Abstract:

In order to predict coal gas permeability more accurately and ensure coal mine safety production, a prediction model of coal gas permeability based on ISSA-optimized BPNN was constructed. Firstly, the sparrow search algorithm (SSA) was improved by introducing Sine chaotic mapping and Gaussian mutation to enhance its global search capability and local optimization accuracy, thereby optimizing the weight and threshold configuration of BPNN. Secondly, the data on the factors affecting gas permeability were processed using Pearson correlation coefficient matrix and kernel principal component analysis (KPCA) to improve the computational efficiency and accuracy of the model. Three principal components with a cumulative variance of 88.59% were extracted as model inputs, and permeability was used as the output for the experiment. Finally, the model was applied to a coal mine in Shanxi for case verification. The experimental results show that ISSA-BPNN outperforms PSO-BPNN, PSO-SVM, PSO-LSSVM, and SSA-BPNN models in four indicators: mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square error (RMSE), root mean square error (RMSE), and coefficient of determination (R²). Compared with other models, ISSA-BPNN has reduced MAE by 0.032 7, 0.022, 0.017 9, and 0.018 2 in the test samples, respectively. MAPE decreases by 5.15%, 3.14%, 2.76%, and 2.36% respectively. RMSE decreases by 0.031 6, 0.027 9, 0.018 8, and 0.022 2 respectively. R² increases by 0.077 5, 0.065 8, 0.040 1, and 0.049 3, respectively. Finally, the case verification shows that its reliability and stability are high in practical applications.

Key words: improved sparrow search algorithm (ISSA), back propagation neural network (BPNN), coal gas, permeability, prediction model