Search

Abstract With the continuous development and combined application of cloud computing and artificial intelligence, some new methods have emerged to reduce task execution time for training neural network models in a cloud-edge collaborative environment. The most attractive method is neural network model segmentation. However, many factors affect the segmentation point, such as resource allocation, system energy consumption, load balancing, and network Bandwidth allocation. Some segmentation methods consider the shortest task execution time, which ignores the utilization of resources at the edge and can result in resource waste. Additionally, these factors are difficult to measure, which presents a challenge in calculating the best segmentation point to achieve the goal of maximum resource utilization and minimum task execution time. To solve this problem, this paper proposes a cloud-edge collaborative task scheduling method based on model segmentation (CECMS). This method first analyzes the factors affecting the segmentation point of the model and then obtains accurate factors that affect the segmentation point calculation through the pre-execution method. Furthermore, a multi-objective solution algorithm is improved to calculate the optimal model segmentation point. And tasks are separately offloaded to the edge and cloud based on the optimal model segmentation point. Finally, the experiments are conducted to verify the effectiveness of this method. Finally, the effectiveness of the CECMS method was verified through simulation experiments. Compared with the Dynamic Adaptive DNN Surgery (DADS) method and an adaptive DNN inference acceleration framework algorithm with end–edge–cloud collaborative computing algorithm (ADC), CECMS achieves the same effectiveness as DADS and ADC in optimizing task execution time by comprehensively considering the utilization of edge resources and minimizing task execution time, while also effectively ensuring resource utilization.

Abstract Accurate estimation of evaporation is of great significance for understanding regional drought, and managing and applying limited water resources in dryland. However, the application of the traditional estimation approaches is limited due to the lack of required meteorological parameters or experimental conditions. In this study, a novel hybrid model was proposed to estimate the monthly pan Ep in dryland by integrating long short-term memory (LSTM) with grey wolf optimizer (GWO) algorithm and Kendall-τ correlation coefficient, where the GWO algorithm was employed to find the optimal hyper-parameters of LSTM, and Kendall-τ correlation coefficient was used to determine the input combination of meteorological variables. The model performance was compared to the performance of other methods based on the evaluation metrics, including root mean squared error (RMSE), the normalized mean squared error (NMSE), the mean absolute error (MAE), the mean absolute percentage error (MAPE), and Nash–Sutcliffe coefficient of efficiency (NSCE). The results indicated that the optimal input meteorological parameters of the hybrid Kendall-τ-GWO-LSTM models are the monthly average temperature, the minimum air temperature, the maximum air temperature, the minimum values of RMSE, NMSE, MAE, and MAPE are 38.28, 0.20, 26.62, and 19.96%, and the maximum NSCE is 0.89, suggesting that the hybrid Kendall-τ-GWO-LSTM exhibit better model performance than the other hybrid models. Thus, the hybrid Kendall-τ-GWO-LSTM model was highly recommended for estimating pan Ep with limited meteorological information in dryland. The present investigation provides a novel method to estimate the monthly pan Ep with limited meteorological variables in dryland by coupling a deep learning model with meta-heuristic algorithms and the data preprocessing techniques.

Abstract The stability of artificial sand-binding vegetation determines the success or failure of restoration of degraded ecosystem, accurately evaluating the stability of artificial sand-binding vegetation can provide evidence for the future management and maintenance of re-vegetated regions. In this paper, a novel data-driven evaluation model was proposed by combining statistical methods and a neural network model to evaluate the stability of artificial sand-binding vegetation in the southeastern margins of the Tengger Desert, where the evaluation indexes were selected from vegetation, soil moisture, and soil. The evaluation results indicate that the stability of the artificially re-vegetated belt established in different years (1956a, 1964a, 1981a, and 1987a) tend to be stable with the increase of sand fixation years, and the artificially re-vegetated belts established in 1956a and 1964a have almost the same stability, but the stability of the artificially re-vegetated belt established in 1981a and 1987a have a significant difference. The evaluation results are reliable and accurate, which can provide evidence for the future management of artificial sand-binding vegetation.

Abstract The sustainability of artificial sand-binding vegetation is determined by the water balance between evapotranspiration (ET) and precipitation in desert regions. Consequently, accurately estimating ET is a critical prerequisite for determing the types and spatial distribution of artificial vegetation in different sandy areas. For this purpose, a novel hybrid estimation model was proposed to estimate monthly ET by coupling the deep learning long short term memory (LSTM) with variational mode decomposition (VMD) and whale optimization algorithm (WOA) (i.e., VMD-WOA-LSTM) to estimate the monthly ET in the southeast margins of Tengger Desert. The superiority of LSTM was selected due to its capability of automatically extracting the nonlinear and nonstationary features from sequential data, WOA was employed to optimize the hyperparameters of LSTM, and VMD was used to extract the intrinsic traits of ET time series. The estimating results of VMD-WOA-LSTM has been compared with actual ET and estimation of other hybrid models in terms of standard performance metrics. The results reveale that VMD-WOA-LSTM provide more accurate and reliable estimating results than that of LSTM, the support vector machine (SVM), and the variants of those models. Therefore, VMD-WOA-LSTM could be recommended as an essential auxiliary method to estimate ET in desert regions.

The air quality index (AQI) indicates the short-term air quality situation and changing trend of the city, which includes six air pollutants: PM2.5, PM10, CO, NO2, SO2 and O3. Due to the diversity of pollutants and the fluctuation of single pollutant time series, it is a challenging task to find out the main pollutants and establish an accurate forecasting system in a city. Previous studies primarily focused on enhancing either forecasting accuracy or stability and failed to analyze different air pollutants at length, leading to unsatisfactory results. In this study, a model selection forecasting system is proposed that consists of data mining, data analysis, model selection, and multi-objective optimized modules and effectively solves the problems of air pollutants monitoring. The proposed system employed fuzzy C-means cluster algorithm to analyze 13 original AQI series, and fuzzy comprehensive evaluation is used to find out the main air pollutants in each city. And then multiple artificial neural networks are used to forecast the main air pollutants for each category and find the optimal models. Finally, the modified multi-objective optimization algorithm is used to optimize the parameters of optimal models and model selection to obtain final forecasting values from optimal hybrid models. The experiment results of datasets from 13 cities in the Beijing–Tianjin–Hebei Urban Agglomeration demonstrated that the proposed system can simultaneously obtain efficient and reliable data for air quality monitoring.

Seed bank persistence plays a significant role in maintaining the sustainability of artificial vegetation in desert regions. However, the periods of droughts after each rainfall event form a series of stochastic shocks and result in depletion of the seed bank. Understanding seed bank persistence under stochastic shocks is a crucial prerequisite to evaluation of the sustainability of artificial vegetation under ongoing drought, but few studies have addressed the response of soil seed bank persistence to frequent stochastic shocks in deserts. In this paper, a stochastic shock model describing the input–output process of the seed bank is proposed based on modern risk theory to explore seed bank persistence of artificial vegetation in desert regions. The existence condition of the seed bank and the lower bound of the probability of seed bank persistence under stochastic shock after a long time were theoretically determined. As a verification of the theoretical model, seed bank persistence of artificial sand-binding vegetation on the southeastern margin of the Tengger Desert in China was investigated based on field observation and experimental and empirical data. The numerical results reveal that the seed bank of artificial sand-binding vegetation will always exist under stochastic shocks with a probability greater than 95%, suggesting that in the long-term succession process the current artificial sand-binding vegetation will likely maintain the development trend toward the adjacent natural vegetation.