Research On Short-term Load Forecasting Based On Deep Learning

Short-term load forecasting is an important link in power network production planning and actual dispatching operation,and it is also a hot topic in the academic circle.Power system is a random and fluctuating system,and the historical load data generated by it is also random and fluctuating,which brings great difficulties to the load prediction of power grid.Through the study of deep learning,this thesis explores the advantages of deep learning and the problem of short-term load prediction of power system.The main work of this thesis is as follows:(1)A variety of deep learning and data preprocessing methods are deeply discussed,the current research status of load forecasting at home and abroad is discussed,and some current load forecasting methods at home and abroad are summarized.According to the influence of load characteristics,type,temperature,climate,date type and other influencing factors on the power load,the power load data collected by the intelligent instrument is preprocessed,the lost data is filled,the abnormal data is identified and corrected,and the corrected load data is normalized.(2)Construct short-term load forecasting based on Differential Evolution-Convolutional Neural Networks-Long and Short Term Memory Network(DE-CNN-LSTM).By learning and understanding the characteristics of convolutional neural network and long and short term memory network,the CNN-LSTM hybrid neural network is firstly constructed.The convolutional neural network is used to convolve the power load sequence data,and the long input sequence data is processed into the characteristic short sequence data.The new characteristic short sequence after the convolution operation is then input into the LSTM network.The superparameters composed of the number of neurons and the number of iterations in the hidden layer will have a certain impact on the prediction accuracy,and the adjustment of the superparameters is a key link of the algorithm.The differential evolution algorithm is used to adjust the superparameters of the CNN-LSTM network,so as to optimize its network structure.The optimized CNN-LSTM neural network is compared with conventional CNN-LSTM and traditional CNN model.(3)Based on the above content,this thesis combines convolutional neural network,long and short term memory network and differential evolution algorithm to solve the problem of short-term load prediction.By comparing with the traditional CNN-LSTM and CNN network model and using differential evolution method to optimize,it is found that DE-CNN-LSTM can not only effectively improve the prediction accuracy,but also accelerate the convergence rate of the network model in the training process.This study provides a new idea for short-term load forecasting of power system and lays a solid foundation for the connection between load forecasting and power market forecasting.