ISBOA-Transformer Coal Mine Roadway Gas Concentration Prediction with Biological Intelligence Improvement

Abstract

In the domain of coal mine safety, precisely forecasting the gas concentration in roadways holds utmost significance. It serves as a crucial safeguard for miners' lives and well - being and also plays a pivotal role in augmenting the economic performance of mining enterprises. To achieve this accurate prediction of coal mine roadway gas concentration, this research centers around the Bultai coal mine for in - depth investigation. A state - of - the - art concentration monitoring sensor is utilized to perform real - time gas concentration measurements within the coal mine roadways. Moreover, a novel ISBOA - Transformer prediction model for coal mine roadway gas concentration, enhanced through bio - intelligence techniques, is introduced. Initially, improvements are made to the traditional Secretary Bird Optimization Algorithm (SBOA). Latin hypercubic sampling is adopted to ensure a more uniform population initialization, and a fixed - point recombination and mutation strategy is implemented. This strategy aims to boost the evolutionary probability of species and diversify the population. Finally, the optimized algorithm is utilized to precisely tune the hyperparameters associated with the number of attention heads in the Transformer model. Through this process, a highly effective gas concentration prediction model is developed. When validating the model with a training dataset and a testing dataset, the Mean Absolute Error (MAE) of the proposed model is determined to be 0.00062531 for the training dataset and 0.0005832 for the testing dataset. The Root Mean Squared Error (RMSE) registers at 0.0008848 for the training dataset and 0.00082342 for the testing dataset. The Goodness of Fit (R2) values are 0.9712 for the training dataset and 0.9800 for the testing dataset. Clearly, these evaluation metrics vividly demonstrate that the ISBOA - Transformer model performs better than other models in the comparison. Subsequently, the model is implemented on extra gas concentration datasets obtained from the Bultai coal mine.