Short-Term Load Forecasting for Microgrid Considering Weather Characteristics and Source-Load Correlation

Within microgrids, the utilization of distributed energy resources primarily caters to internal loads. Given the restricted spatial footprint of microgrids and the coexistence of sources and loads within identical regional weather contexts, it becomes imperative to forecast the power dynamics of both distributed energy resources and loads concurrently, under the consistent ambient settings. To address this, the present study investigates the influence of weather characteristics on source and load power and extracts pertinent weather feature data for short-term load forecasting within microgrids. Initially, a Multivariate Maximal Information Coefficient (Mv-MIC) is introduced to discern weather features exhibiting high correlation with source and load power sequences. Subsequently, a similar day dataset is established through grey relational analysis. Furthermore, factor analysis is deployed to supplant highly correlated feature sequences with common factors. These common factors, along with photovoltaic power sequences, undergo adaptive noise complete ensemble empirical mode decomposition (CEEMDAN). Finally, a composite prediction model integrating BPNN, LSTM, and Bidirectional Long Short-Term Memory Network (BiLSTM) architectures is constructed. Simulation validation employing DTU 7K 47-node data indicates enhanced prediction accuracy compared to scenarios exclusively considering the impact of weather features on distributed energy resources within microgrids, with RMSE and MAE metrics improved by 4.04% and 12.37% respectively, for the former, and 2.26% and 0.65% respectively, for the latter.