Abstract: The traditional neural network has low accuracy in load forecasting with strong time dependence. This paper provided a load prediction method based on the convolutional neural network (CNN) and the long short-term memory network (LSTM) to improve the accuracy of short-term power load.Moreover, it collected 5-dimensional load characteristic data and extracted spatial coupling interaction features of data by using CNN convolution layer and pooling layers feature extraction units.In addition, it inputted the reconstructed data into the LSTM network to mine the load timing characteristics and used dropout technology to increase the model generalization ability. Besides, it used an adaptive moment estimation (Adam) optimizer to train the model. It entered the test data into the trained neural network model to predict the electric load in the next 1 h and 12 h. The experimental results show that the proposed model is better than the improved neural network forecasting models,such as improved BP neural network and LSTM, from the convergence speed and forecasting accuracy perspective. The prediction accuracy of 1 h load forecasting is 98.66%, and the 12 h load forecasting accuracy is 96.81%, which improves the accuracy of short-term power load forecasting.

    Key words: long short-term memory network; short-term load forecasting; Dropout technology; convolutional neural network; adaptive moment estimation