dynamic, volatile, and non-parametric nature. Machine learning models are considered effective for stock forecasting, yet, their hyperparameters need tuning with the latest market data to incorporate the market's complexities. Usually, models are trained and tested in batches, which smooths the correction process and speeds up the learning. When making intraday stock predictions, the models should forecast for each instance in contrast to the whole batch and learn simultaneously to ensure high accuracy. In this paper, we propose a strategy based on two different learning approaches: incremental learning and Offline–Online learning, to forecast the stock price using the real-time stream of the live market. In incremental learning, the model is updated continuously upon receiving the stock's next instance from the live-stream, while in Offline-Online learning, the model is retrained after each trading session to make sure it incorporates the latest data complexities. These methods were applied to univariate time-series (established from historical stock price) and multivariate time-series (considering historical stock price as well as technical indicators). Extensive experiments were performed on the eight most liquid stocks listed on the American NASDAQ and Indian NSE stock exchanges, respectively. The Offline–Online models outperformed incremental models in terms of low forecasting error. [ABSTRACT FROM AUTHOR]

Increasingly scarce water resources and growing global populations have exacerbated the problems of water quality in river systems and freshwaters in general. New monitoring methodologies and tools to democratize access to water quality information are needed if we are to reach ambitious societal objectives such as the UN Sustainable Development Goals and the European Green Deal. Here we present a cloud-based system for producing publicly accessible real time water quality forecasts coupled to novel biosensor technology. Short term forecasts of water quality impairments, e.g., as cyanobacteria blooms, sediment plumes and toxic pollution incidents are increasingly relevant both to citizens and stakeholders. Here, we present a new cloud based system that utilizes satellite data to produce real time forecasts of flow and water quality using a chain of dynamic catchment-scale models at multiple locations in a river network. We demonstrate this new system using two case studies: the River Thames and the Essex Colne River (United Kingdom). These rivers are key water supply sources for London and South-East England, respectively and are of high interest to recreational water users. We show how the new system can predict and forecast water quality, estimate toxicity and connect to citizen science observations using an App (www.aquascope.com) for information synthesis and delivery.

AbstractThis study proposes an integrated hydrometeorological modelling framework approach and methodology for flash flood Early Warning Systems in the Chugoku region of Japan. Unprecedented rainfall-induced hydrometeorological disasters and flash floods are increasingly occurring worldwide. Comprehensive efforts are conducted to simultaneously combine multiple disciplines into integrated modelling framework approaches to reduce disaster resilience. This enables more accurate hindcasts, reanalyses, real-time forecasts or nowcasts for flash floods. This study integrates proposed hydrological calibration approach with meteorological input. Two real-time rainfall forecasts by the Weather Research and Forecasting model forced by the Atmospheric Reanalysis v5 (ERA5) and the Japanese 55-year Reanalysis (JRA55) were used as input data to the hydrological model ensemble parameterized previously. This approach was applied to seven major rivers to evaluate river discharges real-time forecasts accuracy during the Heavy Rainfall Event of July 2018. Long lead-times of up to 29 h with a satisfactory reproducible range of Nash-Sutcliffe Efficiency were obtained using both meteorological forecast for all rivers cumulatively. This indicates that the proposed integrated hydrometeorological approach enables accurate flash flood real-time forecasting for this event. Similarly, the joint hydrometeorological approach enables framework for development of real-time flash food forecasting application in Japan and presumably worldwide.

With the continuous development of civil aviation industry in recent years, the demand for flight training has been increasing and the flight training safety requirements have been improving. To address the problem that the mining of flight training data is not deep enough and the research of flight training data is delayed, the study proposes a flight training data prediction model based on incremental learning to achieve real-time flight training data prediction and ensure flight training safety. The model firstly uses a small amount of data to generate a base model; secondly, on this base model, a specific amount of data is input to the model for learning according to a certain time frequency to continuously optimize the base model; finally, the model is compared with several models for experiments. The experimental results show that compared with the traditional model, the model has high prediction accuracy and good real-time performance in flight training data prediction, which can better ensure flight training safety. [ABSTRACT FROM AUTHOR]

This study marks a significant step toward the future development of river discharges forecasted in real time for flash flood early warning system (EWS) disaster prevention frameworks in the Chugoku region of Japan, and presumably worldwide. To reduce the disaster impacts with EWSs, accurate integrated hydrometeorological real-time models for predicting extreme river water levels and discharges are needed, but they are not satisfactorily accurate due to large uncertainties. This study evaluates two calibration methods with 7 and 5 parameters using the hydrological Cell Distributed Runoff Model version 3.1.1 (CDRM), calibrated by the University of Arizona's Shuffled Complex Evolution optimization method (SCE-UA). We hypothesize that the proposed ensemble hydrological parameter calibration approach can forecast similar future events in real time. This approach was applied to seven major rivers in the region to obtain hindcasts of the river discharges during the Heavy Rainfall Event of July 2018 (HRE18). This study introduces a new historical extreme rainfall event classification selection methodology that enables ensemble-averaged validation results of all river discharges. The reproducibility metrics obtained for all rivers cumulatively are extremely high, with Nash–Sutcliffe efficiency values of 0.98. This shows that the proposed approach enables accurate predictions of the river discharges for the HRE18 and, similarly, real-time forecasts for future extreme rainfall-induced events in the Japanese region. Although our methodology can be directly reapplied only in regions where observed rainfall data are readily available, we suggest that our approach can analogously be applied worldwide, which indicates a broad scientific contribution and multidisciplinary applications. [ABSTRACT FROM AUTHOR]

Present tsunami warning systems have been specialized for earthquake-generated tsunamis, but rapidly evaluating the tsunamis caused by volcanic eruptions and/or volcanic sector collapses remains a challenge. In this study, we applied a numerical model to the 2018 Anak Krakatau tsunami event, which was generated by the sector collapse, investigated a tsunami prediction skill by the model, and developed a real-time forecasting method based on a pre-computed database for future tsunamis accompanied by such eruption of Anak Krakatau. The database stores spatiotemporal changes in water surface level and flux, which are simulated under various collapse scenarios, for confined areas in the vicinity of potential source. The areas also cover the locations of observation stations that are virtually placed on uninhabited island surrounding the source area. During an actual volcanic tsunami event, a tsunami is expected to be observed at the observation stations. For real-time tsunami forecasting, the most suitable scenarios to reproduce the observed waveforms are searched quickly in the database. The precomputed results under the identified scenarios are further provided as input for rapid tsunami propagation simulation. Therefore, an effective real-time forecasting can be conducted to densely populated coastal areas located at a considerable distance from the source, such as the coasts of Java and Sumatra. The forecasting performance was examined by applying the method for three hypothetical collapse scenarios assuming different sliding directions. We demonstrated that the tsunamis along the coasts were successfully forecasted. Moreover, we showed that the combination of a pre-computed database and the existence of observation stations near the source area was able to produce appropriate tsunami forecasting for the coastal area even in a volcanic event.

Real-time algorithms have to adapt and adjust to new incoming patterns to provide timely and accurate responses. This paper presents a new distributed forecasting algorithm for streaming time series called StreamWNN. StreamWNN starts with an offline stage in which a forecasting model based on tuples of information fusion is created with historical data. In particular, this model consists of the fusion of patterns composed of past values of the time series with the future values of their k-nearest neighbors. Afterwards, streaming data starts to arrive. The model is incrementally updated in the online stage using a buffer with streaming data that more accurately matches the current model patterns. The model can be updated daily, monthly, quarterly or based on error thresholds. The methodology has been applied to Spanish electricity demand time series providing more accurate results when the model is updated incrementally. The best error results are obtained with the daily update of the model, resulting in an error between 2% and 3.5% depending on the prediction horizon. The model provides better error results than other algorithms. • StreamWNN is a real-time forecasting algorithm for time series received in streaming. • The prediction model is based on the fusion of relevant information related to the nearest neighbors. • Incremental learning deals with the change of distribution of streaming data. • StreamWNN predicts accurately Spanish electricity consumption in real-time. [ABSTRACT FROM AUTHOR]

Subsurface inclusions are one of the most common defects that affect the inner quality of continuous casting slabs. This increases the defects in the final products and increases the complexity of the hot charge rolling process and may even cause breakout accidents. The defects are, however, hard to detect online by traditional mechanism-model-based and physics-based methods. In the present paper, a comparative study is carried out based on data-driven methods, which are only sporadically discussed in the literature. As a further contribution, a scatter-regularized kernel discriminative least squares (SR-KDLS) model and a stacked defect-related autoencoder back propagation neural network (SDAE-BPNN) model are developed to improve the forecasting performance. The scatter-regularized kernel discriminative least squares is designed as a coherent framework to directly provide forecasting information instead of low-dimensional embeddings. The stacked defect-related autoencoder back propagation neural network extracts deep defect-related features layer by layer for a higher feasibility and accuracy. The feasibility and efficiency of the data-driven methods are demonstrated through case studies based on a real-life continuous casting process, where the imbalance degree drastically vary in different categories, showing that the defects are timely (within 0.01 ms) and accurately forecasted. Moreover, experiments illustrate the merits of the developed scatter-regularized kernel discriminative least squares and stacked defect-related autoencoder back propagation neural network methods regarding the computational burden; the F1 scores of the developed methods are clearly higher than common methods. [ABSTRACT FROM AUTHOR]

Copyright of Water Conservancy Science & Techonlogy & Economy is the property of Water Conservancy Science & Technology & Economy Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Summary: Macroeconomic data are subject to data revisions. Yet, the usual way of generating real‐time density forecasts from Bayesian Vector Autoregressive (BVAR) models makes no allowance for data uncertainty from future data revisions. We develop methods of allowing for data uncertainty when forecasting with BVAR models with stochastic volatility. First, the BVAR forecasting model is estimated on real‐time vintages. Second, the BVAR model is jointly estimated with a model of data revisions such that forecasts are conditioned on estimates of the 'true' values. We find that this second method generally improves upon conventional practice for density forecasting, especially for the United States. [ABSTRACT FROM AUTHOR]

As climate change increases the occurrences of extreme weather events, like flooding threaten humans more often. Hydrodynamic models provide spatially distributed water depths as inundation maps, which are essential for flood protection. Such models are not computationally efficient enough to deliver results before or during an event. To ensure real-time prediction, we developed a feature-informed data-driven forecast system (FFS), which interpreted the forecasting process as an image-to-image translation, to predict the maximum water depth for a fluvial flood event. The FFS combines a convolutional neural network (CNN) and feature-informed dense layers to allow the integration of the distance to the river of each cell to be predicted into the FFS. The aim is to ensure training for the whole study area on a standard computer. A hybrid database with pre-simulated scenarios is used to train, validate, and test the FFS. The FFS delivers predictions within seconds making a real-time application possible. The quality of prediction compared with the results of the pre-simulated physically-based model shows an average root mean square error (RMSE) of 0.052 for thirty-five test events, and of 0.074 and 0.141 for two observed events. Thus, the FFS provides an efficient alternative to hydrodynamic models for flood forecasting.

Many urban areas in tropical Southeast Asia, e.g., Bangkok in Thailand, have recently been experiencing unprecedentedly intense flash floods due to climate change. The rapid flood inundation has caused extremely severe damage to urban residents and social infrastructures. In addition, urban Southeast Asia usually has inadequate capacities in drainage systems, complicated land use patterns, and a large vulnerable population in limited urban areas. To reduce the urban flood risk and enhance the resilience of vulnerable urban communities, it has been of essential importance to develop real-time urban flood forecasting systems for flood disaster prevention authorities and the urban public. This paper reviewed the state-of-the-art models of real-time forecasting systems for urban flash floods. The real-time system basically consists of the following subsystems, i.e., rainfall forecasting, drainage system modelling, and inundation area mapping. This paper summarized the recent radar data utilization methods for rainfall forecasting, physical-process-based hydraulic models for flood inundation prediction, and data-driven artificial intelligence (AI) models for the real-time forecasting system. This paper also dealt with available technologies for modelling, e.g., digital surface models (DSMs) for the finer urban terrain of drainage systems. The review indicated that an obstacle to using process-based hydraulic models was the limited computational resources and shorter lead time for real-time forecasting in many urban areas in tropical Southeast Asia. The review further discussed the prospects of data-driven AI models for real-time forecasting systems. [ABSTRACT FROM AUTHOR]

The stock market's volatility, noise, and information overload necessitate efficient prediction methods. Forecasting index prices in this environment is complex due to the non-linear and non-stationary nature of time series data generated from the stock market. Machine learning and deep learning have emerged as powerful tools for identifying financial data patterns and generating predictions based on historical trends. However, updating these models in real-time is crucial for accurate predictions. Deep learning models require extensive computational resources and careful hyperparameter optimization, while incremental learning models struggle to balance stability and adaptability. This paper proposes a novel hybrid bidirectional-LSTM (H.BLSTM) model that combines incremental learning and deep learning techniques for real-time index price prediction, addressing these scalability and memory challenges. The method utilizes both univariate time series derived from historical index prices and multivariate time series incorporating technical indicators. Implementation within a real-time trading system demonstrates the method's effectiveness in achieving more accurate price forecasts for major stock indices globally through extensive experimentation. The proposed model achieved an average mean absolute percentage error of 0.001 across nine stock indices, significantly outperforming traditional models. It has an average forecasting delay of 2 s, making it suitable for real-time trading applications. [ABSTRACT FROM AUTHOR]

Subsurface inclusions are one of the most common defects that affect the inner quality of continuous casting slabs. This increases the defects in the final products and increases the complexity of the hot charge rolling process and may even cause breakout accidents. The defects are, however, hard to detect online by traditional mechanism-model-based and physics-based methods. In the present paper, a comparative study is carried out based on data-driven methods, which are only sporadically discussed in the literature. As a further contribution, a scatter-regularized kernel discriminative least squares (SR-KDLS) model and a stacked defect-related autoencoder back propagation neural network (SDAE-BPNN) model are developed to improve the forecasting performance. The scatter-regularized kernel discriminative least squares is designed as a coherent framework to directly provide forecasting information instead of low-dimensional embeddings. The stacked defect-related autoencoder back propagation neural network extracts deep defect-related features layer by layer for a higher feasibility and accuracy. The feasibility and efficiency of the data-driven methods are demonstrated through case studies based on a real-life continuous casting process, where the imbalance degree drastically vary in different categories, showing that the defects are timely (within 0.01 ms) and accurately forecasted. Moreover, experiments illustrate the merits of the developed scatter-regularized kernel discriminative least squares and stacked defect-related autoencoder back propagation neural network methods regarding the computational burden; the F1 scores of the developed methods are clearly higher than common methods.

• A real-time multi-step forecasting system is developed in container freight market forecasting. • A data preprocessing strategy with multiple rolling decomposition is proposed for streaming data. • Accurate and stable forecasting results are provided under the impact of COVID-19. • The scientific experiments and comprehensive evaluation module are established. • The developed real-time system is validated well in different datasets. The management and decision-making of the shipping market rely heavily on the accurate forecasting of the China Containerized Freight Index (CCFI); however, this is still a challenging task. Furthermore, CCFI is influenced by many factors and exhibits complex nonlinear characteristics, which may cause the evolution of its distributions over time, i.e., the concept drift issue. However, previous studies mainly focused on the simple application of single data preprocessing and machine learning, while ignoring the significance of real-time and multi-step forecasting, which may not effectively deal with the concept drift issue and result in unsatisfactory results. Additionally, some studies disregarded the data leakage problem caused by the one-time decomposition of all datasets. Therefore, a novel real-time multi-step forecasting system with multiple rolling decomposition strategy based on streaming data is designed. It includes data preprocessing, forecasting, and evaluation modules. A three-stage data preprocessing strategy is proposed in the first module to overcome the single preprocessing defect. It can eliminate the adverse effects of outliers, extract main features, and reduce system complexity. Moreover, the forecasting and evaluation modules are designed to enhance the forecasting performance and verify the system's validity. Considering the obvious concept drift problem of CCFI during the Coronavirus Disease 2019 (COVID-19) epidemic, datasets from the COVID-19 epidemic are selected for experimental verification to verify the system's ability to solve this problem. Empirical studies demonstrate the superiority of the developed system over the compared models in terms of universality and robustness. Therefore, the proposed system is an effective prediction method for the containerized freight market in the post epidemic period and an alternative for forecasting other streaming data with the concept drift issue. [ABSTRACT FROM AUTHOR]

In this study, an improved version of the outlier robust extreme learning machine (IORELM) is introduced as a new method for multi-step-ahead hourly air temperature forecasting. The proposed method was calibrated and used to estimate the hourly air temperature for one to ten hours in advance after finding its most optimum values (i.e., orthogonality effect, activation function, regularization parameter, and the number of hidden neurons). The results showed that the proposed IORELM has an acceptable degree of accuracy in predicting hourly temperatures ten hours in advance (R = 0.95; NSE = 0.89; RMSE = 3.74; MAE = 1.92).

In this study, an improved version of the Extreme Learning Machine, namely the Improved Weighted Regularization ELM (IWRELM), is proposed for hourly precipitation forecasting that is multi-steps ahead. After finding the optimal values of the proposed method, including the number of hidden neurons, the activation function, the weight function, the regularization parameter, and the effect of orthogonality, the IWRELM model was calibrated and validated. Thereafter, the calibrated IWRELM model was used to estimate precipitation up to ten hours ahead. The results indicated that the proposed IWRELM (R = 0.9996; NSE = 0.9993; RMSE = 0.015; MAE = 0.0005) has acceptable accuracy in short-term hourly precipitation forecasting up to ten hours ahead.

Multiple factors including rainfall and underlying surface conditions make river basin real-time flood forecasting very challenging. It is often necessary to use real-time correction techniques to modify the forecasting results so that they reach satisfactory accuracy. There are many such techniques in use today; however, they tend to have weak physical conceptual basis, relatively short forecast periods, unsatisfactory correction effects, and other problems. The mechanism that affects real-time flood forecasting error is very complicated. The strongest influencing factors corresponding to this mechanism affect the runoff yield of the forecast model. This paper proposes a feedback correction algorithm that traces back to the source of information, namely, modifies the watershed runoff. The runoff yield error is investigated using the principle of least squares estimation. A unit hydrograph is introduced into the real-time flood forecast correction; a feedback correction model that traces back to the source of information. The model is established and verified by comparison with an ideal model. The correction effects of the runoff yield errors are also compared in different ranges. The proposed method shows stronger correction effect and enhanced prediction accuracy than the traditional method. It is also simple in structure and has a clear physical concept without requiring added parameters or forecast period truncation. It is readily applicable in actual river basin flood forecasting scenarios. HIGHLIGHTS From the perspective of system theory, an RSCR method to modify production flow is proposed.; This method uses the system response curve feedback of the production flow in each period to correct the production flow in each period.; The validity and practicability of the method are proved by numerical simulation experiments with different random distribution errors.;

Regardless of the existing governmental and public preventive actions for surveillance and controlling the air quality in several regions of the Chennai city in India, the air quality does not meet the desired standard. In this regard, this study employs an ARMA/ARIMA modelling approach for forecasting Respirable Suspended Particulate Matter (RSPM), Sulphur dioxide (SO2) and Nitrogen dioxide (NO2) concentration for three most polluted sites in Chennai city. A total of nine univariate linear stochastic models have been developed, three for each of the stations which includes one for each of the specific pollutants in order to forecasts the concentration of each pollutant. The evaluation of the model statistics R2 values and index of agreement values evince that a significant level of real-time forecasting of the pollutants can be achieved by employing the developed ARMA/ARIMA models. Moreover, the comparisons of actual air pollutant concentration have been carried out with the permissible limit as prescribed by the National ambient air quality standards (NAAQS) of India for assessing the level of pollution of all three locations.

Many urban areas in tropical Southeast Asia, e.g., Bangkok in Thailand, have recently been experiencing unprecedentedly intense flash floods due to climate change. The rapid flood inundation has caused extremely severe damage to urban residents and social infrastructures. In addition, urban Southeast Asia usually has inadequate capacities in drainage systems, complicated land use patterns, and a large vulnerable population in limited urban areas. To reduce the urban flood risk and enhance the resilience of vulnerable urban communities, it has been of essential importance to develop real-time urban flood forecasting systems for flood disaster prevention authorities and the urban public. This paper reviewed the state-of-the-art models of real-time forecasting systems for urban flash floods. The real-time system basically consists of the following subsystems, i.e., rainfall forecasting, drainage system modelling, and inundation area mapping. This paper summarized the recent radar data utilization methods for rainfall forecasting, physical-process-based hydraulic models for flood inundation prediction, and data-driven artificial intelligence (AI) models for the real-time forecasting system. This paper also dealt with available technologies for modelling, e.g., digital surface models (DSMs) for the finer urban terrain of drainage systems. The review indicated that an obstacle to using process-based hydraulic models was the limited computational resources and shorter lead time for real-time forecasting in many urban areas in tropical Southeast Asia. The review further discussed the prospects of data-driven AI models for real-time forecasting systems.

In this paper we survey a number of recent empirical findings regarding the usefulness of including diffusion indexes in dynamic Nelson-Siegel (DNS) type models used to predict the term structure of interest rates (see e.g., Diebold and Li (2007) and Diebold and Rudebusch (2013)). We also survey various empirical methods used in the construction of DNS models, and used to specify and estimate diffusion index augmented DNS models. In particular, we review (sparse) principal component analysis, factor augmented autoregression, and various dimension reduction, variable selection, machine learning, and shrinkage methods, such as the least absolute shrinkage operator (lasso), the elastic net, and independent component analysis, among others. Finally, we discuss the importance of using real-time data in contexts where datasets are subject to revision; and we compare and contrast the use of targeted versus un-targeted specification methods when including diffusion indexes in DNS type prediction models. Interestingly, as noted in Swanson and Xiong (2018a, 2018b), the usefulness of diffusion indexes is crucially dependent upon whether real-time data are used or not. Specifically, when real-time data are used to estimate the weights in di usion indexes, it is found that relatively few “data rich” models that use big data are preferred to simpler DNS models, post 2010. Instead, pure DNS models that rely only on historical interest rate data deliver mean square error “best” forecasts. However, when data are not real-time, diffusion indexes always have marginal predictive content for interest rates. Moreover, it is clear that in more volatile interest rate regimes, such as prior to 2010, machine learning and related methods have much to offer, regardless of the type of dataset used in their construction.

Abstract The Goddard Earth Observing System composition forecast (GEOS‐CF) system is a high‐resolution (0.25°) global constituent prediction system from NASA's Global Modeling and Assimilation Office (GMAO). GEOS‐CF offers a new tool for atmospheric chemistry research, with the goal to supplement NASA's broad range of space‐based and in‐situ observations. GEOS‐CF expands on the GEOS weather and aerosol modeling system by introducing the GEOS‐Chem chemistry module to provide hindcasts and 5‐days forecasts of atmospheric constituents including ozone (O3), carbon monoxide (CO), nitrogen dioxide (NO2), sulfur dioxide (SO2), and fine particulate matter (PM2.5). The chemistry module integrated in GEOS‐CF is identical to the offline GEOS‐Chem model and readily benefits from the innovations provided by the GEOS‐Chem community. Evaluation of GEOS‐CF against satellite, ozonesonde and surface observations for years 2018–2019 show realistic simulated concentrations of O3, NO2, and CO, with normalized mean biases of −0.1 to 0.3, normalized root mean square errors between 0.1–0.4, and correlations between 0.3–0.8. Comparisons against surface observations highlight the successful representation of air pollutants in many regions of the world and during all seasons, yet also highlight current limitations, such as a global high bias in SO2 and an overprediction of summertime O3 over the Southeast United States. GEOS‐CF v1.0 generally overestimates aerosols by 20%–50% due to known issues in GEOS‐Chem v12.0.1 that have been addressed in later versions. The 5‐days forecasts have skill scores comparable to the 1‐day hindcast. Model skills can be improved significantly by applying a bias‐correction to the surface model output using a machine‐learning approach.

The Goddard Earth Observing System composition forecast (GEOS‐CF) system is a high‐resolution (0.25°) global constituent prediction system from NASA's Global Modeling and Assimilation Office (GMAO). GEOS‐CF offers a new tool for atmospheric chemistry research, with the goal to supplement NASA's broad range of space‐based and in‐situ observations. GEOS‐CF expands on the GEOS weather and aerosol modeling system by introducing the GEOS‐Chem chemistry module to provide hindcasts and 5‐days forecasts of atmospheric constituents including ozone (O3), carbon monoxide (CO), nitrogen dioxide (NO2), sulfur dioxide (SO2), and fine particulate matter (PM2.5). The chemistry module integrated in GEOS‐CF is identical to the offline GEOS‐Chem model and readily benefits from the innovations provided by the GEOS‐Chem community. Evaluation of GEOS‐CF against satellite, ozonesonde and surface observations for years 2018–2019 show realistic simulated concentrations of O3, NO2, and CO, with normalized mean biases of −0.1 to 0.3, normalized root mean square errors between 0.1–0.4, and correlations between 0.3–0.8. Comparisons against surface observations highlight the successful representation of air pollutants in many regions of the world and during all seasons, yet also highlight current limitations, such as a global high bias in SO2 and an overprediction of summertime O3 over the Southeast United States. GEOS‐CF v1.0 generally overestimates aerosols by 20%–50% due to known issues in GEOS‐Chem v12.0.1 that have been addressed in later versions. The 5‐days forecasts have skill scores comparable to the 1‐day hindcast. Model skills can be improved significantly by applying a bias‐correction to the surface model output using a machine‐learning approach. Plain Language Summary: Accurate forecasting of the compostion of the atmosphere is important for a variety of applications, including air pollution mitigation, support of satellite and other remote‐sensing observations, and research applications. Producing such forecasts is computationally expensive due to the complexity of atmospheric chemistry, which interacts with weather on all scales. Here we present the NASA Goddard Earth Observing System composition forecast (GEOS‐CF) system, which produces global forecasts of major atmospheric constituents such as ozone (O3), nitrogen dioxide (NO2), and fine particulate matter (PM2.5). On a daily basis, the model tracks the atmospheric concentrations of more than 250 chemical species in more than 55 million model grid cells, computing the interactions between those species using the state‐of‐the‐science GEOS‐Chem chemistry model. We present an in‐depth evaluation of the GEOS‐CF model through comparison against independent observations. We show how the model captures many observed features of atmospheric composition, such as spatio‐temporal variations in air pollution due to changes in pollutant emissions, weather, and chemistry. We also highlight some of the model deficiencies, for example, with respect to the simulation of aerosol particles. Finally, we demonstrate how surface observations and model data can be combined using machine learning to provide improved local air quality forecasts. Key Points: GEOS‐CF is a new modeling system that produces global forecasts of atmospheric composition at 25 km2 horizontal resolutionGEOS‐CF model output is freely available and offers a new tool for academic researchers, air quality managers, and the public [ABSTRACT FROM AUTHOR]

Using real-time data from 1985:Q1 to 2017:Q3 and simple vector autoregression (VAR) models, we show that there is a substantial payoff in combining credit supply indicators with house prices for forecasting real economic activity in the USA. Consistent with the findings in the literature, we show that the forecasts from a bivariate VAR model of real activity and credit conditions dominate the forecasts from a univariate model of real activity. The most interesting finding of the paper is that once real house price growth is added to the bivariate VAR model of real activity and credit conditions, the forecasting performance improves significantly. The forecasts from the model that contains credit supply indicator and real house price growth are also competitive with the forecasts of the Survey of Professional Forecasters. These results provide further evidence in support of the recent theoretical and empirical research on the dynamic relationship between housing market and credit conditions and its role in explaining real economic activity fluctuations in the USA. [ABSTRACT FROM AUTHOR]

The accuracy of two-dimensional hydrodynamic models (2D models) is improved when high-resolution Digital Elevation Models (DEMs) are used. However, the entailed high spatial discretisation results in excessive computational expenses, thus prohibiting their implementation in real-time forecasting especially at a large scale. This paper presents a sub-model approach that adapts 1D static models to tailor high-resolution 2D model grids relevant to specified targets, such that the tailor-made 2D hydrodynamic sub-models yield fast processing without significant loss of accuracy via a GIS-based multi-scale simulation framework. To validate the proposed approach, model experiments were first designed to separately test the impact of two outcomes (i.e., the reduced computational domains and the optimised boundary conditions) towards final 2D prediction results. Then, the robustness of the sub-model approach was evaluated by selecting four focus areas with distinct catchment terrain morphologies as well as distinct rainfall return periods of 1–100 years. The sub-model approach resulted in a 45–553 times faster processing with a 99% reduction in the number of computational cells for all four cases; the goodness of fit regarding predicted flood extents was above 0.88 of F2, flood depths yield Root Mean Square Errors (RMSE) below 1.5 cm and the discrepancies of u- and v-directional velocities at selected points were less than 0.015 ms−1. As such, this approach reduces the 2D models' computing expenses significantly, thus paving the way for large-scale high-resolution 2D real-time forecasting. [ABSTRACT FROM AUTHOR]

This study uses a deep learning approach to forecast ozone concentrations over Seoul, South Korea for 2017. We use a deep convolutional neural network (CNN). We apply this method to predict the hourly ozone concentration on each day for the entire year using several predictors from the previous day, including the wind fields, temperature, relative humidity, pressure, and precipitation, along with in situ ozone and NO2 concentrations. We refer to a history of all observed parameters between 2014 and 2016 for training the predictive models. Model-measurement comparisons for the 25 monitoring sites for the year 2017 report average indices of agreement (IOA) of 0.84–0.89 and a Pearson correlation coefficient of 0.74–0.81, indicating reasonable performance for the CNN forecasting model. Although the CNN model successfully captures daily trends as well as yearly high and low variations of the ozone concentrations, it notably underpredicts high ozone peaks during the summer. The forecasting results are generally more accurate for the stations located in the southern regions of the Han River, the result of more stable topographical and meteorological conditions. Furthermore, through two separate daytime and nighttime forecasts, we find that the monthly IOA of the CNN model is 0.05–0.30 higher during the daytime, resulting from the unavailability of some of the input parameters during the nighttime. While the CNN model can predict the next 24 h of ozone concentrations within less than a minute, we identify several limitations of deep learning models for real-time air quality forecasting for further improvement. [ABSTRACT FROM AUTHOR]

In this study, we use a deep convolutional neural network (CNN) to develop a model that predicts ozone concentrations 24 h in advance. We have evaluated the model for 21 continuous ambient monitoring stations (CAMS) across Texas. The inputs for the CNN model consist of meteorology (e.g., wind field, temperature) and air pollution concentrations (NO x and ozone) from the previous day. The model is trained for predicting next-day, 24-hour ozone concentrations. We acquired meteorological and air pollution data from 2014 to 2017 from the Texas Commission on Environmental Quality (TCEQ). For 19 of the 21 stations in the study, results show that the yearly index of agreement (IOA) is above 0.85, confirming the acceptable accuracy of the CNN model. The results also show the model performed well, even for stations with varying monthly trends of ozone concentrations (specifically CAMS-012, located in El-Paso, and CAMS-013, located in Fort Worth, both with IOA=0.89). In addition, to ensure that the model was robust, we tested it on stations where fewer meteorological variables are monitored. Although these stations have fewer input features, their performance is similar to that of other stations. However, despite its success at capturing daily trends, the model mostly underpredicts the daily maximum ozone, which provides a direction for future study and improvement. As this model predicts ozone concentrations 24 h in advance with greater accuracy and computationally fewer resources, it can serve as an early warning system for individuals susceptible to ozone and those engaging in outdoor activities. [ABSTRACT FROM AUTHOR]

Real-time crime forecasting is important. However, accurate prediction of when and where the next crime will happen is difficult. No known physical model provides a reasonable approximation to such a complex system. Historical crime data are sparse in both space and time and the signal of interests is weak. In this work, the authors first present a proper representation of crime data. The authors then adapt the spatial temporal residual network on the well represented data to predict the distribution of crime in Los Angeles at the scale of hours in neighborhood-sized parcels. These experiments as well as comparisons with several existing approaches to prediction demonstrate the superiority of the proposed model in terms of accuracy. Finally, the authors present a ternarization technique to address the resource consumption issue for its deployment in real world. This work is an extension of our short conference proceeding paper [Wang, B., Zhang, D., Zhang, D. H., et al., Deep learning for real time Crime forecasting, 2017, arXiv: 1707.03340]. [ABSTRACT FROM AUTHOR]

Watershed is the basic unit for studying different hydrologic processes. Flow forecasting in a watershed is dependent upon the rainfall. The effect of erroneous rainfall prediction is a source of uncertainty in flow forecasting. In this study, a model is proposed to improve the flow forecasting on real-time basis. The proposed model has three components (1) Adaptive Grey Rainfall Forecasting Model, (2) Rainfall-Runoff Model and (3) Fuzzy Updating Model. The proposed forecasting model is tested for lead periods of 1 to 3 h with hourly rainfall and discharge data. In this study, four different cases using combination of three models are discussed and the results are compared. The study has been carried out on three Indian watersheds namely Banha, Harsul and Khadakohol. The performance of the model is measured using Nash Sutcliffe Efficiency (E), Correlation Coefficient (r), Error of Peak Discharge (EQpeak) and Error of Time to Peak (ETpeak). It is observed that the case with integration of all three models performed good with a forecasting efficiency of E = 0.950, 0.861, 0.564; and r = 0.991, 0.972, 0.897 for lead-1, 2, 3 respectively for Banha watershed. For Harsul watershed, E = 0.898, 0.704, 0.367; and r = 0.985, 0.949, 0.834 for lead-1, 2, 3 respectively. For Khadakohol watershed, E = 0.968, 0.932, 0.787; and r = 0.994, 0.987, 0.951 for lead-1, 2, 3 respectively. EQpeak is less than 10% for lead-1 for most of the events and increased slightly for lead-2 and lead-3. ETpeak is 0 h for all lead periods of the three watersheds. The proposed model is useful for farmers in planning and monitoring of water resources for crop management and helps in taking necessary actions during heavy rains and floods. [ABSTRACT FROM AUTHOR]

The accuracy of two-dimensional hydrodynamic models (2D models) is improved when high-resolution Digital Elevation Models (DEMs) are used. However, the entailed high spatial discretisation results in excessive computational expenses, thus prohibiting their implementation in real-time forecasting especially at a large scale. This paper presents a sub-model approach that adapts 1D static models to tailor high-resolution 2D model grids relevant to specified targets, such that the tailor-made 2D hydrodynamic sub-models yield fast processing without significant loss of accuracy via a GIS-based multi-scale simulation framework. To validate the proposed approach, model experiments were first designed to separately test the impact of two outcomes (i.e., the reduced computational domains and the optimised boundary conditions) towards final 2D prediction results. Then, the robustness of the sub-model approach was evaluated by selecting four focus areas with distinct catchment terrain morphologies as well as distinct rainfall return periods of 1–100 years. The sub-model approach resulted in a 45–553 times faster processing with a 99% reduction in the number of computational cells for all four cases; the goodness of fit regarding predicted flood extents was above 0.88 of F2, flood depths yield Root Mean Square Errors (RMSE) below 1.5 cm and the discrepancies of u- and v-directional velocities at selected points were less than 0.015 ms−1. As such, this approach reduces the 2D models’ computing expenses significantly, thus paving the way for large-scale high-resolution 2D real-time forecasting.

This paper introduces the rescaled representation of VAR models (R-VARs) and demonstrates its application in forecasting mixed-frequency macroeconomic data. We develop the model, illustrate how to implement it, and derive the asymptotic properties of the estimates. We show that R-VARs provide reliable estimates of the prediction error bands while maintaining the precision of the point forecasts. We illustrate these features by comparing it to a mixed-frequency Bayesian VAR model, the leading alternative in the existing literature. [ABSTRACT FROM AUTHOR]

Accurate and reliable prediction of smoke development process is critical for building fire protection design. Although modern computers allow us to use sophisticated models, e.g. CFD, to simulate certain fire scenarios, the simulation time is often drastically longer than zone models. This paper applies a real-time forecasting method, Ensemble Kalman filter (EnKF) algorithm, to a zone model, CFAST, for the simulation of a full-scale fire experiment in a two-story building structure. The results of the CFAST simulation with and without the EnKF algorithm are compared to the experimental data. It shows that the EnKF algorithm can significantly improve the accuracy of the CFAST simulation of the smoke layer temperature and height. To solve the problem of filter divergence and further reduce the computational cost, a general statistical interference method, the bootstrap method, is added to the EnKF algorithm. It shows that the bootstrap method improves the accuracy with fewer ensemble members and less computing time, thus even achieving a real-time forecasting on a personal computer. A sensitivity analysis is also conducted in nine case studies for demonstrating practical engineering applications of the real-time forecasting process. [ABSTRACT FROM AUTHOR]

Time series data can be defined as a chronological sequence of observations on a variable of interest. A streaming time series is a time series that arrives continuously at high speed and has a data distribution that may change over time. Streaming time series data usually comes from electronic devices such as sensors and many of the applications dealing with streaming data in Industry 4.0 require real-time responses. Performing real-time forecasting offers the possibility to consider new types of patterns in the incoming streaming data, which is not possible when working with batch models. This paper presents a new approach to detect novelties and anomalies in real-time using a nearest-neighbors based forecasting algorithm. The algorithm works with an offline base model that is updated as stream data arrives following an incremental learning approach. It detects unknown patterns called novelties and anomalies. Novelties are included in the model in an online way and anomalies trigger an alarm as they present unexpected behaviors that need to be specifically analyzed. The algorithm has been tested with Spanish electricity demand data. Results show that the prediction errors obtained when the model is updated considering novelties and anomalies are lower than the errors obtained when the model is not updated. Thus, the model adjusts in real-time to the new patterns of data providing accurate errors and real-time predictions. [ABSTRACT FROM AUTHOR]

Financial markets are highly complex and volatile; thus, accurate forecasting of such markets is vital to make early alerts about crashes and subsequent recoveries. People have been using learning tools from diverse fields such as financial mathematics and machine learning to make trustworthy forecasting on such markets. However, the accuracy of such techniques had not been adequate until artificial neural network frameworks such as long short-term memory (LSTM) were utilized. Moreover, making accurate real-time forecasting, also known as nowcasting, of financial time series is highly subjective to the LSTM's architecture in use and the procedure of training it. Herein, we forecast financial markets in real-time by training a dual version of LSTM which forecasts only one time step at each iteration so that the forecast for this iteration will be in the input for the next iteration. Semi-convergence is a prominent issue in a recurrent LSTM setup as the error could propagate through iterations; however, the duality of this LSTM aids in dwindling this issue. Especially, we employ one LSTM to find the best number of epochs associated with the least loss and train the second LSTM only through that many epochs to make forecasting. We treat the current forecast as a part of the training set for the next forecast and train the same LSTM. While classic ways of training cause more error when the forecast is made further away through the test period, our approach offers superior accuracy as the training increases when it proceeds through the testing period. The forecasting accuracy of our approach is validated using three time series from each of the three diverse financial markets: stock, cryptocurrency, and commodity. The results are compared with those of a single LSTM, an extended Kalman filter, and an autoregressive integrated moving average model. • Classic tools may be incapable of forecasting some complex financial markets. • We use a sequentially trained many-to-one dual-LSTM to make real-time predictions. • One LSTM is employed to reveal the number of epochs associated with the least loss. • The other LSTM makes the forecasting after training with the best number of epochs. • The current forecast is included in the training set to forecast for the next time step. [ABSTRACT FROM AUTHOR]