Method Of Large Consumer Load Forecasting And Its Application

With the development of economy and the adjustment of industrial structure,enterprise equipments upgrade to be high-capacity and high-parameter devicesgradually, leading to higher voltage level and larger load of the big consumers. Sincelarge consumers occupy a high proportion in regional load, its fluctuations wouldimpose obvious impacts on the regional load, increasing the difficulty of forecastingrelated load. Therefore, effective large consumer load forecasting is a key toimprove the entire load forecasting accuracy and also is of great significance to thesafety and economy of the power grid operation.Through analyzing load curve and daily electricity consumption curve ofregional large consumers, the conclusion that the load of large consumers, which hasgreat fluctuation and randomness is made. Meantime, the influence frommaintenance activities of enterprise equipments on load is very big. Besides,although the load of large consumers could not be affected by factors such as theclimate, the weather and the holidays, it has close relationship with productiontechnology and schedule of enterprises. On the basis of production process,production equipment and load curve of enterprises including iron and steel, cement,chemical fiber and paper-making, their load characteristics are analyzed.The grey model, BP neural network, generalized regression neural network andgrey neural network are utilized separately and unitarily to forecast the load of largeconsumers. Simulation results show that general algorithm could achieve less idealperformance due to great impact, randomness and fluctuation of the load, whereasthe combined forecast can eliminate load fluctuations of large consumers amongeach other and enable the load to be more smoothly to obtain higher predictionprecision.In view of the deficiency that the parameters chosen for traditional vectormachine (SVM) have great impacts on the prediction result, particle swarmoptimization algorithm(PSO) was adopted to optimize the punish parameters and thekernel function parameters. Compared with other algorithms, the outcomes show thatthis algorithm not only accelerates the convergence speed and improves theconvergence precision, but also processes good prediction effect, demonstrating itssuperiority in forecasting load of large consumers.In order to enhance the level of regional short-term load forecasting and control load characteristics of regional typical large consumers, an regional typical loadanalysis and management system for large consumers is designed and developed.Aiming at serving the load of regional large consumers as main targets, the systemputs its core in electric power market demand analysis and forecasting theory,providing an integrated information platform on the basis of the computer, networkcommunication, information processing technology and safety management mode.Based on MyEclipse platform, Oracle10g database and the multilayer systemstructure of Browser/Server (B/S), the system can acquire operation information ofelectric power system automatically and authentically as well as forecast the load oflarge consumers, providing a basis for regional load forecasting.