Net Load Forecasting Method for Distribution Substation Based on an Improved MHGSA Mechanism

With the promotion and application of distributed photovoltaic power generation in the distribution substation, the net load of the distribution substation is affected by both the total electricity consumption load of users and the photovoltaic output. Considering various influencing factors such as photovoltaic power generation, meteorology, and non-working day user electricity consumption behavior, as well as the temporal characteristics of net load in distribution substation, this paper proposes a net load forecasting model for distribution substation based on an improved Multi-Head Gumbel-softmax Sparse self-Attention (MHGSA) mechanism. The model first decomposes the net load time series data through variational mode decomposition. Then the training dataset is formed by overlaying factors such as photovoltaic power generation and meteorology, and the MHGSA mechanism is introduced into a dual-layer LSTM model to establish a net load forecasting model. Experimental results have shown that the method proposed in this paper has high prediction accuracy and can achieve accurate prediction of net load in distribution substation.