Household Electricity Load Forecasting Based On Prophet And Machine Learning

Today,more and more attention has been paid to electric power energy,and more and more new energy enterprises have emerged,increasing the problems of supply and consumption of power energy,so the work of electricity load forecasting has become a challenge that needs to be solved nowadays.In reality,load forecasting needs to be optimised repeatedly,each improvement in accuracy followed by the efforts of researchers.With the advent of various models,numerous new forecasting methods have been introduced,which greatly improve the weakness of traditional time series methods for power forecasting problems.At the same time,with the development of machine learning,all kinds of models are well adapted to the characteristics of wide sources,large quantities and varieties of power big data.In this paper,we use the features of Prophet model and machine learning model to construct a fusion model and combine meteorological and temporal features to model the prediction of electric load data.Firstly,an exploratory analysis of the dataset was conducted to determine the medium-term electricity load forecast.By pre-processing and feature engineering,on the one hand,abnormal data were processed and normalized,changing the time dimension from hour to day;on the other hand,the meteorological factors were determined by heat map and ranking importance,so as to effectively reduce the dimensionality of the data.On this basis,the data of the last two months were used as the test set and the rest as the training set.The Prophet model was used in training set,and each decomposition item was analyzed,and then the test set was preliminarily predicted by combining the advantages of Neural Prophet model.In order to investigate the impact of meteorological factors on load data,this paper used four single machine learning models for prediction,and found that the fitting effect of single model was poor,indicating that the meteorological data was not sufficiently interpreted and the connection between the data before and after the electricity load data was not considered.In order to solve the problem of single model prediction,four fusion models based on Prophet model and four single machine learning models were proposed,by processing the data,studying the connection between the residual terms of the models and the real data,modelling the prediction of the residual terms,and the effect of the fusion models were generally better than that of the single model.Among the fusion models,Prophet-XGBoost model has the best fitting,with the MAPE value of 9.26%,The MAPE value of the other fusion models were also lower than that of the single model.Finally,the stability of the Prophet-XGBoost model was verified by crossvalidation method,indicating that the model was feasible.