基于改进BP神经网络的火电厂实时数据预测模型研究

提出了一种基于改进BP神经网络的火电厂实时数据预测模型,即在标准BP算法中引入动量因子和自适应学习速率,以减少收敛振荡过程,加快学习速度。选用某电厂300MW机组主给水流量实时数据进行网络训练学习和校核,分析了输入和隐含层节点数、学习样本数和动量因子对模型预测精度的影响。实例分析表明,该模型有较好容错性,能满足火电机组性能分析的要求。

STUDY ON REAL-TIME DATA-FORECASTING MODEL BASED ON MODIFIED BP NEURAL NETWORK FOR THERMAL POWER PLANTS

A real-time data forecasting model based on modified BP neural network for thermal power plants has been put forward, that means momentum factors and adaptive learning rate have been introduced into standard BP algorithm, so as to reduce the process of convergence oscillation, and to speed up the learning speed. Selecting real - time data of feed-water flow rate of 300 MW unit in one power plant to train the said network for learning and verifying, the influence of nodal point number, learning specimen number, and momentum factors in the input and hidden layers upon forecasting precision of said model has been analysed. Practical examples show that the said model to boast better error - tolerant performance, can satisfy the requirements of performance analysis for thermal power units.