基于DWT-CNN-Informer模型的液压支架压力多步长预测  

Multi-step prediction of hydraulic support pressure based on DWT-CNN-Informer model

作　　者：张传伟[1,2] 张刚强 路正雄 李林岳 ZHANG Chuanwei;ZHANG Gangqiang;LU Zhengxiong;LI Linyue(School of Mechanical Engineering,Xi’an University of Science and Technology,Xi’an Shaanxi 710054,China;Shaanxi College of Communications Technology,Xi’an Shaanxi 710018,China)

机构地区：[1]西安科技大学机械工程学院,陕西西安710054 [2]陕西交通职业技术学院,陕西西安710018

出　　处：《中国安全生产科学技术》2025年第4期57-63,共7页Journal of Safety Science and Technology

基　　金：陕西省重点研发计划项目(2022GD-TSLD-63,2022GD-TSLD-64);陕西省教育厅项目(23JP100)。

摘　　要：为了实现液压支架压力多步长精准预测,提出1种基于DWT-CNN-Informer模型的压力多步长预测方法,该方法利用离散小波变换(discrete wavelet transform, DWT)将预处理后的压力时序数据分解为趋势项和周期项频率分量;各频率分量输入卷积神经网络(CNN)模型提取频率特征;提取的频率特征输入Informer编码器,经位置编码和多头概率稀疏自注意力机制捕捉时序变化特征,并结合自注意力蒸馏减少特征冗余;将Informer解码器改为全连接层,直接输出各分量多步长预测结果;重构叠加各分量多步长预测结果得到液压支架压力多步长预测结果。研究结果表明:在预测步长分别为6,12,24时,DWT-CNN-Informer模型相比LSTM、Informer、CNN-Informer模型在平均绝对误差(MAE)、均方根误差(RMSE)、对称平均绝对百分比误差(SMAPE)指标上均表现出更高预测精度。研究结果为液压支架压力精准预测提供有效方法。In order to realize the accurate multi-step prediction of hydraulic support pressure,a multi-step pressure prediction method based on DWT-CNN-Informer model was proposed.The discrete wavelet transform(discrete wavelet transform,DWT)was used to decompose the preprocessed pressure time series data into the trend and period term frequency components,and each frequency component was input intothe convolutional neural network(CNN)model to extract the frequency features.The extracted frequency features were then input intothe Informer encoder,which captured the time series change features by position encoding and the multinomial probabilistic sparse self-attention mechanism,andthe feature redundancy was reduced combining with the self-attention distillation.The Informer decoder was changed to a fully connected layer,which directly output the multi-step prediction results of each component.The multi-step prediction results of hydraulic supportpressure were obtained by reconstructing and superimposing the multi-step prediction results of each component.The results show that the DWT-CNN-Informer modelpresents higher prediction accuracy in terms of mean absolute error(MAE),root mean square error(RMSE)and symmetric mean absolute percentage error(SMAPE)compared with the LSTM,Informer and CNN-Informer modelsunder the prediction steps of 6,12 and 24.The research results provide an effective method for the accurate prediction of hydraulic support pressure.

关 键 词：液压支架压力 多步长预测 离散小波变换 CNN模型 Informer模型 

分 类 号：X936[环境科学与工程—安全科学]