基于LSTM与ARIMA组合模型的高炉煤气产生量预测

Prediction of Blast Furnace Gas Output Based on Combined Model of LSTM and ARIMA

通过分析煤气产生量的数据特点,采用LSTM循环神经网络和ARIMA两种预测模型建立钢铁企业高炉煤气产生量预测模型,经验证LSTM模型性能明显优于BP神经网络,但是其预测结果值比真实值普遍偏低,而ARIMA模型的预测结果值比真实值普遍偏高。基于上述现象,提出了一种基于LSTM与ARIMA组合预测模型,将两种模型的预测结果采用CRITIC方法进行融合处理。结果表明,组合模型明显改善了两种模型在预测特性上的弊端,将预测结果的均方根误差减小为2.325,更贴近真实值,提高了预测精度。

By analyzing the data characteristics of gas production,the model of bf gas production in iron and steel enterprises was established by using the LSTM circulation neural network and ARIMA prediction models.It is proved that the performance of LSTM model is obviously better than that of BP neural network.However,the predicted value was generally lower than the true value,while the predicted value of ARIMA model was generally higher than the true value.Based on the above phenomena,a combined prediction model based on LSTM and ARIMA was proposed,and the prediction results of the two models were processed by CRITIC method.The results show that the combined model significantly improved the defects of the two models in the prediction characteristics,and reduce the root mean square error of the prediction results to 2.325,which is closer to the real value and improve the prediction accuracy.