Search

When considering agricultural commodity transaction data, long sampling intervals or data sparsity may lead to small samples. Furthermore, training on small samples can lead to overfitting and makes it hard to capture the fine-grained fluctuations in the data. In this study, a multi-scale forecasting approach combined with a Generative Adversarial Network (GAN) and Temporal Convolutional Network (TCN) is proposed to address the problems related to small sample prediction. First, a Time-series Generative Adversarial Network (TimeGAN) is used to expand the multi-dimensional data and t-SNE is utilized to evaluate the similarity between the original and synthetic data. Second, a greedy algorithm is exploited to calculate the information gain, in order to obtain important features, based on XGBoost. Meanwhile, TCN residual blocks and dilated convolutions are used to tackle the issue of gradient disappearance. Finally, an attention mechanism is added to the TCN, which is beneficial in terms of improving the forecasting accuracy. Experiments are conducted on three products, garlic, ginger and chili. Taking garlic as an example, the RMSE of the proposed method was reduced by 1.7% and 1% when compared to the SVR and RF models, respectively. Its R 2 accuracy was also improved (by 4.3% and 3.4%, respectively). Furthermore, TCN-attention and TCN were found to require less time compared to GRU and LSTM. The accuracy of the proposed method increased by about 5% when compared to that without TimeGAN in the ablation study. Moreover, compared with TCN, the Gated Recurrent Unit (GRU), and the Long Short-term Memory (LSTM) model in the multi-scale price forecasting task, the proposed method can better utilize small samples and high-dimensional data, leading to improved performance. Additionally, the proposed model is compared to the Transformer and TimesNet models in terms of its accuracy, deployment cost, and other metrics. [ABSTRACT FROM AUTHOR]

When considering agricultural commodity transaction data, long sampling intervals or data sparsity may lead to small samples. Furthermore, training on small samples can lead to overfitting and makes it hard to capture the fine-grained fluctuations in the data. In this study, a multi-scale forecasting approach combined with a Generative Adversarial Network (GAN) and Temporal Convolutional Network (TCN) is proposed to address the problems related to small sample prediction. First, a Time-series Generative Adversarial Network (TimeGAN) is used to expand the multi-dimensional data and t-SNE is utilized to evaluate the similarity between the original and synthetic data. Second, a greedy algorithm is exploited to calculate the information gain, in order to obtain important features, based on XGBoost. Meanwhile, TCN residual blocks and dilated convolutions are used to tackle the issue of gradient disappearance. Finally, an attention mechanism is added to the TCN, which is beneficial in terms of improving the forecasting accuracy. Experiments are conducted on three products, garlic, ginger and chili. Taking garlic as an example, the RMSE of the proposed method was reduced by 1.7% and 1% when compared to the SVR and RF models, respectively. Its R2 accuracy was also improved (by 4.3% and 3.4%, respectively). Furthermore, TCN-attention and TCN were found to require less time compared to GRU and LSTM. The accuracy of the proposed method increased by about 5% when compared to that without TimeGAN in the ablation study. Moreover, compared with TCN, the Gated Recurrent Unit (GRU), and the Long Short-term Memory (LSTM) model in the multi-scale price forecasting task, the proposed method can better utilize small samples and high-dimensional data, leading to improved performance. Additionally, the proposed model is compared to the Transformer and TimesNet models in terms of its accuracy, deployment cost, and other metrics.

Considerable improvement has taken place in forecasting tropical cyclones at 24–48 hr leads; however, improving the accuracy of tropical cyclone forecasts at longer leads is still a major scientific challenge. The major bottleneck in accurate tropical cyclone forecasts using limited area models (LAMs) comes from the use of artificial lateral boundary conditions, especially at longer leads. Although global circulation models (GCMs) still cannot match the horizontal resolution that can be implemented in a LAM over a smaller domain, it is possible that better representation of scales and thus scale interactions in a global domain can lead to better simulation of tropical cyclones with a GCM even with relatively coarser resolution. This hypothesis is tested in the present work with a GCM and a LAM configuration. Thirty cyclones over the north Indian Ocean that represent different seasons and intensities during 1999–2012 are considered. Analysis of forecast skills at three leads (24 hr, 48 hr and 96 hr) show that while the LAM has better skill compared to the GCM at shorter leads (<48 hr), the GCM has significantly higher skill at longer leads (96 hr). The two configurations are found to exhibit somewhat complementary skills in terms of forecast lead and the severity of the cyclones. Therefore, it is suggested that a methodology combining both LAMs and GCMs can provide more reliable forecasts. For a large sample of cyclones of various intensities formed over the north Indian Ocean, the forecast skills of a global and a regional model with different forecast leads are compared. Although a global circulation model (GCM) cannot match the resolution of a limited area model (LAM) over a smaller domain, it is possible that better representation of scales and thus scale interactions can lead to better simulation of tropical cyclones with a GCM. This hypothesis is tested in the present work with a GCM and a LAM configuration.Histogram of errors in landfall for different leads (24 hr, 48 hr and 96 hr) in a LAM and a GCM. The number in parentheses represents the number of cases with <200 km for the respective lead hours. [ABSTRACT FROM AUTHOR]

Highly wind penetrated future power system will couple to the variabilities and nonlinear correlations of wind. Reliable wind power forecasting (WPF) for a region is critical to the security and economics of the power system operation. Therefore, this paper proposes a multiscale WPF method by establishing a multi-to-multi (m2m) mapping network and the use of stacked denoising autoencoder (SDAE). The concerned forecast time horizon is 24–72 hours. First, multi-NWPs in a region are corrected based on SDAE to generate better inputs for the following regional WPF. Second, a number of SDAEs with diverse model parameters and input features are integrated into ensemble SDAE for predicting the wind power generated from various wind farms in a region. Two sets of data are utilized in this case study to validate the proposed method. The results show that the proposed m2m mapping and SDAE are able to capture the real correlations of wind at multiple sites, and outperform the other counterparts in terms of multi-NWPs correction as well as the WPF for both the region and individual concerned wind farm. Moreover, the ensemble SDAE performs better than any other individual regional WPF model. [ABSTRACT FROM PUBLISHER]