Load Forecasting based on CNN-LSTM for 96 Time Points

Load forecasting is very important for power planning and dispatching. Especially with the large-scale new energy connected to the large power grid, the uncertainty of the impact on the large power grid is further increased. How to accurately grasp the load changes in a short time and make scheduling strategies according to load changes has become an important problem at present. In this paper, a method based on CNN-LSTM is proposed to predict the load at 96 time points. Considering 11 load influencing factors, the optimal network structure is obtained by constantly adjusting the CNN-LSTM structure. Experiments show that when the depth of LSTM and CNN are both 1, the corresponding MAPE value is the smallest (0.13). When the depth of CNN is changed, it is always at the time of 300 iterations, and the corresponding Loss is 0.0005. When the CNN depth is consistent and the number of iterations reaches 300, the Loss is the best when the LSTM depth is 3, 4, 5 and 6 (0.0005).