A novel method for predicting tunnel blasting dust yield based on numerical inversion and WOA-DELM (Whale Optimization Algorithm-Deep Extreme Learning Machine) neural network.

Drilling and blasting during the construction of tunnels generate large amounts of dust. Accurately predicting dust yield helps to formulate effective dust control measures that can be employed to reduce worker exposures. However, current dust yield prediction methods mainly rely on empirical formulas for open-pit blasting, which do not fully account for the combined effects of ventilation, rock type, and environmental conditions. This paper uses a railway tunnel as a case study and analyzes the dust dispersion patterns based on on-site measurements. Through numerical inversion, a dynamic correlation model between dust concentration at monitoring points and total dust generation was established, and sensitivity analysis of influencing factors was conducted using the Whale Optimization Algorithm-Deep Extreme Learning Machine (WOA-DELM) neural network. A new calculation formula for tunnel blasting dust yield is proposed. The results showed a bimodal dispersion pattern: the initial dispersion was driven by the blasting shock wave, while the secondary dispersion was caused by mechanical ventilation. The surrounding rock type was the most significant factor affecting dust yield, with a variation of up to 8.04 kg. Other factors influencing yield included tunnel temperature (0.97 kg/°C), humidity (0.74 kg/%), tunnel water-content level (0.51 kg per level on a field-based 1-10 scale) and distance from the monitoring point to the tunnel face and ventilation duct outlet (0.17 kg/m and 0.02 kg/m, respectively). By using the WOA-DELM neural network to optimize the traditional open-pit blasting dust calculation formula, an average prediction error of 4.6% was achieved. The research results provide a scientific basis and parameter guidance for developing data necessary to implement engineering controls for tunnel dust control.