Research On Office Building Electricity Forecasting Based On Stacking Model Fusion

In modern society,as the economy grows,the demand for electricity is gradually increasing,so it is necessary to continuously improve the differential electricity price in the sales link,realize the medium and long-term electricity transaction,and encourage electricity users,electricity users and power generation companies to coordinate through centralized bidding Market-oriented operation.Due to the large scale of electricity consumption in office buildings,if you participate in electricity market transactions according to the rules,you need an effective electricity forecast to declare electricity demand.When the deviation between the declared electricity demand of the office building and the actual electricity consumption is within 2%,then there is no need to bear the deviation fee,otherwise,the electricity fee needs to be paid according to the prescribed compensation unit price.Therefore,in response to this problem,this paper studies the electricity forecasting from the aspects of data preprocessing,feature engineering and data mining algorithms to improve the accuracy of electricity forecasting models.In this electricity forecasting study of office buildings,the absolute average percentage error between the actual electricity consumption and the forecast electricity consumption in a month is used for evaluation.In terms of forecasting methods,the forecasting model predicts electricity consumption in hours,and then aggregates the forecast results in months to realize the demand for forecasting monthly electricity consumption.Secondly,the overall design of the modeling steps,from data preprocessing to feature engineering to the selection of algorithms,and finally to compare the effects of prediction models,select the best model to achieve prediction.The implementation steps are first to read the csv file data to get the predicted label,and then use the web crawler technology to crawl the meteorological data on the meteorological website,and extract the time features from the time data,and then use Python and related libraries pandas and sklearn to carry out these data preprocessing,using the recursive feature elimination method for feature selection on the preprocessed data to obtain the optimal feature subset.This research no longer uses the traditional method of single model modeling and prediction,but uses the stacking model fusion method.At the same time,in the selection of the first-layer basic learner of Stacking model fusion,the previous practice of using several similar algorithm fusions is discarded.After experimental comparison,a combination of algorithms with high degree of difference and strong learning ability are combined to make the prediction of Stacking model fusion reach the optimal effect.The electricity forecasting research based on the integration of the Stacking model has been verified after the company has launched it.It has a certain reliable reference value for the company’s subsequent electricity transaction,and it can save the power supply operation cost of the office building itself and has a certain significance for the power supply stability of the area where the office building is located.