关键词:小波神经元网络;隶属度;短期负荷预测;电力系统
SHORT-TERM LOAD FORECASTING BASED ON WAVELET NEURAL NETWORK
ABSTRACT:Wavelet neural network (WNN) possesses more degree of freedom and better adaptivity than multi-layer FP neural network. To better reflect the influence of climate factors on load and improve the precision of load forecasting, the Morlet wavelet is chosen to establish a wavelet neuron network, the back propagate algorithm is adopted to train the WNN network, a new method of analyzing clustering by self-study membership is used to train the samples. The load data and climatic data of Wuhan power network in recent years are applied in modeling and load forecasting. The forecasting results show that the established WNN model possesses better convergence and the forecasting precision can be improved by choosing training samples with analyzing clustering by self-study membership.
KEY WORDS:Wavelet neural network;Membership;Short-term load forecasting;Power system
1 引言
短期负荷预测是负荷预测的重要组成部分,是电力系统运行调度中的重要内容。国内外已提出了多种短期负荷预测方法,如多元回归、ARMA模型、人工神经元网络方法等。可归类为:①利用负荷的自身发展规律,如ARMA模型[1]等;②负荷发展规律与气象因素相结合,如ANN(Artificial Neural Network)方法[2];③其他方法,如小波分解法[3-5]、模糊聚类法[6]及混沌算法[7]。
人工神经网络以其强大的多元性映射能力能够准确捕捉并学习负荷值与天气之间的非线性关系,使考虑气象因素的电力系统短期负荷预测成为可能。近年来它一直受到密切关注,且已成为解决电力负荷预测问题的有效计算工具。小波在分析非固定信号和构造非线性函数模型方面具有卓越性能,因此结合了小波基函数的小波神经元网络(WNN)比一般神经网络具有更多的优越性。
为更好地反映气象因素对负荷的影响及提高负荷预测的精度,本文构建了一种小波神经元网络负荷预测模型,以Morlet小波取代Sigmoid函数,采用误差反传学习算法来训练网络,采用自学习隶属度分析聚类方法来选择训练样本。
2 小波及小波变换
基本小波或母小波定义为满足相容性条件(如式(1)所示)的平方可积函数φ(t)∈
注册
个人空间
