Research On Method Of Short-term Load Forecasting Based On Improved Convolutional Neural Network

In the development process of smart grid,rapid and accurate load forecasting is the main basis for grid adjustment and generation.The emergence of artificial intelligence and machine learning has transformed load forecasting from human guessing into a more rapid and accurate intelligent algorithm.Nowadays,with the increasing scale of power consumption and the increasing complexity of power system,higher requirements are put forward for the accuracy of load forecasting.In this context,short-term load forecasting is selected as the research content of this paper,various aspects of short-term load forecasting are analyzed,and a new forecasting model is proposed.The main work of this paper is as follows:Firstly,various methods of short-term load forecasting are discussed and their advantages and disadvantages are studied in this article.Then,by exploring the characteristics of power system load and its influencing factors,the actual load data are pre-processed,and some conclusions that need to be used later are obtained.Based on the analysis of intelligent algorithm for short-term load forecasting,the principle of neural network is explored.Neural network has good non-linear approximation ability.However,when the data scale increases,the parameters of neural network increase sharply,resulting in slow forecasting speed and easy to fall into local minimum.In order to improve the accuracy of short-term load forecasting,a load forecasting model based on convolution neural network is proposed in this paper.The local connection and weight sharing mechanism of convolution neural network can reduce the number of parameters,reduce the data dimension,and improve the forecasting accuracy.In this paper,experiments are carried out with actual load data.The results show that compared with BP neural network,convolution neural network can achieve better results.Good prediction results,but there is still room for improvement.Then,aiming at the application of convolutional neural network.An improved method is proposed to optimize the initial parameters of convolution neural network by using improved cuckoo search.The initial parameters of the network are determined by using the global optimization ability of ICS.The prediction model of ICS-CNN is constructed.The short-term load forecasting is carried out by using the optimized convolution neural network.The BP-NN model and the unoptimized CNN model are compared and simulated.Result The MRE of ICS-CNN model is 1.37%,which is 1.86% lower than that of BP neural network(3.23%)and meets the requirement of short-term load forecasting accuracy.Compared with the traditional neural network,it can predict short-term load changes more quickly and accurately,with stable output,high prediction accuracy and good robustness.