基于神经网络和模糊理论的短期负荷预测

Short-term Load Forecasting Based on Artificial Neural Network and Fuzzy Theory

电力系统负荷预测是能量管理系统(EMS)的重要组成部分,它对电力系统的运行、控制和计划都有着非常重要的影响,提高电力系统负荷预测的准确度既能增强电力系统运行的安全性,又能改善电力系统运行的经济性,但负荷预测的复杂性、不确定性使传统的基于解析模型和数值算法的模型难以获得精确的预测负荷。为提高电力系统短期负荷预测准确度,构建了一种新型的负荷预测模型。该模型首先采用多层前馈神经网络,以与预报点负荷相关性最大的几种因素作为输入因子,以改进BP算法作为预测算法,来获得预报日相似日负荷曲线;然后引入自适应模糊神经网络,用于预测预报日的最大、最小负荷;针对模糊神经元的权值更新问题,采用一种新的权值更新算法———一步搜索寻优法,在获得预报日相似日各点负荷和最大、最小负荷的基础上,通过纵向变换,对预报日的负荷修正,进一步减小预测误差。用上述模型和算法预测某地区电网的短期负荷,取得了良好的预测效果。

Electric power system load forecasting plays an important role in the Energy Management System （ EMS）, which has great influence on the operation, controlling and planning of electric power system. A precise electric power system short term load forecasting will result in economic cost saving and improving security operation condition. With the development of deregulation in electric power system, the method of short term load forecasting with high accuracy is becoming more and more important. Due to the complicacy and uncertainty of load forecasting, electric power load is difficult to be forecasted precisely if analysis model and numerical value algorithm model is applied. In order to improve the precision of electric power system short term load forecasting, a new load forecasting model is put foreword in this paper. First, according to the features of electric power load and considering the combined influence of historical load data, weather and day type, Multi-layer Feed Foreword Neural Network （MFNN） and improved Back Propagation （BP） algorithm are adopted here to obtain the similar daily load curves of forecasting day. Then, an Adaptive Fuzzy Neural Network {AFNN} architecture based on the adaptive multi-layer structure with fuzzy neurons in its layers which integrates the merits of fuzzy logic and neural network is designed to forecast the daily max/min load of forecasting day. A novel algorithm-a pace search for optimum to renew the weight value of fuzzy neurons is proposed, which solves the difficult problem of the parameters renewal in fuzzy neurons. Finally, vertical direction variation is applied to amend the load of forecasting day, which will furthermore reduce the forecast error effectively. The application of proposed short term load forecasting method to some areas power system demonstrates its better performance.