Research On Forecasting System Of Classified Load Of Power Energy Consumption

Load forecasting is vitally important for power system administrative departments including dispatch, consumption and planning department etc. It is influenced by many factors such as economy, policy, weather and people's production and living, so differences in load forecasting between different periods, areas, developing conditions and consumers' classes are its distinctive characteristic. How to implement load forecasting with high quality and efficiency has always been emphatically paid attention to by the electricity industry and relative scholars. Existing researches mainly focus on synthesis load forecasting other than classified load forecasting, which affects not only accuracy of load forecasting but also administrative departments' efficient management of different types of load. In this dissertation, based on the research of the practical project of Urban Power Supply Bureau, Chongqing Electric Power Corp., "Development and research on forecasting system of electricity sales of classified load", research is made on medium- and long-term forecasting models and methods and a corresponding software is preliminarily developed by fully utilizing accumulated data of monthly electricity sales of classified load and weather data from the bureau. Obviously, this research is of important theoretical and practical value. Main achievements of the research are as following:Basic principles of back propagation neural network (BPNN) and its applications on classified load forecasting are introduced and variety of impact factors such as weather, time, economy, load disturbance are analyzed in detail. Two methods: the bad data pretreatment method and the temperature data fuzzification method are proposed, which lay the foundation of reasonably proposing the neural network sample model of classified load forecasting.Based on the different types of electricity price, nature of power consumption and data source of monthly electricity sales, the synthesis load of Urban Power Supply Bureau is classified into 9 categories. The neural network forecasting model is proposed in full consideration of impact of time, weather and load disturbance on each category of load. Finally, the optimal input model of neural network is determined by simulation and comparison of different combinations of samples.A software of classified load forecasting of Urban Power Supply Bureau is preliminarily developed in Matlab. This software is able to meet the requirement of data input of monthly electricity sales of classified load and enquiry of forecasting result by neural network.