Research On Power Load Forecasting Problem Based On Deep Learning Neural Network

In the 21st century,with the continuous development of the global economy and modern industrialization,electric energy has become the most important form of energy utilization in the world because of its great advantages of easy transmission and conversion and wide application.Today's social life has long been inseparable from the intimacy.This paper mainly studies the forecasting of power load,which is an important reference for power management in power grid.High-precision power load forecasting has many advantages in ensuring efficient use of electric energy,reducing operating costs,accurately reflecting the laws of electricity consumption,optimizing the market environment for electricity use,ensuring power stability,reducing power outage losses,and so on.Power system load forecasting plays an important role in China's social production and economic development.The issue of higher quality and higher precision power load forecasting is becoming more and more urgent and important.Firstly,this paper builds a power load model based on ELMAN neural network on MATLAB platform.ELMAN neural network is a widely used feedback neural network model.The model uses the delay operator to achieve local self-loop,which enables the internal feedback network to have dynamic information processing capability,which can better adapt to the historical data of the previous step.The experiment uses the sample data of the global electric load forecasting competition.The experiment is divided into two parts:(1)in the span of several consecutive years,the power data of key months are trained and learned,and the accuracy of prediction can reach a very high level(the error rate is within 4%);(2)in the whole period of a year,the accuracy of prediction has declined,but the error can still be within 7%.Experimental results show that ELMAN model is more accurate in short-term power load forecasting,and it is more suitable to deal with power load forecasting problems with obvious timing characteristics.Secondly,in the framework of Tensorflow,this paper constructs a load forecasting model based on LSTM(Long-Short Term Memory).Firstly,the data sets of time,power load and air temperature are selected from the global power load forecasting competition data,and then divided into training sets and test sets after preprocessing and shaping.Finally,three groups of experiments are designed according to the different size of data sets and the different number of hidden layers to import LSTM model for training and forecasting.The experimental results show that the prediction error rate of the model is 2%.Finally,by comparing the above two models with the accuracy of power load forecasting as the standard,it can be found that LSTM model has obvious advantages in training and forecasting accuracy,which is worth further study in the future.