Top

TD3-MPC Model for Mineral Dust Concentration Estimation Based on Hot-Air Flow and Particle Sensors

Authors

Files

PDF

Abstract

Real-time estimation of mineral dust concentration in thermally disturbed mine roadways remains difficult because hot-air ventilation changes particle trajectories, delays emission-measurement coupling, and slows optical sensor recovery. This paper develops a twin delayed deep deterministic policy gradient and a model predictive control model for concentration estimation based on coupled hot-air flow and particle-sensor streams. The method introduces delay-compensated feature reconstruction, a physics-guided state transition layer, twin critic learning and receding-horizon correction. A 42-hour laboratory and pilot-gallery dataset was organised into 151,200 synchronous samples of thermal anemometers, particle counters, humidity probes and gravimetric reference instruments. Compared with support vector regression, long short-term memory, temporal convolution and soft actor-critic baselines, the proposed model reduced the mean absolute error under rapidly varying hot-air velocity from 2.91-4.76 mg/m³ to 1.68 mg/m³ and improved R² from 0.914 to 0.972. During high-load cutting episodes above 80 mg/m³, the 95th percentile error was still 4.9 mg/m³, and the best learning baseline reached 7.8 mg/m³. Based on the above results, the two can work together to reduce thermal-flow drift and transient sensor lag, especially when dust plumes evolve faster than conventional particle sensors can recover. Therefore, the above estimator is suitable for warning-oriented monitoring, calibration transfer and ventilation coordination in underground mining areas with considerable thermal disturbance.

PDF

Details