Research On Resident Electricity Consumption Characteristics Based On Machine Learning

In recent years,the proportion of residential electricity demand to the total demand of social electricity is gradually rising.Compared with industrial electricity demand,residential electricity consumption demand is more flexible and easier to be regulated by policies.Therefore,reducing the unreasonable resident's electricity consumption will help to reduce the whole society's electricity consumption,which is conducive to the healthy and sustainable development of the economy.From the perspective of demand side,studying the pattern of household electricity consumption and electricity consumption behaviour is helpful to make effective and reasonable intervention strategies.Nowadays,big data and artificial intelligence play an important role in energy consumption and energy forecasting fields.Therefore,this dissertation takes advantage of machine learning technology to study residential electricity consumption characteristics.In this dissertation,residential electricity consumption characteristics and residential electricity consumption probability density prediction are studied respectively,and a case study of electricity consumption in Nanjing,Jiangsu Province,2014 is given.Firstly,residential electricity consumption patterns at festivals are explored by k-means clustering algorithm.The results show that: there are three typical electricity consumption patterns during the Spring Festival and two typical electricity consumption patterns during the Labour Day and the National Day,and they are consistent with people's behaviour pattern during the period.Secondly,we study on the seasonal electricity consumption patterns,the result indicates that: electricity consumption in winter and summer is higher than its in spring and autumn;the volatility of electricity consumption profiles is higher in winter and summer;electricity consumption in winter and summer is vulnerable to extreme high temperature and extreme low temperature;high temperature will lead to a significant increase in electricity consumption in summer,and the peak of electricity consumption always lags behind high temperature;As the number of high temperature increases,the response time of residential electricity consumption to high temperature becomes shorter.Finally,this dissertation studies residential electricity consumption probability density prediction by combining deep forward network with kernel density estimation.We construct influencing features as many as possible by means of feature engineering,such as temperature related features,air quality related features,weather related features,date related features,etc.On this basis,residential electricity consumption point prediction and probability density prediction is constructed.The results show that: the prediction accuracy of deep neural network is higher compared with random forest and gradient boosting machine.Probability density prediction can give accurate prediction interval,thus providing more effective information for decision makers.According to the above analysis results,this dissertation puts forward relevant policy suggestions to encourage residential electricity consumption behaviour.