CN-121998327-A - Double-flow gas concentration prediction method and device based on time sequence decomposition

CN121998327ACN 121998327 ACN121998327 ACN 121998327ACN-121998327-A

Abstract

The invention discloses a method and a device for predicting double-flow gas concentration based on time sequence decomposition, which are used for acquiring a multi-channel gas concentration data historical sequence in a first time period, decoupling the multi-channel gas concentration data historical sequence into a trend subsequence and a period subsequence based on channels, obtaining a trend prediction sequence and a period prediction sequence of gas concentration, and fusing the trend prediction sequence and the period prediction sequence to obtain a predicted gas concentration sequence in a time period to be predicted.

Inventors

TIAN FENG
LIU YICHEN
LIU XIAOPEI
ZHANG XIAOHONG

Assignees

西安科技大学

Dates

Publication Date: 20260508
Application Date: 20260119

Claims (9)

1. The method for predicting the concentration of the double-flow gas based on time sequence decomposition is characterized by comprising the following steps of: The method comprises the steps of acquiring a multi-channel gas concentration data historical sequence of a first time period, wherein the first time period is a continuous time period of an immediately adjacent preamble of a time period to be predicted, and the multi-channel gas concentration data historical sequence comprises at least one of fully-mechanized coal face gas concentration and upper corner gas concentration, return airway gas concentration, temperature, air inlet gas concentration, wind speed, oxygen concentration and coal cutter cutting speed; the multi-channel gas concentration data history sequence is decoupled into a trend subsequence and a period subsequence based on channels, the trend subsequence is predicted based on a trend prediction sub-module to obtain a trend prediction sequence of the gas concentration, the period subsequence is predicted based on a period prediction sub-module to obtain a period prediction sequence of the gas concentration, and the trend prediction sequence and the period prediction sequence are fused to obtain a predicted gas concentration sequence of a time period to be predicted.
2. The method for predicting the concentration of double-flow gas based on time sequence decomposition according to claim 1, wherein the trend prediction sub-module performs reversible normalization and inverse normalization on the trend sub-sequence corresponding to each channel, wherein a linear sub-network is arranged between the normalization and the inverse normalization, and the linear sub-network is provided with a plurality of fully connected layers connected in series.
3. The dual-flow gas concentration prediction method based on time series decomposition according to claim 1, wherein said linear subnetwork comprises a first fully-connected layer, a second fully-connected layer and a third fully-connected layer; The first full-connection layer is used for mapping the normalized trend subsequence to a high-dimensional feature space to obtain t ∈ , Representing the real space with the channel number of C and the sequence length of D, t Representing a high-dimensional feature at time t; The second full connection layer is used for connecting t Mapping to prediction space to obtain t ∈ , Representing the real space with the channel number of C and the sequence length of T, t Channel prediction characteristics at the time t are represented; The third full connection layer is used for connecting t Mapping to single channel prediction space to obtain trend ∈ , Representing the real space of sequence length T in a predetermined channel, trend A trend prediction sequence at time t is shown.
4. The method for predicting double-flow gas concentration based on time sequence decomposition according to claim 2 or 3, wherein predicting the periodic subsequence based on the periodic prediction sub-module to obtain a periodic prediction sequence of gas concentration comprises: Reconstructing the normalized periodic subsequence into a patch sequence; sequentially carrying out depth separable convolution and dynamic convolution on the patch sequence to obtain an integrated feature map; And carrying out residual connection on the integrated features and the patch sequence, mapping the residual connection to a prediction space, and fusing feature maps of the prediction space of each channel to obtain a periodic prediction sequence.
5. The dual-stream gas concentration prediction method based on time series decomposition according to claim 4, wherein said dynamic convolution is implemented by a first dynamic convolution layer and a second dynamic convolution layer connected in series; The first dynamic convolution layer is used for capturing nonlinear time sequence variation of a single channel and outputting a first integrated characteristic diagram , A real space with a channel of C, a sequence length of D and a patch sequence number of N is represented; The second dynamic convolution layer is used for realizing the integrated expression of the multi-channel information and outputting a second integrated characteristic diagram 。
6. The dual-flow gas concentration prediction method based on time sequence decomposition according to claim 5, wherein the residual connected feature map is mapped to a prediction space through a fourth full connection layer; and fusing the feature maps of the prediction space of each channel through a fifth full-connection layer to obtain a periodic prediction sequence.
7. The method for predicting double-flow gas concentration based on time series decomposition according to claim 6, wherein the method for fusing the trend prediction sequence and the period prediction sequence is as follows: , Wherein, the The predicted gas concentration sequence is indicated, Representation of trend The trend prediction sequence after the inverse normalization, Representing the denormalized periodic predicted sequence, α and β represent the adjustment factors for adaptively adjusting the contribution ratio.
8. The method for predicting dual-flow gas concentration based on time series decomposition of claim 2, further comprising, prior to decoupling the multi-channel gas concentration data history sequence: detecting and eliminating abnormal data in the multi-channel gas concentration data history sequence; Interpolation is carried out on missing data in the multi-channel gas concentration data history sequence; eliminating high-frequency disturbance components in the multi-channel gas concentration data history sequence; And eliminating the gas concentration data history sequence of the weak related channel in the multi-channel gas concentration data history sequence.
9. A dual-flow gas concentration prediction method based on time series decomposition, comprising a memory, a processor and a computer program stored in the memory and running on the processor, characterized in that the processor implements the method according to any one of claims 1-8 when executing the computer program.

Description

Double-flow gas concentration prediction method and device based on time sequence decomposition Technical Field The invention belongs to the technical field of coal mine safety monitoring, and particularly relates to a method and a device for predicting double-flow gas concentration based on time sequence decomposition. Background The gas concentration is a core index of coal mine safety, and the prediction accuracy of the gas concentration directly influences the effectiveness of disaster prevention and control. Traditional monitoring methods rely on single-point threshold value overrun alarm, and lack prospective characterization of risk evolution process. Statistical models (e.g., ARIMA, SVM) while able to capture timing laws to some extent, have difficulty in effectively characterizing complex nonlinear dynamic features. The deep learning method (such as LSTM, CNN, transformer) is gradually applied to the field of gas concentration prediction, so that the self-adaptive modeling capacity of the model is remarkably improved. When dealing with sequence non-stationarity, multi-scale coupling and multi-variable complex interaction relations existing in an underground environment, the method still has the problems of limited prediction precision, insufficient long-range dependent modeling capability, high calculation complexity and the like. Therefore, when the coal mine underground environment faces strong noise disturbance and high fluctuation sequences, the prediction precision is obviously reduced, and the safety early warning requirement of high reliability is difficult to meet. Disclosure of Invention The invention aims to provide a method and a device for predicting double-flow gas concentration based on time sequence decomposition, so as to improve prediction accuracy in the face of strong noise disturbance and high fluctuation sequences. The invention adopts the following technical scheme that the double-flow gas concentration prediction method based on time sequence decomposition comprises the following steps: The method comprises the steps of acquiring a multi-channel gas concentration data historical sequence of a first time period, wherein the first time period is a continuous time period of an immediate preamble of a time period to be predicted, and the multi-channel gas concentration data historical sequence comprises at least one of fully-mechanized coal face gas concentration and upper corner gas concentration, return airway gas concentration, temperature, air inlet gas concentration, wind speed, oxygen concentration and coal cutter cutting speed; The method comprises the steps of decoupling a multi-channel gas concentration data history sequence into a trend subsequence and a period subsequence based on channels, predicting the trend subsequence based on a trend prediction sub-module to obtain a trend prediction sequence of gas concentration, predicting the period subsequence based on a period prediction sub-module to obtain a period prediction sequence of gas concentration, and fusing the trend prediction sequence and the period prediction sequence to obtain a predicted gas concentration sequence of a time period to be predicted. The multi-channel gas concentration data historical sequence is decoupled into the trend subsequence and the periodic subsequence, so that the multi-scale feature capturing capability is improved, the modeling and fusion capability of time-varying interaction features among multiple channels can be enhanced, the adaptability in a complex environment is improved, and the prediction stability and the robustness are improved. Drawings FIG. 1 is a schematic diagram of a dual-flow gas concentration prediction method based on time-series decomposition according to an embodiment of the present invention; FIG. 2 is a schematic diagram of a dual stream prediction module according to an embodiment of the present invention; FIG. 3 is a schematic diagram showing the comparison of effects during single-time prediction in the embodiment of the present invention; FIG. 4 is a graph showing the comparison of the effects of model prediction with a sliding step of 4 in the embodiment of the present invention; FIG. 5 is a graph showing the comparison of the effects of model prediction with a sliding step size of 8 in the embodiment of the present invention; FIG. 6 is a graph showing the comparison of the effects of model predictive sliding step size of 16 in an embodiment of the invention. Detailed Description The invention will be described in detail below with reference to the drawings and the detailed description. The underground coal mine gas concentration time sequence is very suitable for analysis by a time sequence decomposition method, and the core reasons are that the data characteristics of the underground coal mine gas concentration time sequence completely fit the application premise of time sequence decomposition, and the underground coal mine gas concentration time seque