燃煤电站锅炉优化配煤神经网络构建的研究  

Study on Construction and Predication of Neural Network in Coal-Blending in Coal-fired Power Station

作　　者：吴江[1] 潘雷[1] 潘卫国[1] 任建兴[1] 郑莆燕[1] 刘莉[1] 黄锦杰[2] 

机构地区：[1]上海电力学院能源与环境工程学院,上海200090 [2]华能广东分公司汕头电厂运行部,广州汕头515071

出　　处：《锅炉技术》2010年第1期1-4,8,共5页Boiler Technology

基　　金：国家自然科学基金(50806041);上海市"曙光计划"(08SG53);上海市教委科研创新重点项目(09ZZ184);上海市地方能力建设项目(09230501300);上海高校本科高地建设项目资助

摘　　要：随着电力工业快速增长,煤的消耗量增加,配煤逐步成为燃煤电站锅炉的一种常态。由于煤种煤质的波动,配煤是一个在不确定的条件下的优化问题,传统的线性规划模型很难解决,而BP神经网络这一非线性优化工具已成功地应用于混煤煤质特性的预测模型。分析了不同BP神经网络模型的预测效果以及制约其预测效果的主要因素(网络结构、学习样本数量、隐层节点数、学习精度)并在此基础上构建了用于预测混合煤种低位发热量的BP神经网络,得到极高的预测精度,进行了3次实验误差分别为0.043 05%,0.038 05%,0.116 267%。同时利用穷举法对配煤进行优化选取。本神经网络的结构是由输入、输出样本的数据结构决定的,具有很强的通用性和拓展性。Along with the rapid development of our country＇s industry, the factors in duce that the power plant combusts the single coal impossibly for a long time, which are the increase wastage of coal, the difficulty transportation, the asymmetric distributing of coal mine and the policy of nation about that power boiler combusts inferior coal, therefore, the power plant must combusts blended coal. As blending of coal-doped fluctuations, is a blending under conditions of uncertainty in the optimization problem. The traditional linear programming model that has not solve nonlinear problems, and BP neural network this non- linear optimization tool has been successfully applied to mixed coal, coal quality of prediction model. This paper describes the BP neural network theory, analysis of different BP neural network model of prediction and control of its forecast results the main factor （Network structure, study sample size, hidden nodes, learning accuracy）. And on this basis, build a BP neural networks in forecasting for the low heat of the mixed coals BP neural networks, and to very high precision. In this paper, carried out three experimental error was 0. 043 05%, 0. 038 05%, 0. 116 267%o. At the same time,the use of coal exhaustive method to optimize selection. As the neural network structure by the input and output data structure for the decision, therefore, has a strong expansion.

关 键 词：BP神经网络 优化配煤 穷举法 

分 类 号：TK229.6[动力工程及工程热物理—动力机械及工程]