Abstract:

With the development of machine learning， neural network technology has become a solution to making analysis and prediction on massive data in coal-fired power plants. To facilitate the dispatching of power resources in power grid， it is key to predicting power unit output. Having modeled a 300 MW CFB based on BP neural network， the impact of abnormal data on the model accuracy was alleviated through data preprocessing， and the number of input variables in principal component analysis（PCA） was reduced through dimension reduction. Relative error and root-mean- square error between the output value and expected value under different numbers of hidden neurons were compared and analyzed in an experiment， and seven hidden neurons were selected for their comprehensive advantages. Studying the corresponding data of the unit in the experiment， the results show that the relative error between the output value and the expected value obtained by the model is around ±2%， and the root-mean-square error is around 13.4 MW. Therefore， this model has good accuracy and stability in power unit prediction.

Key words: BP neural network, power unit output, prediction model, relative error, machine learning