基于BP神经网络的火电厂实时负荷优化分配_李勇.pdf

第 39 卷 第 17 期 电力系统保护与控制 Vol.39 No.17 2011 年 9 月 1 日 Power System Protection and Control Sept. 1, 2011 基于 BP 神经网络的火电厂实时负荷优化分配 李 勇，王建君，曹丽华 （东北电力大学能源与动力工程学院，吉林吉林 132012 ） 摘要：满足电网调度要求的实时分配方法是火电厂负荷优化分配得到推广应用的关键，综合考虑了燃料成本及污染物排放成 本对传统数学模型进行改进。根据 BP 网络算法中误差函数、权值、阈值的计算公式推导出一种能够真正达到自适应的学习 率对算法进行了改进，引入多变异位自适应遗传算法计算样本数据对 BP 网络进行训练。分析了实时煤耗特性曲线的获取， 并探讨了基于BP网络的实时负荷优化分配方法的实现。实例计算结果：10组样本测试结果的平均相对误差仅为0.185%，且 平均耗时仅为0.011s，表明了该方法在降低发电成本的同时，可有效地缩短优化计算时间。 关键词：火电厂；负荷优化分配；实时调度；BP神经网络；多变异位自适应遗传算法 Real time optimal load dispatch of power plant based on back propagation neural network LI Yong ，WANG Jian-jun ，CAO Li-hua （School of Energy and Power Engineering，Northeast Dianli University ，Jilin 132012 ，China ） Abstract ：Real time dispatching method which can meet the requirements of power network dispatching is critical to use load optimal dispatch of power plant widely. The costs of fuel and pollutant emissions are comprehensively considered to improve the traditional mathematical model. According to the calculation formulae of error function ，weight and threshold in BP network algorithm ，the learning rate which can really reach adaptation is derived to improve the algorithm. Meanwhile，BP network is trained by the sample data that multiple mutation adaptive genetic algorithm is introduced to train the BP network. The acquisition of real time coal consumption characteristic curve is analyzed，and the implement of real time load optimal dispatch system based on BP network is discussed. The example calculation result is that the average relative error of 10 sets testing samples is only 0.185 percent and the average time consuming is only 0.011 s. It shows that the power generation cost is reduced，as well as the o