Search

Introduction: The characteristics of intermittency and volatility brought by a high proportion of renewable energy impose higher requirements on load forecasting in modern power system. Currently, load forecasting methods mainly include statistical models and machine learning methods, but they exhibit relative rigidity in handling the uncertainty, volatility, and nonlinear relationships of new energy, making it difficult to adapt to instantaneous load changes and the complex impact of meteorological factors. The Transformer model, as an algorithm used in natural language processing, with its self-attention mechanism and powerful nonlinear modeling capability, can help address the aforementioned issues. Methods: However, its current performance in time series processing is suboptimal. Therefore, this paper improves the Transformer model through two steps, namely, Data-Slicing and Channel-independence, enhancing its adaptability in load forecasting. Results: By using load data from Northern Ireland as an example, we compared GRU, CNN, and traditional Transformer models. We validated the effectiveness of this algorithm in short-term load forecasting using MAPE and MSE as indicators. Discussion: The results indicate that, in short-term load forecasting, the MDS method, compared to GRU, CNN, and traditional Transformer methods, has generally reduced the MSE by over 48%, and achieved a reduction of over 47.6% in MAPE, demonstrating excellent performance. [ABSTRACT FROM AUTHOR]

With the popularity of electric vehicles (EVs), vehicle-to-grid (V2G) technology is attracting increasing attention due to its crucial merit of enabling bidirectional power flows between EVs and grid, so as to enhance the grid security and stability by regulated dispatching. However, the existing V2G approaches are confronted with several unrealizable challenges because of high computational complexity for large-scale EVs and impracticality for future power data acquisition. In this article, an edge computing framework is proposed in a distributed manner to ensure the dispatching efficiently and provide the raw dataset flexibly. Meanwhile, the long short-term memory network is applied to prediction merely by the past and present power data. Moreover, attention mechanism and data clustering are utilized to improve the prediction accuracy and operation robustness. Experiments involving real dataset demonstrated that the proposed V2G scheme is able to achieve very satisfactory dispatching performance with the prediction accuracy up to 98.89%. [ABSTRACT FROM AUTHOR]

The rising popularity of electric vehicles (EVs) underscores the potential of vehicle-to-grid (V2G) technology to contribute to load peak-shaving, valley-filling, and photovoltaic (PV) self-consumption. Effective V2G control strategies can be obtained by data-driven techniques, which is able to leverage historical and current data to inform future decision-making amidst uncertainties. However, the centralized collection and sharing of data among charging stations face challenges due to data asset concerns. Furthermore, even if data sharing hurdles are overcome, the non-independent and non-identically distributed (Non-IID) nature of data across charging stations can still negatively impact performance. In this study, we introduce FedPT-V2G, a security-enhanced federated transformer learning approach for real-time V2G dispatch that addresses Non-IID data. We employ deep learning models trained on historical and current data to enable real-time decision-making, facilitating both load shifting and PV self-consumption. Additionally, we utilize federated learning to jointly train a global model across all charging stations without collecting or sharing any local private data. We pioneer the application of the Proximal algorithm and Transformer model to tackle data distribution discrepancies within the V2G scheduling prediction task. The Proximal algorithm employs regularization techniques to align local models at each charging station more closely with the global model during updates. Concurrently, the multi-head attention mechanism within the Transformer model allows learned feature vectors to diverge, enabling better exploitation of variations across the entire feature space. Finally, we validate the proposed FedPT-V2G approach through extensive numerical simulations, demonstrating comparable performance to centralized learning on both balanced (98.93% vs 98.65%) and imbalanced (92.15% vs 92.20% in label skew) datasets. • End-to-end learning based on historical data to make V2G decisions in real-time. • Enable load shifting and PV self-consumption under diverse uncertainties. • Apply Proximal algorithm and Transformer model to handle Non-IID data in V2G task. • Verify the proposed FedPT-V2G approach through extensive numerical simulations. • Achieve similar performance to centralized learning on both IID and Non-IID data. [ABSTRACT FROM AUTHOR]