基于LSTM的煤场热值损失多因素线性回归预测

Multi-factor Linear Regression Prediction for Heat Value Loss of Coal Yard Based on LSTM

针对火力发电厂煤场热值损失问题,提出了一种基于长短期记忆(LSTM)的多因素线性回归综合热值损失预测模型。使用LSTM神经网络预测煤场所在地的环境因素,建立煤堆热值变化与时间累积下各因素总值之间的回归方程,代入预测因素的影响值以计算未来的热值损失。在煤场进行实地数据采样,对高位热值进行同值灰分处理,利用所记录的环境因素数据对因变量热值进行多因素回归分析,建立热值损失数学模型。通过构建LSTM神经网络预测多因素变化,将预测数据导入回归模型进行热值变化的预测。仿真结果表明:多因素LSTM神经网络对环境因素的预测精度要高于单因素及传统差分自回归移动平均(ARIMA)模型,并且具有较强的稳定性。

To solve the problem of the heat value losses in the coal yards of thermal power plants, a comprehensive multi-factor linear regression model for heat value loss prediction based on long short-term memory(LSTM) neural network was proposed. The environmental factors in the location of the coal yard were predicted by LSTM neural network, and a regression equation between the change of coal pile heat value and the total value of various factors within the cumulative time was established, so as to calculate the heat value losses in future through considering the influencing values of predictive factors. With the on-spot collected data of the coal yard and after the ash content equivalence processing on the high heat value, analyses were conducted on the multi-factor regression for the dependent variable heat value by the recorded environmental factor data, while a mathematical model for heat value loss was established. Based on the construction of LSTM neural network to predict the multi-factor change, the change of heat value was predicted through importing prediction data into the the regression model. Simulation results show that the prediction accuracy for environmental factors of multi-factor LSTM neural network is higher than those of single factor and traditional autoregressive integrated moving average(ARIMA) models, and the multi-factor LSTM neural network has a strong stability.