Abstract: In view of the low accuracy and hysteresis of the traditional short-term power load forecasting model, this paper proposes a hybrid neural network model based on CNN (convolutional neural network), LSTM (long short-time memory network) and attention mechanism. The convolutional layer is used to extract the influence features of multidimensional power data, filter the non-important factors, complete the mapping transformation of relevant features of power data. Then the cycle of the long short-time memory network layer can selectively forget and remember the temporal data. Finally, the attention mechanism is used to add the weight of important features, and the results will be exported by Adam optimization. This method relies on the GPU’s big and powerful computing to solve the real-time problem of prediction, improving its accuracy of prediction by means of multi-fusion neural networks. The proposed model is proved to be true and reliable in the light of an example, and the quality of prediction is significantly better than other traditional models.

Key words: short-term power load forecasting, convolutional neural network, long short-time memory neural network, attention mechanism