Short-Term Load Forecasting Technology with Distributed Energy Timing Uncertainty

With the rapid growth of distributed photovoltaic scale, the impact of random fluctuation characteristics of its output on urban load is also increasing. Facing the load forecasting scenario of large-scale distributed PV access, this paper proposes a short-term load forecasting method considering the influence of distributed PV. The side load of the grid under photovoltaic access is the difference between the actual electricity load and photovoltaic output. Therefore, before constructing input data, this paper first adopts big data mining technology to analyze the characteristics of photovoltaic output and user side load and the correlation between them and their respective influencing factors. The influential factors with high correlation are selected as the input feature set of the load forecasting model through feature construction. Then, the LSTM neural network prediction model with self-attention mechanism is constructed to deeply explore the load sequence characteristics. Grey Wolf algorithm was used to optimize the prediction model, and the model with the best prediction effect was determined. The numerical results show that the proposed method can effectively improve the prediction accuracy of net load with distributed PV.