An ala-tcn-timexer-based coal spontaneous combustion temperature prediction method

Abstract

The application discloses a coal spontaneous combustion temperature prediction method based on ALA-TCN-TimeXer, and belongs to the technical field of coal mine fire prediction. An integrated prediction framework is constructed by fusing an artificial traveling wave optimization algorithm ALA, a time convolution network TCN and a TimeXer model. The method first preprocesses coal spontaneous combustion gas concentration and temperature data and constructs samples; then, key hyperparameters such as a self-adaptive optimization learning rate and model dimension of the ALA are learned; local time series and multi-scale dynamic features are extracted by using the TCN; global dependence and coupling relationship modeling of a temperature target sequence and gas exogenous variables is realized by using the TimeXer; and finally, the predicted temperature is output through a full connection layer. The application can effectively capture temperature mutation and long-term trend, improve prediction accuracy and stability, and provide reliable technical support for early warning of coal spontaneous combustion and mine safety monitoring.

Claims

1. A coal spontaneous combustion temperature prediction method based on ALA-TCN-TimeXer, characterized in that, Includes the following steps: Step S1: Obtain gas concentration data and coal temperature sequence data from the coal spontaneous combustion programmed heating experiment to construct an original dataset; normalize the gas characteristic data and coal temperature data in the original dataset, and construct a time feature matrix based on the time information in the original data; Step S2: Divide the normalized data into training and testing sets, and construct a supervised learning sample sequence using a sliding window method; Step S3: Construct a coal spontaneous combustion temperature prediction model based on TCN-TimeXer. The coal spontaneous combustion temperature prediction model includes an input layer, a TCN layer, a TimeXer layer, and a fully connected layer. The input layer is used to receive gas feature sequences and time feature sequences. The TCN layer is used to extract local time-series features. The TimeXer layer is used to model the global time-series dependency and coupling relationship between the coal temperature target sequence and the gas exogenous variable sequence. The fully connected layer is used to output the predicted value of coal spontaneous combustion temperature. Step S4: The Artificial Lemming Algorithm (ALA) is used to globally adaptively optimize the key hyperparameters of the coal spontaneous combustion temperature prediction model to obtain the optimal hyperparameter combination; the key hyperparameters include the learning rate. Model Dimension and feedforward layer dimension ; wherein the model dimension The feature dimension of the TimeXer layer, the feedforward layer dimension The number of hidden units in the TimeXer feedforward network; Step S5: Train the coal spontaneous combustion temperature prediction model based on the optimal hyperparameter combination, apply the trained model to the test set, and output the coal spontaneous combustion temperature prediction result.

2. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, In step S1, the original gas characteristic data and coal temperature data are processed using Min-Max normalization; the time feature matrix is ​​used to encode date information and periodic time information.

3. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, In step S2, the normalized data is divided into a training set and a test set in an 8:2 ratio; a sliding window method is used to construct model input and output samples, using historical coal temperature sequences, gas feature sequences and time feature sequences at multiple consecutive times as model input, and the coal spontaneous combustion temperature at the time corresponding to the preset prediction step size as the prediction target; the constructed sample sequences are packaged into batch data.

4. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, In step S3, the TCN layer uses an extended causal convolutional structure to extract short-term fluctuation features and multi-scale local dependency features from the input sequence; the TimeXer layer uses a self-attention mechanism and a cross-attention mechanism to jointly model the local temporal features and time features output by the TCN layer, and outputs a global feature representation.

5. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, In step S3, the fully connected layer performs a linear mapping on the global feature representation output by the TimeXer layer to obtain the predicted value of coal spontaneous combustion temperature; the loss function of the coal spontaneous combustion temperature prediction model adopts the mean squared error (MSE), and its formula is as follows: ； in, This represents the model loss value. Indicates the first The predicted value for each sample, Indicates the first The true value of each sample This represents the total number of samples.

6. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, The learning rate The initialization range is 0.001 to 0.01, and the model dimension... The initialization range is 32 to 128, and the feedforward layer dimension The initialization range is 32 to 128.

7. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, In step S4, the Artificial Lemming Algorithm (ALA) constructs multiple surrogate individuals to represent different combinations of hyperparameters, and uses the model prediction error as the fitness value; in the... The energy coefficient is calculated in the next iteration. The energy coefficient The calculation formula is as follows: ； in, Indicates the initial energy parameters. Indicates the current iteration number. This indicates the maximum number of iterations.

8. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, In step S5, the optimized learning rate is... Model Dimension and feedforward layer dimension Substitute the values ​​into the TCN-TimeXer model for final training; After training, the predicted coal spontaneous combustion temperature is output on the test set, and the root mean square error (RMSE) is used to evaluate the model's prediction performance.

9. The method for predicting coal spontaneous combustion temperature based on ALA-TCN-TimeXer according to claim 1, characterized in that, The trained coal spontaneous combustion temperature prediction model is deployed in a mine safety monitoring platform or edge computing device. The mine safety monitoring platform or edge computing device receives gas concentration data collected by gas monitoring devices in goaf areas, sealed fire zones, return airways of fully mechanized mining faces, coal bunkers, or belt conveyor roadways. Based on gas concentration data and time characteristic data from multiple consecutive sampling times, it continuously outputs predicted coal spontaneous combustion temperatures for future preset prediction step sizes. When the predicted coal spontaneous combustion temperature reaches a preset temperature threshold, or when the rate of increase of the predicted coal spontaneous combustion temperature reaches a preset rate of increase threshold within multiple consecutive prediction periods, a coal spontaneous combustion early warning information is output.

Images