Research And Implementation Of Short-term Load Forecasting Based On Neural Network

Short-term load forecasting is an important component of the Energy Management System (EMS), and it is an important sector of the day-to-day work for the scheduling of power system operators. The level of prediction accuracy of its direct have high impact on the security, economy and quality of the operation of power system.This thesis describes the causes of the historical load data and the impact of load forecast, proposes a difference method based on the idea of Differential in order to amend the bad data, and eliminate the curve burr. In-depth analysis of the cyclical changes of power load and the short-term load based on the impact of summer temperature, the thesis designs a short-term load forecasting model based on BP neural network , considering the temperature factor. The model selected the Momentum -adaptive learning rate adjustment algorithm to improve BP network, which effectively improved the convergence rate of BP network, and inhibited the BP networks vulnerable to the shortcomings of the local minimum point .And it uses the unique temperature treatment which take full consideration to the load impact of the temperature .By using Hunan historical load data, and compared with the temperature factors not included in neural network prediction model, it shows that the prediction model can effectively improve the accuracy of short-term load forecasting, with good practicability and feasibility.Finally, short-term load forecasting software is designed and implemented. The software system uses B / S model, is developed based on .Net platform. The forecast results of the system are satisfactory. In actual use, it effective mitigated the pressure of workers of power companies whose work is load forecasters predict. And it gets good appraisal of the users.