Abstract:In order to improve the accuracy of distribution substation heavy overload warning, this paper proposes a short-term load prediction method for distribution substation based on VMD-CNN-BiLSTM-CBAM model. The method firstly extracts similar days using K-mean clustering, adopts variational modal decomposition (VMD) to decompose the load on similar days to obtain IMF components, and adopts the model combining convolutional neural network (CNN)-bi-directional long and short-term memory network (BiLSTM)-double self-attention mechanism (CBAM) for each IMF component to make a prediction, and finally, utilizes the sample entropy to reconstruct the predicted daily load curves of distribution substation to obtain the predicted daily load curves of distribution substation. Finally, the sample entropy is used to reconstruct each prediction component to obtain the daily load curve of distribution variable prediction. Examples show that the method proposed in this paper has high prediction accuracy compared with the methods of convolutional neural network (CNN) and adaptive fuzzy neural network model (ANFIS), which can provide technical support for the subsequent heavy overload warning of distribution substation.