Household Power Consumption Mode Analysis And Load Forecasting Based On Power Big Data

With the development of smart grid technology,digital smart meters have been used widely.The power grid data acquisition system has collected a large number of household power consumption data.How to mine the value of power consumption data is one of the main problems in power grid companies at this stage.As an important means of intelligent power management and power marketing,home power consumption mode analysis and home load forecasting based on artificial intelligence play a key role.The main problems in the current research are:(1)the research on household power consumption mode lacks in-depth analysis on the clustering results,and the data sets used in experiments are too small to be representative enough.(2)There are fewer studies on load forecasting on household granularity,and the accuracy of forecasting needs to be further improved.This paper studies the above problems.The research is divided into the following three major parts:(1)The k-means algorithm is used to mine and analyze the household power consumption patterns of the monthly power consumption data of nearly 20000 households in a city.These households are classified into appropriate clusters using kmeans according to the power consumption characteristics of each household,followed by a clustering result analysis according to the household’s label(region,tariff policy adopted,whether there is abnormal power consumption fluctuation,etc.),and some reasonable suggestions are put forward,It is of great significance for formulating personalized energy services and optimizing energy use strategies.(2)In order to further improve the accuracy of household load forecasting,a household short-term load forecasting method based on state frequency memory network is proposed.This method decomposes the memory state into multiple frequency components by introducing discrete Fourier transform.Each frequency component can simulate the specific frequency behind the household load fluctuation,and then predict the future power consumption through the nonlinear mapping of the combination of these frequency components by inverse Fourier transform.This method effectively improves the accuracy of household short-term load forecasting by adding the frequency information of household power in load forecasting.The mean square error(abbr.MSE),root mean square error(abbr.RMSE)and mean absolute error(abbr.MAE)were used to evaluate the proposed model.Taking the load forecasting of the next day for example,comparing the results obtained by LSTM,which behaves the best in this field,with those obtained by the proposed model,the error of forecasted results of three kinds of household power load has reduced by21.6%,11.4% and 15.4% respectively,thus,the effectiveness of the proposed model is fully verified.(3)In the multivariable load forecasting,considering weather,temperature and other factors,the existing models cannot selectively focus on some specific driving sequences.To solve this problem,a recursive neural network with two-stage attention mechanism is proposed for household power load forecasting.By introducing the input attention mechanism and time attention mechanism,the model can adaptively select the most relevant input features and remember the long-term dependence of time series.In the household load forecasting of the next day with a time window of 8,compared with the SVR model,the three kinds of errors of the model are reduced by39.1%,21.9% and 37.4% respectively,and compared with the encoder decoder model without attention mechanism,the three kinds of errors are reduced by 25.1%,13.6%and 28.7% respectively.Compared with other models,the model shows good performance in multivariable household power load forecasting.Finally,all the studies in this paper use the real data provided by the power marketing department of a city.Experiments verify the effectiveness of the proposed model,which has important practical significance.