Objective To obtain more satisfactory fitting effect and forecasting accuracy of load prediction for photovoltaic storage direct-supply distribution networks (PSDDNs), this paper proposes a load prediction method based on snow ablation optimization (SAO) and long short-term memory network (LSTM).

Method The 3σ statistical method was adopted in the paper to normalize the load data of the PSDDNs. The key structural parameters of the three layers of the LSTM, i.e., the output gate, the forgetting gate, and the input gate were optimized by establishing a typical LSTM network and utilized the SAO algorithm to achieve regular load information extraction and load prediction. In addition, the performance evaluation metrics for load prediction including root-mean-square error and squared error were utilized to quantitatively evaluate the fitting effect and forecasting accuracy of load prediction.

Result Lower prediction errors are achieved by employing the load prediction method proposed in this paper, which featured more refined prediction results, faster convergence and better adaptation compared to the load prediction based on Particle Swarm Optimization (PSO)-LSTM.

Conclusion The proposed method can help to effectively extract the timing information of load sequences and eliminate the influence of abnormal load data, providing a reference for the energy management and optimal operation of the PSDDNs.