A coal mine gas concentration prediction method and system based on a dynamic factor graph network

Abstract

This invention discloses a method and system for predicting coal mine gas concentration based on a dynamic factor graph network, belonging to the field of coal mine safety monitoring technology. Addressing the problem that existing technologies cannot explicitly and dynamically model the nonlinear coupling relationships of multiple factors, leading to low prediction accuracy, poor robustness, and insufficient interpretability of gas concentration, this invention constructs a coal mine gas concentration prediction network: the front end explicitly captures the dynamic coupling relationships between multiple factors and outputs high-order feature sequences by embedding the original data of each factor in the dynamic factor graph network, gated dynamic message passing, and a GRU state update mechanism; the back end learns the temporal dependencies through GRU and finally outputs the predicted value. The overall system can adaptively adapt to the evolution of factor coupling relationships during mining, significantly improving the model's adaptability to complex systems, and enhancing the model's learning efficiency and prediction accuracy; the entire process is automated with a fast response speed, meeting the needs of real-time safety monitoring in coal mines and possessing significant engineering application value.

Description

Technical Field

[0001] This invention belongs to the field of coal mine safety monitoring technology, specifically relating to a method for predicting coal mine gas concentration based on dynamic factor graph networks. Background Technology

[0002] Coal mine gas disasters are one of the major hazards threatening safe production in coal mines. Accurate prediction of underground gas concentration trends is crucial for early warning and risk avoidance. Gas concentration is influenced by a variety of factors, including but not limited to gas emission rate, ventilation volume, ambient temperature, ambient humidity, and mining activity intensity. These factors exhibit complex nonlinear coupling relationships, which dynamically evolve as mining progresses. Therefore, gas concentration prediction is a typical time-series prediction problem for complex systems.

[0003] Existing gas concentration prediction technologies are mainly divided into two categories: traditional statistical methods and deep learning methods. Traditional statistical methods can only capture the linear time-series patterns of data and are difficult to handle the nonlinear coupling relationships between multiple factors, resulting in low prediction accuracy and poor adaptability. Deep learning methods (such as models based on Long Short-Term Memory (LSTM) networks and Time-Gated Recurrent Networks (GRU) networks) utilize the nonlinear fitting ability of neural networks to learn time-series dependencies, showing a significant improvement over traditional methods. However, existing deep learning methods typically treat all influencing factors as flat feature vectors and directly input them into the time-series network, forcing the time-series network to "implicitly" learn the interaction relationships between multiple factors. This approach has the following drawbacks: First, it increases the learning burden on the time-series network; second, the learned interaction relationships are static and cannot adapt to the dynamic evolution of factor coupling relationships during the mining process; third, the model has poor interpretability and cannot clearly define the specific impact paths of each factor on the prediction results. Summary of the Invention

[0004] To address the aforementioned problems, the purpose of this invention is to provide a method and system for predicting coal mine gas concentration based on dynamic factor graph networks, thereby solving the problems of low prediction accuracy, poor adaptability, and weak interpretability in existing technologies.

[0005] To achieve the above objectives, the technical solution adopted by the present invention includes:

[0006] A method for constructing a coal mine gas concentration prediction model based on a dynamic factor graph network includes the following steps: S1. Obtain multi-factor time-series data affecting gas concentration in underground coal mines, and preprocess the multi-factor time-series data to obtain a training sample set; the multi-factor time-series data includes multiple time steps, each time step contains raw data of multiple factors, including gas concentration, ventilation volume, ambient temperature, ambient humidity and mining machine power; obtain the corresponding future gas concentration real value as a label. S2, Construct a coal mine gas concentration prediction network, which includes a dynamic factor graph network, a time-gated cyclic network, and a fully connected output layer; The dynamic factor graph network includes a factor node embedding layer, a dynamic message passing layer, a node state update gated loop unit layer and an output fusion layer connected in series. The output of the factor node embedding layer is also connected to the input of the node state update gated loop unit layer. The factor node embedding layer maps the original data of each factor at each time step to a node state vector in a high-dimensional feature space. The dynamic message passing layer calculates a gating message representing the interaction between any two different factors using their node state vectors at the current time step. It then selects any one of the multiple factors as the target factor, sends the gating message to all other factors (excluding the target factor), and sums the results to obtain the aggregated message vector corresponding to the target factor. The node state update gating loop unit layer takes the aggregated message vector corresponding to the target factor as input, combines it with the current node state vector of the target factor, and updates the node state vector of the target factor at the next time step. It iterates through all factors to obtain the node state vector of each factor as the target factor at the next time step. The output fusion layer fuses the node state vectors of all factors at the next time step to obtain a high-order feature vector. The time-gated recurrent network receives the high-order feature sequence arranged by time step output by the dynamic factor graph network, iteratively processes the high-order feature sequence by time step, and obtains and outputs the time-series feature vector after processing the last time step. The fully connected output layer is used to receive the temporal feature vector output by the temporal gated recurrent network and generate the gas concentration prediction value for the next time step. S3 takes the training sample set obtained in S1 as input, the label as output, the mean squared error (MSE) as the loss function and the Adam optimization algorithm as the training method to train the coal mine gas concentration prediction network constructed in S2 until convergence, and obtains the coal mine gas concentration prediction model.

[0007] Preferably, the dynamic message passing layer specifically performs the following steps: S21, For factors i and j, the node state vector of factor i at the current time step and the node state vector of factor j at the current time step are concatenated along the feature dimension and input into the first multilayer perceptron to calculate the initial interaction message of factor i to factor j. S22, the node state vector of factor i at the current time step and the node state vector of factor j at the current time step are concatenated along the feature dimension, input into the second multilayer perceptron and activated by the Sigmoid function to obtain the gating coefficients, and the gating coefficients are multiplied element-wise with the initial interaction message obtained in S21 to obtain the gating message; S23, specify factor j as the target factor, sum the gating messages passed to it by all other factors i (i≠j) to obtain the aggregated message vector pointing to factor j.

[0008] Preferably, the node state update gated loop unit layer adopts GRUCell, which takes the aggregated message vector corresponding to the target factor as input, takes the current node state vector of the target factor as the hidden state, and generates the node state vector of the target factor in the next time step through the update gate and reset gate mechanism of GRUCell.

[0009] Preferably, the output fusion layer obtains the higher-order feature vector by fusing the node state vectors of all factors at the next time step and performing average pooling.

[0010] Preferably, the timing-gated cyclic network is a GRU network.

[0011] Preferably, the preprocessing includes alignment, completion, denoising, and normalization.

[0012] A method for predicting coal mine gas concentration based on dynamic factor graph networks includes the following steps: Step 1: Collect multi-factor time-series data on gas concentration in underground coal mines and perform preprocessing; Step 2: Input the preprocessed multi-factor real-time monitoring data into the coal mine gas concentration prediction model for prediction, and obtain the predicted gas concentration value for the next time step; the coal mine gas concentration prediction model is the coal mine gas concentration prediction model constructed by the construction method disclosed in this application. Step 3: Perform reverse normalization on the predicted gas concentration value for the next time step to obtain the predicted gas concentration result for the next time step.

[0013] A coal mine gas concentration prediction system based on dynamic factor graph network, comprising: The data acquisition module is used to collect multi-factor time-series data on gas concentration in underground coal mines in real time. The data processing module is used to perform alignment, completion, denoising, and normalization processing on the multi-factor time series data; The coal mine gas concentration prediction module is used to input preprocessed multi-factor time series data into the coal mine gas concentration prediction model constructed by the method disclosed in this application, to obtain the gas concentration prediction value for the next time step, and to perform reverse normalization on the gas concentration prediction value for the next time step to obtain the gas concentration prediction result for the next time step. The early warning display module is used to visualize the gas concentration prediction results for the next time step and trigger an early warning when the gas concentration prediction results for the next time step exceed a preset threshold.

[0014] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the coal mine gas concentration prediction method based on dynamic factor graph networks disclosed in this application.

[0015] A computer program product includes a computer program / instructions that, when executed by a processor, implement the coal mine gas concentration prediction method based on dynamic factor graph networks disclosed in this application.

[0016] Compared with the prior art, the advantages of the present invention are: (1) The coal mine gas concentration prediction method and system based on dynamic factor graph network of the present invention innovatively designs dynamic factor graph network, realizes explicit and dynamic modeling of multi-factor nonlinear coupling relationship, breaks through the limitation of the existing technology of "implicit learning interaction relationship", and can adapt to the evolution of factor coupling relationship during mining by dynamically adjusting the message transmission strength between factors through gating mechanism, significantly improving the model's adaptability to complex systems.

[0017] (2) The coal mine gas concentration prediction method and system based on dynamic factor graph network of the present invention decomposes the complex multi-factor time series prediction problem into two sub-problems: "interactive modeling" and "time series learning", which reduces the learning burden of individual modules and improves the learning efficiency and prediction accuracy of the model. Experimental verification shows that the prediction accuracy (MAE, RMSE) of the model of the present invention is 15% higher than that of the existing LSTM and bidirectional GRU models.

[0018] (3) The coal mine gas concentration prediction method and system based on dynamic factor graph network of the present invention can clarify the influence weight of different factors on gas concentration by analyzing the gating coefficients and gating messages between factors, and provide decision-making basis for coal mine safety management. However, most existing deep learning models are "black box" models and cannot provide interpretability support.

[0019] (4) The coal mine gas concentration prediction method and system based on dynamic factor graph network of the present invention realizes full-process automation of data acquisition, processing, prediction and early warning, with fast response speed, which can meet the needs of real-time safety monitoring in coal mines and has important engineering application value. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the following detailed description to explain the invention, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a curve comparing the prediction results of the model of this invention with those of existing models; Figure 2 This is a schematic diagram of the distribution of gating coefficients among factors in a dynamic factor graph network. Detailed Implementation

[0021] The invention is not limited to the specific embodiments described below. All equivalent modifications made based on the technical solutions of this application fall within the protection scope of this invention. Unless otherwise specified, all components and devices in this invention utilize components and devices known in the prior art.

[0022] Example 1 This embodiment discloses a method for constructing a coal mine gas concentration prediction model based on a dynamic factor graph network. The hardware environment is as follows: CPU is Intel Core i7-12700H, GPU is NVIDIA RTX 3060 (6GB), and memory is 32GB. The software environment is as follows: Python 3.9, PyTorch 1.12.1, NumPy 1.23.5, Pandas 1.5.3, and Matplotlib 3.6.2.

[0023] This embodiment specifically includes the following steps: S1. Obtain multi-factor time-series data on the impact of gas concentration in underground coal mines, and preprocess the multi-factor time-series data to obtain a training sample set; the multi-factor time-series data includes multiple time steps, each time step contains the original data of multiple factors, including gas concentration, ventilation volume, ambient temperature, ambient humidity and mining machine power. Obtain the actual gas concentration value at the corresponding future moment as the label.

[0024] This embodiment collects 1000 hours of real-time monitoring data (i.e., multi-factor time series data) from a coal mine at 1-hour intervals, including 5 factors: gas concentration, ventilation volume, ambient temperature, ambient humidity, and mining machine power. The multi-factor time series data is preprocessed as follows: Data alignment: Align the data of the five factors according to the timestamp and delete outlier data with mismatched timestamps; Missing value handling: Use linear interpolation to complete a small number of missing data with a missing rate of <3%; Outlier handling: Use the 3σ principle to identify abrupt values ​​caused by sensor failure and replace them with the mean of two adjacent normal data; Normalization: Use Min-MaxScaler to normalize all data to the [0,1] interval; Set the time step T=24 (predict the gas concentration for the next hour using data from the past 24 hours), and construct an input sequence of shape (976,24,5,1) from the preprocessed data.

[0025] S2, Construct a coal mine gas concentration prediction network, which includes a dynamic factor graph network, a time-gated cyclic network, and a fully connected output layer; The dynamic factor graph network includes a factor node embedding layer, a dynamic message passing layer, a node state update gated loop unit layer and an output fusion layer connected in series. The output of the factor node embedding layer is connected to the input of the node state update gated loop unit layer. The factor node embedding layer is used to map the original data of each factor at each time step to a node state vector in a high-dimensional feature space. Specifically, an independent linear embedding layer is set for each factor to map the 1-dimensional scalar original data to a 64-dimensional high-dimensional feature space. For the i-th factor at time step t... Calculate its node state vector in the high-dimensional feature space: ,in , Let be the learnable weights and biases of the i-th factor.

[0026] The dynamic message passing layer is used to calculate a gating message representing the interaction relationship between any two different factors through their node state vectors at the current time step. It takes any one of the multiple factors as the target factor, and sums the gating messages sent by all other factors to the target factor to obtain the aggregated message vector corresponding to the target factor. The dynamic message passing layer specifically performs the following steps: S21, for factors i and j, the node state vector of factor i at the current time step and the node state vector of factor j at the current time step are concatenated along the feature dimension and input into the first multilayer perceptron. In this embodiment, the first multilayer perceptron is a 2-layer MLP. The initial interaction message of factor i to factor j is calculated. ; S22, the node state vector of factor i at the current time step and the node state vector of factor j at the current time step are concatenated along the feature dimension, input into the second multilayer perceptron and activated by the Sigmoid function. In this embodiment, the second multilayer perceptron is a single linear layer to obtain the gating coefficients. The gating coefficient is multiplied element-wise with the initial interaction message obtained in S21 to obtain the gating message. ； S23, designate factor j as the target factor, and sum the gated messages passed to it by all other factors i (i≠j) to obtain an aggregated message vector pointing to the target factor j. .

[0027] The node state update gated loop unit layer takes the aggregated message vector corresponding to the target factor as input, and combines it with the current node state vector of the target factor to update and generate the node state vector of the factor in the next time step. In this embodiment, the node state update gated loop unit layer adopts GRUCell, which takes the aggregated message vector corresponding to the target factor as input, and updates and generates the node state vector of the factor in the next time step. As input, the current node state vector of the target factor As a hidden state, the node state vector of the target factor in the next time step is generated through the update gate and reset gate mechanism of GRUCell. Iterate through all factors to obtain the node state vector at the next time step for each factor as the target factor.

[0028] The output fusion layer is used to fuse the node state vectors of all factors at the next time step and perform average pooling to obtain the high-order feature vector at time step t. ; Obtain higher-order feature sequences In this embodiment, t=5.

[0029] The time-gated recurrent network employs a two-layer stacked GRU network, receiving the high-order feature sequences arranged by time steps output by the dynamic factor graph network. Iterative processing of high-order feature sequences by time step Output the hidden state at the last time step. The hidden state of the last time step is used as the temporal feature vector and output. Where t is 1, This represents the initial hidden state (set as a vector of 0).

[0030] The fully connected output layer is used to receive the time-series feature vector and generate a predicted gas concentration value for the next hour.

[0031] S3 takes the training sample set obtained in S1 as input, the label as output, the mean squared error (MSE) as the loss function, and the Adam optimization algorithm as the model parameter update. The coal mine gas concentration prediction network constructed in S2 is trained until convergence, and the coal mine gas concentration prediction model is obtained.

[0032] The coal mine gas concentration prediction model (DF-GN-GRU model) of this embodiment was compared with the existing mainstream models (LSTM model and GRU model) using mean absolute error (MAE) and root mean square error (RMSE) as evaluation indicators. The experimental results are shown in Table 1. The MAE and RMSE of the DF-GN-GRU model of this invention are significantly lower than those of the LSTM model and GRU model, and the prediction accuracy is significantly improved, which verifies the effectiveness of the invention.

[0033] Table 1 Comparison of DF-GN-GRU, LSTM, and GRU

[0034] The gating coefficients at all time steps in the dynamic factor graph network are extracted, and the average gating coefficient of each factor is calculated. The results are as follows: Figure 2 The results show the following parameters: gas concentration (0.68), ventilation volume (0.72), ambient humidity (0.35), ambient temperature (0.28), and mining machine power (0.45). These results are entirely consistent with the conclusion in coal mine safety theory that "ventilation volume and gas emission are the core factors affecting gas concentration," verifying the model's interpretability and providing a basis for decision-making in coal mine safety management.

[0035] Example 2 This embodiment also discloses a method for predicting coal mine gas concentration based on dynamic factor graph networks, including the following steps: Step 1: Collect multi-factor time-series data on gas concentration in underground coal mines and perform preprocessing; The underground sensor network collects data on gas concentration, ventilation volume, ambient temperature, ambient humidity, and mining machine power every hour. The real-time data is processed according to the same preprocessing procedure as the training phase (alignment, completion, noise reduction, and normalization) to construct a single-sample input sequence of shape (1,24,5,1) (containing data from the current moment and the previous 23 hours). Step 2: Input the preprocessed multi-factor real-time monitoring data into the coal mine gas concentration prediction model for prediction to obtain the predicted gas concentration value for the next hour; the coal mine gas concentration prediction model is the coal mine gas concentration prediction model constructed by the construction method in Example 1. Step 3: Perform reverse normalization on the predicted gas concentration for the next hour to obtain the predicted gas concentration for the next hour.

[0036] The predicted methane concentration for the next hour is compared with the preset methane concentration safety threshold. If the predicted value exceeds the threshold, an audible and visual warning is immediately triggered, and the warning information is pushed to the coal mine safety monitoring center.

[0037] The comparison curves of the predicted gas concentration for the next hour are as follows: Figure 1 As shown, the predicted curve of the model of this invention has the highest degree of fit with the actual gas concentration curve and can accurately capture the sudden change trend of gas concentration, while the LSTM model and GRU model have obvious lag and bias.

[0038] Example 3 This embodiment discloses a coal mine gas concentration prediction system based on a dynamic factor graph network, including: The data acquisition module is used to collect multi-factor time-series data on gas concentration in underground coal mines in real time. The data acquisition module in this embodiment consists of a gas sensor, a wind speed sensor, a temperature sensor, a humidity sensor, and a mining machine power sensor deployed in the underground fully mechanized mining face. All sensors comply with the explosion-proof standards for underground coal mines. Each sensor collects real-time data of five factors once per hour and transmits it to the data processing module. The data processing module receives multi-factor time series data and performs alignment, completion, denoising, and normalization processing on it. The coal mine gas concentration prediction module loads the coal mine gas concentration prediction model constructed by the construction method of this application, and inputs the preprocessed multi-factor time series data into the coal mine gas concentration prediction model to obtain the gas concentration prediction value for the next hour. Then, it performs reverse normalization on the gas concentration prediction value for the next hour to obtain the gas concentration prediction result for the next hour. The early warning display module is used to visualize the gas concentration prediction results for the next hour and trigger an early warning when the gas concentration prediction results for the next hour exceed a preset threshold.

[0039] The early warning display module in this embodiment includes a large screen in the coal mine safety monitoring center and a mobile APP for managers. It displays historical data and prediction curves of gas concentration in real time in the form of a line graph, and marks the safety threshold line. When the predicted gas concentration for the next hour exceeds the threshold, the monitoring center triggers an audible and visual warning and pushes a warning message to the manager's mobile APP, which includes the warning location, the predicted gas concentration result, and suggested handling measures.

[0040] The system was tested in a coal mine for three months. During this period, it accurately warned of 12 abnormal gas concentration increases, which bought the coal mine sufficient time to take emergency measures such as ventilation adjustment and personnel evacuation. No gas safety accidents occurred, which verified the practicality and reliability of the system.

[0041] This embodiment also discloses a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the coal mine gas concentration prediction method based on dynamic factor graph network disclosed in this application.

[0042] This embodiment also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the coal mine gas concentration prediction method based on dynamic factor graph network disclosed in this application.

[0043] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various simple modifications can be made to the technical solutions of the present disclosure, and these simple modifications all fall within the protection scope of the present disclosure.

[0044] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0045] Furthermore, the various implementation methods disclosed in this solution can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content invented by this disclosure.

Claims

1. A coal mine gas concentration prediction model construction method based on a dynamic factor graph network, characterized in that, Includes the following steps: S1. Obtain multi-factor time-series data affecting gas concentration in underground coal mines, and preprocess the multi-factor time-series data to obtain a training sample set; the multi-factor time-series data includes multiple time steps, each time step contains raw data of multiple factors, including gas concentration, ventilation volume, ambient temperature, ambient humidity and mining machine power. Obtain the actual gas concentration value at the corresponding future moment as the label; S2, Construct a coal mine gas concentration prediction network, which includes a dynamic factor graph network, a time-gated cyclic network, and a fully connected output layer; The dynamic factor graph network includes a factor node embedding layer, a dynamic message passing layer, a node state update gated loop unit layer and an output fusion layer connected in series. The output of the factor node embedding layer is also connected to the input of the node state update gated loop unit layer. The factor node embedding layer maps the original data of each factor at each time step to a node state vector in a high-dimensional feature space. The dynamic message passing layer calculates a gating message representing the interaction between any two different factors using their node state vectors at the current time step. It then selects any one of the multiple factors as the target factor, sends the gating message to all other factors (excluding the target factor), and sums the results to obtain the aggregated message vector corresponding to the target factor. The node state update gating loop unit layer takes the aggregated message vector corresponding to the target factor as input, combines it with the current node state vector of the target factor, and updates the node state vector of the target factor at the next time step. It iterates through all factors to obtain the node state vector of each factor as the target factor at the next time step. The output fusion layer fuses the node state vectors of all factors at the next time step to obtain a high-order feature vector. The time-gated recurrent network receives the high-order feature sequence arranged by time step output by the dynamic factor graph network, iteratively processes the high-order feature sequence by time step, and obtains and outputs the time-series feature vector after processing the last time step. The fully connected output layer is used to receive the temporal feature vector output by the temporal gated recurrent network and generate the gas concentration prediction value for the next time step. S3 takes the training sample set obtained in S1 as input, the label as output, the mean squared error (MSE) as the loss function and the Adam optimization algorithm as the training method to train the coal mine gas concentration prediction network constructed in S2 until convergence, and obtains the coal mine gas concentration prediction model.

2. The coal mine gas concentration prediction model construction method based on a dynamic factor graph network according to claim 1, characterized in that, The dynamic message passing layer specifically performs the following steps: S21, For factors i and j, the node state vector of factor i at the current time step and the node state vector of factor j at the current time step are concatenated along the feature dimension and input into the first multilayer perceptron to calculate the initial interaction message of factor i to factor j. S22, the node state vector of factor i at the current time step and the node state vector of factor j at the current time step are concatenated along the feature dimension, input into the second multilayer perceptron and activated by the Sigmoid function to obtain the gating coefficients, and the gating coefficients are multiplied element-wise with the initial interaction message obtained in S21 to obtain the gating message; S23, specify factor j as the target factor, sum the gating messages passed to it by all other factors i (i≠j) to obtain the aggregated message vector pointing to factor j.

3. The coal mine gas concentration prediction model construction method based on a dynamic factor graph network according to claim 1, characterized in that, The node state update gated loop unit layer adopts GRUCell, takes the aggregated message vector corresponding to the target factor as input, takes the current node state vector of the target factor as the hidden state, and generates the node state vector of the target factor in the next time step through the update gate and reset gate mechanism of GRUCell.

4. The coal mine gas concentration prediction model construction method based on a dynamic factor graph network according to claim 1, characterized in that, The output fusion layer obtains the higher-order feature vector by fusing the node state vectors of all factors at the next time step and performing average pooling.

5. The coal mine gas concentration prediction model construction method based on the dynamic factor graph network according to any one of claims 1-4, characterized in that, The time-gated cyclic network is a GRU network.

6. The coal mine gas concentration prediction model construction method based on a dynamic factor graph network according to claim 5, characterized in that, The preprocessing includes alignment, completion, denoising, and normalization.

7. A coal mine gas concentration prediction method based on a dynamic factor graph network, characterized in that, Includes the following steps: Step 1: Collect multi-factor time-series data on gas concentration in underground coal mines and perform preprocessing; Step 2: Input the preprocessed multi-factor real-time monitoring data into the coal mine gas concentration prediction model for prediction, and obtain the gas concentration prediction value for the next time step. The coal mine gas concentration prediction model is the coal mine gas concentration prediction model constructed by any of the construction methods in claims 1 to 6; Step 3: Perform reverse normalization on the predicted gas concentration value for the next time step to obtain the predicted gas concentration result for the next time step.

8. A coal mine gas concentration prediction system based on a dynamic factor graph network, characterized in that, include: The data acquisition module is used to collect multi-factor time-series data on gas concentration in underground coal mines in real time. The data processing module is used to perform alignment, completion, denoising, and normalization processing on the multi-factor time series data; The coal mine gas concentration prediction module is used to input preprocessed multi-factor time series data into the coal mine gas concentration prediction model constructed by any of the construction methods described in claims 1 to 6, obtain the gas concentration prediction value for the next time step, and perform reverse normalization on the gas concentration prediction value for the next time step to obtain the gas concentration prediction result for the next time step. The early warning display module is used to visualize the gas concentration prediction results for the next time step and trigger an early warning when the gas concentration prediction results for the next time step exceed a preset threshold.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the coal mine gas concentration prediction method based on a dynamic factor graph network as described in claim 7.

10. A computer program product, characterised in that, It includes a computer program / instruction, which, when executed by a processor, implements the coal mine gas concentration prediction method based on a dynamic factor graph network as described in claim 7.

Images