Short Term Probabilistic Load Forecasting For Customer Side Based On Behavioral Characteristics

With the ongoing development of the electricity market transaction and construction of smart grid,behavior information and load forecasts from customer side will lay great foundation for ensuring the safe and economic operation of the smart distribution network,enhancing energy efficiency,and scientifically developing the mechanism for demand response.Based on this background,this thesis paper gets the help from the massive fine-grained electric load data provided by smart meters,and proposes a behavioral-characteristic-based short-term probabilistic forecasting framework for customer side.Under this framework,several issues,such as load profiling,hierarchical load point forecasting and probabilistic customer-side load forecasting,have been formally studied.Specifically,this thesis paper contributes to the following points:1.It investigates the state-of-the-art technique on analyzing customers' behavioral characteristics and probabilistic load forecasting and summarizes the challenges and blind spots in customer-side load forecasting.2.It proposes an analysis framework for customer-side behavioral characteristics,including the meteorological,periodic and holiday characteristics of their electric loads,as well as mining unique electricity patterns with distributed clustering.After that,external factor driving characteristics,typical load profiling and hierarchical structures are obtained.3.Based on predictive models,such as generalized linear regression model,artificial neural network and gradient boosting regression tree,the customer-side differentiated short-term load forecasting models considering behavioral characteristics are constructed.After that,the revised hierarchical point load forecasts,for customers and their aggregated load,are produced with the aggregation constraints of the hierarchical structure.4.Three kinds of probabilistic generation methods from input,model and output component are proposed.With these generation methods,point forecasting models and Copula-framework-based hierarchical probabilistic forecasting strategy,short-term probabilistic load forecasting models for individual and aggregated load is constructed.5.From the perspectives of predictive model,probabilistic generation methods and customer classification,the relationship between customer-side point load forecasts and probabilistic forecasts is comprehensively investigated.It is concluded that high point forecasting accuracy has a consistent relationship with probabilistic forecasting accuracy.Taking the Irish residents and Shanghai industrial and commercial customers as case study,the feasibility and effectiveness of the customer-side load forecasting modeling strategy,model applicability and probabilistic load forecasting schemes are formally analyzed and validated.