Load Forecasting Method Considering Distributed Photovoltaics and Multi-Dimensional Temporal Characteristics

With the rapid popularization of distributed photovoltaic (PV) systems, traditional load forecasting is facing more and more challenges due to various problems caused by distributed PV grid connection. Distributed PV significantly affects the net load distribution and weakens the effectiveness of traditional forecasting methods. Therefore, improving the accuracy of load forecasting under high PV penetration has become a key issue for the safe and efficient operation of modern power systems. To solve this problem, a load forecasting method based on Particle Swarm Optimization and Bi-directional Long Short-Term Memory (PSO-BiLSTM) network with distributed PV identification and multi-dimensional temporal characteristics is proposed. Firstly, the multi-dimensional temporal characteristics are constructed by combining the lagged characteristics, cyclical characteristics and calendar characteristics, which enhances the ability of the model to capture the time correlation between load data. At the same time, using the net load characteristics under different weather conditions, the distributed PV identification characteristic is constructed to effectively identify whether users install distributed PV. Then, PSO algorithm is used to automatically optimize the key hyperparameters of BiLSTM, so as to improve the robustness of the model. Finally, the effectiveness of the method is verified by the actual load data. The experimental results show that the PSO-BiLSTM model is superior to other models in many evaluation indexes. Especially in the case of distributed PV grid connection, the prediction accuracy of the model has been significantly improved, and the time relationship and load changes caused by distributed PV have been effectively captured. These results confirm the effectiveness and practicability of the proposed method in power system load forecasting with high distributed PV penetration.