改进模糊神经网络在负荷预测中的应用研究

Research on the Use of Improved Fuzzy Artificial Neural Network in Load Forecasting

电力系统短期负荷预测是电网调度中一项重要的工作,精确的负荷预测可以为调度员提供必要的基础数据,电网运行安排也都是以负荷预测的数据作为根据。利用人工神经网络可以任意逼近非线性系统的特性,将其用于短期负荷预测。研究了改进的误差反向传播算法——动量及自适应lrBP的梯度递减训练算法,预测结果表明比标准BP算法具有更好的性能。同时,针对大量无法用精确数值来量化的信息的影响,引入模糊理论的方法,定义了不同的隶属度函数,模糊化后输入到网络中进行训练和预测,结果表明其精度比输入量非模糊化的人工神经网络更高。

Short term load forecasting is an important work in the power network dispatch. Accurate load forecasting provides necessary basic data for the dispatcher. The data of load forecasting is also used as the basis to the power network operation arrangement. For the characteristics of artificial neural networks which can approximate to nonlinear systems arbitrarily, in this paper,it is used for short term load forecasting. The improved error back-propagation algorithm - gradient descent with momentum and adaptive learning rate back-propagation training algorithms is researched. The results of forecasting show that it has better performance than that of the standard BP algorithm. At the same time, for the impact of a lot of information which can not be quantified by precise number, the fuzzy theory method is introduced to this method. Definitions of different membership functions are made. After being fuzzy,they are inputted to the network for training and forecasting. The results show that it is more accurate than the artificial neural network with non-fuzzy input.