基于多特征融合的煤自燃温度深度预测模型

Deep prediction model of coal spontaneous combustion temperature based on multi-feature fusion

为了有效预防煤矿采空区煤自燃灾害,提高煤自燃灾害预测模型精度,提出了基于多特征融合的煤自燃温度深度预测模型。首先,通过自编码器网络对煤自燃数据的每一个特征进行降噪处理,增强数据的鲁棒性;其次,按时间序列顺序将降噪后的数据转成二维特征矩阵,采用滑动窗口对特征矩阵进行切片,并采用深度学习中的卷积神经网络提取特征矩阵上的有效特征,进行特征数据融合,在降噪和特征提取与融合的过程中采用差分进化算法对降噪自编码器和卷积神经网络的参数进行优化;最后,将融合后的数据输入门控循环单元神经网络进行煤自燃温度预测。实验结果表明,降噪后和特征融合后的数据在平均绝对误差上比直接采用原始数据预测误差分别降低6.55%和69.26%,均方根误差分别降低13.23%和63.49%,说明经过编码器降噪以及特征融合处理后能够有效提升煤自燃温度预测的准确度。

In order to effectively prevent coal spontaneous combustion disasters in goaf of coal mines and improve the accuracy of coal spontaneous combustion disaster prediction models,a deep prediction model of coal spontaneous combustion temperature based on multi-feature fusion is proposed.Firstly,each feature of coal spontaneous combustion data is denoised through an autoencoder network to enhance the robustness of the data.Secondly,the denoised data is transformed into a two-dimensional feature matrix in chronological order.A sliding window is used to slice the feature matrix,and a convolutional neural network in deep learning is used to extract effective features from the feature matrix,followed by feature data fusion.In the process of denoising,feature extraction and fusion,differential evolution algorithm is used to optimize the parameters of the denoising autoencoder and convolutional neural network.Finally,the fused data is inputed into the gated recurrent unit neural network for predicting coal spontaneous combustion temperature.The experimental results show that the mean absolute error of the data after denoising and feature fusion is reduced by 6.55%and 69.26%respectively compared to directly using the original data,and the root mean square error is reduced by 13.23%and 63.49%respectively.This indicates that the accuracy of coal spontaneous combustion temperature prediction can be effectively improved after encoder denoising and feature fusion processing.