Intelligent Time Series Prediction of Harmful Gas Concentrations during Underground Cavern Construction: Integrated Construction and Geological Factors

Underground cavern operations present safety hazards due to high levels of harmful gases, endangering worker health. However, previous studies on predicting harmful gas concentrations in underground caverns have rarely considered the influence of construction activities, geological factors, and temporal variations. Therefore, this paper proposes a time series prediction model that integrates construction activities and geological factors. Category boosting is used to extract information related to construction activities and geological factors that is closely associated with harmful gas concentrations. A convolutional neural network–long short-term memory module, enhanced with a self-attention mechanism, is utilized to capture temporal information. The beluga whale optimization algorithm is used to optimizes hyperparameters and enhance learning. Results have shown that the model can achieve high-precision prediction of harmful gas concentrations. In addition, the added interpretability enhancement algorithm can effectively explore the impact mechanism of construction and geological factors as well as temporal characteristics on the concentration of harmful gases.