Research On Short-term Power Load Forecasting Based On Deep And Broad Learning

Short term load forecasting is an important work of power dispatching department,and it is the main basis to arrange the start-up and stop of generator units.Accurate short-term load forecasting can reduce power loss and improve power supply quality,so improving the accuracy of forecasting is the key work of researchers.Artificial neural network has strong nonlinear approximation and self-learning ability,which is one of the hot directions of short-term power load forecasting.Taking the power load data of a region from January to June as the simulation data set,the data set is preprocessed,including missing value supplement,outlier correction and normalization.Then the periodic characteristics of load change are analyzed,and the input characteristics of load forecasting are extracted by using autocorrelation function,which lays the foundation for the construction of forecasting model.Broad learning system(BLS)is a new learning paradigm.Its flat structure design makes it have fast learning speed and good generalization ability.In view of the shortcomings of deep neural network,such as large amount of calculation and easy over fitting,the broad learning system is used to build the prediction model,and the parameters of the broad learning system are optimized by using the improved particle swarm optimization algorithm.Through simulation and comparison with some other popular prediction models,it is verified that the BLS model after parameter optimization has the highest prediction accuracy and takes the shortest time compared with other models.In order to achieve higher prediction accuracy,long short term memory network(LSTM)is used to build the prediction model.LSTM can achieve long-term memory and get higher prediction accuracy,but it is greatly affected by the stationarity of the data itself.Due to the strong randomness and volatility of power load,empirical mode decomposition(EMD)technology is proposed to stabilize the original load data set.Due to the characteristics of weekly and daily periodicity of load variation,the single input LSTM model has limitations.A multi-dimensional input LSTM model is constructed.Combined with empirical mode decomposition,the simulation results show that the proposed model has high prediction accuracy.