关键词: 神经网络, 非线性自回归, 时序预测, 工业大气污染

Abstract: For the weak prediction ability of the exhaust gas in the industrial park management, a prediction method of industrial exhaust gas concentration is proposed considering the multiple factors coupling and its nonlinear timing characteristics. A Compound Nonlinear Auto Regressive（CNAR） neural network is proposed based on the numerical characteristics of the predictive indices. The relationship between target prediction index and related influencing factors are used to model the correlation and temporal relationship. Then the short-term prediction of exhaust gas concentration is realized. The atmospheric grid monitoring data of a city in Hebei province is used to train model and verify method, and the experimental results show that the CNAR forecasting model can predict the gas concentration effectively in a short time. The prediction accuracy is higher than the traditional auto regressive neural network, and this method provides a reference basis for the exhaust gas control in industrial park.

Key words: neural network, nonlinear auto regressive, time series forecasting, exhaust gas of industrial park