Short-term Power Load Interval Prediction Based on CEEMDAN-SE-LSTM and KDE

Short-term power load possesses strong randomness and volatility, and the accuracy of its prediction is of importance to raise both power supply reliability and power system operation economy. In allusion to the disadvantage that traditional deterministic prediction cannot reflect future load fluctuation, based on the idea of "point forecast plus interval estimation" a short-term load interval prediction method was proposed. Firstly, based on complete ensemble empirical mode decomposition with adaptive noise method (abbr. CEEMDAN), the load series was decomposed into multi modal components, and according to the calculation results of sample entropy (abbr. SE) of different sequences the sequences were reconstructed to reduce the computation. Secondly, on this basis, a long and short-term memory (abbr. LSTM) neural network prediction model was constructed respectively for each component to obtain the predicted value of the future load point. Finally, based on above-mentioned work, the kernel density estimation (abbr. KDE) method was utilized to estimate the distribution of the prediction error, further, combining with point prediction results the interval prediction of future short-term load was implemented. Comparing the proposed model with other models, the results show that using the proposed model a lower point prediction error can be implemented, meanwhile, the proposed model exerts a better combination property in the interval prediction.