Variable Selection Based on Hybrid Ant Colony Optimization-Simulated Annealing for Spectroscopy Quantitative Analysis of Underground Cable Tunnel

Cable tunnel is widely used in urban electric power construction but environment of cable tunnel may contain toxic and harmful gas. To analysis the densities of different components in mixture gases, near-infrared(NIR) spectroscopy is adopted. However, full spectrum prediction model may be inaccurate and redundant. This paper proposes a hybrid ant colony optimization-simulated annealing variable selection method for spectroscopy quantitative analysis. The proposed method has two parts: the SA part calculate the fitness of new candidate solutions and determine whether to accept the new solution, the ACO part generates new candidate solutions based on the pheromone deposited on each variable and uses new solution to update the pheromone after the SA part has finished. To demonstrate the effectiveness of the proposed method, PLS with full spectrum, uninformative variable elimination by PLS, elastic net, genetic algorithm with PLS, ant colony optimization with PLS and the proposed method are used to build components quantitative analysis models based on real gas data and the results are compared and discussed. The experimental results verify the prediction precision of the proposed method is higher.