Research On Short-Term Load Forecasting Based On Deep Learning

With the continuous development of China’s power grid,the change of power load is becoming more and more complicated,and the research of power load forecasting has become an important content of power grid management.Short-term load forecasting refers to forecasting the load in the next 1-7 days,which is an important basis for the dispatching center to formulate the power generation plan and an important part of the energy management system,and has a very important impact on the operation,control and planning of the power system.Improving the accuracy of short-term load forecasting of power system can not only enhance the safety of power system operation,but also improve the economy of power system operation.In this paper,aiming at improving the accuracy and speed of short-term load forecasting,based on the domestic and international research theories and results,a CEEMD-LSTM-MLR short-term load forecasting model based on complementary set empirical mode decomposition(CEEMD),long-term and short-term memory(LSTM)neural networks and multiple linear regression(MLR)methods is proposed,and the reliability and accuracy of the proposed model are verified by the analysis of the calculation cases.The main work is as follows:(1)The original load data has certain volatility and randomness,and direct forecasting will lead to long forecasting time and low forecasting accuracy.In this paper,the advantages and disadvantages of traditional empirical mode decomposition algorithm,ensemble empirical mode decomposition algorithm and CEEMD algorithm are compared and analyzed,and the CEEMD algorithm with the smallest reconstruction error and the fastest decomposition speed is selected to decompose the power load sequence into high and low frequency components,which effectively reduces the fluctuation of the load sequence.(2)The complex high-frequency components,meteorological data and date type data are predicted by the LSTM neural network with powerful long-time series information mining ability,and the super parameters of the LSTM neural network are optimized by Bayesian algorithm,which significantly improves the prediction accuracy;Using MLR method to predict the low-frequency components with strong periodicity can not only ensure the prediction accuracy,but also avoid the complicated training process and effectively shorten the training time.(3)Through an actual example,the forecasting effect of the combined model with different decomposition algorithms and different forecasting algorithms is compared and analyzed;The improvement of prediction results by Bayesian parameter adjustment is verified.The results show that the forecasting model proposed in this paper has high forecasting accuracy and good applicability,and has certain application value in the field of power system load forecasting.