双重BP神经网络组合模型在实时数据预测中的应用

在回归和延时神经网络的基础上，利用非线性组合预测方法的优点，提出一种新的预测模型--双重BP神经网络组合模型模型，选用某660MW机组的主蒸汽流量数据进行学习训练，实例计算结果表明双重BP神经网络组合模型可提高单项预测模型的精度，校核样本的平均相对误差为1.5%，而单独采用回归神经网络和延时神经网络进行预测的平均相对误差分别为2.7%和1.9%，证明双重BP神经网络组合模型具有很高的预测精度，可应用于火电厂实时数据的有效性验证。

The Application of Double BP Neural Network Combined Forecasting Model in Real-time Data Predicting

As a new data forecasting model,the double BP neural network combined model based on regressive neural network and time-delay neural network was proposed, which has the advantages of nonlinear combined forecasting methods. The model was trained with main steam flow values of one 660MW unit.The calculation results testified that the double BP neural network model improved the forecasting accuracy of a single model.The mean relative forecasting error of checking samples data was 1.5%, while the mean relative forecasting errors of regression neural network and time-delay neural network were 2.7% and 1.9% respectively.It was proved that the double BP neural network combined model had preferable forecasting accuracy and can be applied for validation of real-time data in power plant.