Load Forecasting For Residential Electricity Under The Increasing-block Pricing Tariffs

In order to encourage residents to save electricity,maximize resource utilization and alleviate the energy pressure,a new pricing mechanism,namely increasing block tariff(IBT),for electricity has been advocated and implemented in China.Due to the introduction of the IBT,different types of consumers tend to have different responses for this new tariff.This would result in the change of their consumption behaviours.As a result,the load forecasting becomes more complicated and challenging under the IBT.In an effort to remedy the above pressing issues,this work proposes an innovative clustering-based load forecasting approach to support load forecasting for residential electricity under the IBT in China.The new approach not only can identify different consumption patterns under the IBT,but also can produce better prediction accuracy.To some extent,the research results can provide some feedback and improvement suggestions for the IBT,which could be used to support the IBT design.In addition,it could help the power company to provide reliable and high-quality electricity supply,ensuring safe operation of the power system.The proposed approach innovatively considers the IBT-related attributes in consumer segmentation,and employs fuzzy C-means clustering analysis to group the residents into different segments and capture the consumption patterns of different resident groups.Secondly,for each consumer segment,different load forecasting models are built to predict the usage,respectively.Then the predicted usages of different clusters are aggregated to derive the total usage.In particular,consumer segments can select the most appropriate model based on the features of their consumption behaviours.In this work,Autoregressive Integrated Moving Average Model(ARIMA),support vector regression(SVR),and the self-organising fuzzy neural network(SOFNN)are employed to forecast the short-time load for each segment.Finally,by comparing the model performances of different prediction algorithms,individual segment can select the most accurate predicted result.The winner results are then aggregated to derive the combined predicted usages.By using the hybrid model,the prediction performance is improved than using a single prediction algorithm for all clusters.In order to verify the effectiveness and utility of the proposed approach,an empirical study is conducted in this research.A realistic dataset which includes the daily electricity usage data of 533 households spanning the period from April 2014 to February 2015 is collected.This dataset collects from a certain area in Quanzhou city,Fujian Province in China.To the end,533 households are classified into five clusters with distinctive consumption patterns,including low-demand users and insensitive to high temperature,ordinary users and sensitive to high temperature,ordinary users and sensitive to the IBT,high-demand consumers,luxury consumers.Then an overall forecasting model,a clustering-based forecasting model,and a hybrid forecasting model are built for the short-time load forecasting respectively,and their prediction performances are compared and discussed.The obtained experimental results demonstrate that,the hybrid model not only can produce better prediction accuracy,but also can provide better flexibility for hybrid modeling,which can be used for load forecasting with different characteristics.