Multi-scale Electricity Consumption Prediction Method Based On Adaptive Ensemble Model

Long-term and short-term electrical energy prediction plays an important role in energy management,power plant scheduling,peak demand and grid security conflict.In the past two decades,a variety of data-driven models have been widely used in the prediction of building and large-scale electricity consumption,and achieved satisfactory results.However,for different time and energy scale scenarios,there is no mature prediction scheme with strong generalization ability and high universality.In this paper,an adaptive ensemble prediction model is proposed to deal with different scale electricity consumption prediction tasks.The method includes three main parts:data preprocessing,primary prediction and secondary prediction:Firstly,data preprocessing includes rough selection and machine learning based selection.Rough selection is to collect corresponding characteristic variables according to expert experience.Machine learning based selection uses recursive feature elimination(RFE)and fuzzy c-means clustering(FCM)to further process the roughly selected data.RFE is used for feature selection and FCM for data classification.Secondly,the primary prediction part includes a model library,which contains a variety of typical data-driven models.The five data-driven models selected in this paper are: back propagation neural network(BP),support vector machine(SVM),extreme learning machine(ELM),random forest(RF)and bayesian linear regression(BLR).In order to improve the prediction accuracy of a single model,an algorithm package composed of multi-swarm intelligent evolutionary algorithms(EAs)is used to optimize the key parameters of the prediction model.The algorithm package includes three algorithms: genetic algorithms(GA),particle swarm optimizations(PSO)and teaching and learning optimizations(TLBO).The choice of optimization algorithm depends on the comparison of prediction performance.In the third part of the secondary prediction,the output of the primary prediction is weighted by the linear regression model to provide the final prediction result,and each weight is also selected by the optimization algorithm.In order to verify the effectiveness of the proposed ensemble prediction model,this paper compares and studies three different cases.These three cases represent prediction scenarios of different time and energy scales.Case A performs hourly electrical load prediction of a whole building from University of Wyoming,USA;Case B is a city-scale daily electrical load forecasting(Yizheng City,Jiangsu Province,China),and in Case C,national wide monthly electrical load prediction of the United States is carried out.Results of the three cases indicate that:(1)Compared with five single prediction models,the ensemble model proposed in this paper has the best prediction accuracy in all cases;Compared with the best single prediction model,the evaluation index MAPE of the ensemble model decrease from 3.21 to 2.65;(2)Compared with previous reported ensemble models,the evaluation index MAPE of the proposed model also decrease from 1.37 to 1.13.From the case results,the adaptive ensemble prediction model proposed in this paper has strong generalization ability and prediction accuracy,and is a general multi-scale energy forecasting scheme.