Closed-Loop Probabilistic Forecasting Method of Short-Term Spatial Load Considering Corrupted Data and Field Verification

Le résumé fourni par la source

The rapid development of flexible load technologies brings new challenges to short-term spatial load forecasting (SSLF) in distribution networks (DNs). Conventionally, as an open-loop method, SSLF is unable to dynamically improve the forecast results. However, the nonstationary and volatility features of loads reduce the accuracy of the SSLF, the errors caused by load data loss and distortion further reduce the robustness of the forecasting. In this paper, we propose a closed-loop probabilistic forecasting method for the SSLF by integrating the modified Spatio-Temporal Graph Convolutional network (ST-GCN) and the statistical load baseline profile (SLBP). The modified ST-GCN consists of GCN-Gate Recurrent Unit (GCN-GRU) and GCN in parallel to simultaneously learn the spatio-temporal correlation of loads. To capture the uncertainty of the SSLF, the concrete dropout enabled Bayesian neural networks are applied. The SLBP, which is calculated by statistical method, is used to extract the periodicity of the load and mitigate the impact caused by corrupted data. The closed-loop forecasting architecture is devised to integrate the modified ST-GCN and SLBP by the dynamic clustering technique, to in turn improve the accuracy and the robustness of the forecasting results. Numerical tests are conducted using the real load data of DNs in China. Test results confirmed the superiority of the proposed method.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.