应用级联神经网络预测供热锅炉次日小时热负荷的初步研究

通过对供热锅炉房热负荷的分析，建立了基于两个BP网络的级联神经网络(CNN)。相关性分析表明，可将时间序列负荷数据作纵横向分离，横向相关系列负荷可作为CNN前一BP子网络的输入数据，纵向相关系列负荷可作为CNN后一BP子网络的输入数据。前一BP子网络用于小时负荷的初始预测，其预测结果加入后一BP子网络的输入系列，实现对负荷的精确预测。按照此模型，建立了某一印染厂锅炉房次日小时蒸汽负荷的CNN预测模型。程序运行结果表明该模型在预测时足够准确可靠。

Initial Investigation of Cascaded Neural Networks in Hourly Thermal Load Forecast of Next Day for Boiler Plant

A model of cascaded artificial neural network (CNN) is proposed, constructed by two back-propagation (BP) neural networks, with analyzing properties of the hourly thermal loads of boiler plants. Correlation analysis showed that the time-series load data of a boiler plant could be classified into two categories, right-and-left and pre-post series. In building up the CNN model, the former could thus be the input data to the first BP network and the latter be the input data to the second one. The first BP network is used to forecast preliminarily hourly loads that are also part of input to the second BP network, while the second is used for accurate hourly load forecasting. In this way, a CNN model was completed for a boiler plant in a printing and dyeing mill. The result of running the program indicates that the CNN model has big flexibility, and the thermal loads forecasted are accurate enough.