融合多时间尺度特征的焦炉煤气发生量预测模型

Coke oven gas generation prediction model integrating multi-time scale characteristics

中文关键词  焦炉煤气发生量预测  多时间尺度  滑动窗口逐步分解  经验小波变换  长短期记忆网络

英文关键词  coke oven gas generation prediction  multiple time scales  sliding window stepwise decomposition  empirical wavelet transform  long short-term memory network

中文摘要

      焦炉煤气发生量的准确实时预测是实时监测焦炉生产状态和煤气调度的重要参考. 本文建立了融合多时间尺度特征的焦炉煤气发生量预测模型, 以实现焦炉煤气发生量的有效预测. 首先, 对焦炉煤气发生过程进行特性分析; 然后, 建立滑动窗口逐步分解模式, 在此基础上用经验小波变换对焦炉煤气发生量进行分解, 并根据样本熵对分量进行重构, 使用长短期记忆网络建立重构分量的预测模型; 最后, 利用实际现场数据进行实验. 实验结果显示,对于平均绝对百分比误差指标, 本文所提方法的预测精度达到0.29%, 比单一长短期记忆网络模型提高了0.3%, 相较于逐步分解模式提高了0.22%. 结果验证了所提方法的可行性与有效性.

英文摘要

      Accurate real-time prediction of coke oven gas generation is an important reference for real-time monitoring of coke oven production status and gas scheduling. In this paper, a coke oven gas generation prediction model incorporating multi-time scale features is developed to achieve effective prediction of coke oven gas generation. Firstly, the coke oven gas generation process is characterized, then a sliding window step-by-step decomposition model is established, based on which the coke oven gas generation is decomposed by empirical wavelet transform, and the components are reconstructed according to the sample entropy, and the reconstructed components are reconstructed by using a long and short-term memory network to build a prediction model, and then experiments are carried out by using the actual field data. The experimental results show that the prediction accuracy of the proposed method reaches 0.29% for the average absolute percentage error,which is 0.3% higher than that of the single short-term memory network model, and 0.22% higher than that of the stepwise decomposition model. The results verify the feasibility and effectiveness of the proposed method.