Research On Short-Term Load Forecasting Method Based On Deep Learning

Smart grid is a new generation of future-oriented,sustainable and environmentally friendly power grid.Its foundation is a high-speed bidirectional communication network that integrates advanced sensing measurement technology,decision control technology and other related aspects of intelligence and other professional technologies to achieve safe,stable and efficient operation of the power grid.The smart grid coordinates the needs and capabilities of all generators,grid operators,end users and power market stakeholders to operate all parts of the system as efficiently as possible,minimizing costs and environmental impacts.Some of the benefits of the smart grid include: higher power transmission efficiency,faster power recovery after power outages,better integration of user self-generation systems,improved security and more choices and control for users.Load forecasting is an important link in the development of smart grid.Therefore,this thesis conducts in-depth research on load forecasting.This thesis first analyzes the load forecasting.Through the analysis,we can understand that there are many types of load forecasting.This thesis chooses short-term power load forecasting,and then understands that the characteristics of load forecasting are timeliness and uncertainty.Finally,it analyzes the factors that affect the load forecasting.Through the analysis,it can be known that there are many factors that affect the prediction.This thesis chooses the temperature and rainfall to predict the load.Then,the thesis introduces the principles and characteristics of Convolutional Neural Networks(CNN)and Long Short-Term Memory Networks(LSTM).Through the combination of convolutional neural network and long short-term memory neural network,the power load can be predicted.However,considering the possible problems of over-fitting or gradient disappearance in the combination of convolutional neural network and long short-term memory neural network,this thesis chooses to add attention mechanism to the combination model.The attention mechanism can better complete the extraction of key features of the data.The extraction of data features is completed by convolutional neural network and long-term and short-term memory neural network,and then the focus extraction is completed by the attention mechanism.Finally,short-term power load forecasting can be completed better.Finally,use CNN-LSTM to perform short-term power load forecasting first,and adjust the parameters of the combined model through experiments,,such as Epoch,activation function,etc.After the parameters are adjusted,the attention mechanism is added for comparative experiments.LSTM,CNN-LSTM and CNN-LSTM-ATT to perform load forecasting on the same set of data respectively.Through the comparison of the results of three groups of experiments,it can be concluded that adding attention mechanism can improve the accuracy of short-term power load forecasting and complete the prediction work better.