Short-term Household Electricity Demand Forecasting Based On Deep Learning

Short-term household electricity demand forecasting is an important and challenging subject in smart grid planning,sustainable energy utilization and power market pricing system design.Due to the irregularity of human behaviours and the univariate data recorded by smart meters,traditional load forecasting methods,such as the grey model,regression analysis,support vector regression,etc is difficult to accurately predict the power load of individual households.Long short-term memory neural network(LSTM)of the deep learning theory is an effective method to deal with the problem of time series modeling,especially for household electric power demand data.As a result,the following work is done on short-term household power demand forecast based on long short-term memory neural network:(1)In this dissertation,the theory of recurrent neural network is studied.According to the characteristics of household electricity demand data,a single variable short-term home electricity demand forecasting model is proposed.The model is constructed by using the long short-term memory neural network,which overcomes the long-term dependence problem of the traditional recurrent neural network,and realizes the prediction of 1 hour household electricity consumption.(2)In order to solve the problem of single dimension of household electric power data,a hybrid deep neural network(CLSTM)model is proposed.The convolutional neural network is introduced to extract the feature of electric power sequence data,and the feature dimension is extended to forecast the electric power demand of home in 5 minutes.The validity of the model is proved by comparing it with ARIMA,SVR,LSTM and other methods.(3)In order to introduce more response time to the bidding of electricity market,a K-step continuous power demand forecasting method is proposed based on the CLSTM model proposed earlier.The multi-step prediction of power demand is completed to better analyze the change of household electricity demand over a period of time.(4)The design and implementation of a household load forecasting software.The software consists of four modules: data management,prediction model,graphical interface and auxiliary function.The software can forecast the actual household electricity demand,and can display the forecast results graphically.In addition,it provides user guide,technical support and other auxiliary functions to make the interaction more user-friendly.