Optimization model of tunnel construction machine groups configuration based on neural optimization machine

In order to improve the efficiency of tunneling machine group operation, accelerate the construction progress, and maximize project benefit, the cooperative operation of tunnel construction machine groups was investigated. Unlike conventional multi-objective optimization methods, an optimization model based on a neural network was proposed. Using queueing theory, the machine utilization efficiencies of the mucking, support, and lining subsystems were calculated, after which the corresponding cost and duration expressions were obtained. A time-cost multi-objective optimization model was then established after nondimensionalization, and the problem was solved using a neural optimization machine (NOM) constructed on the basis of a neural network. Compared with conventional metaheuristic algorithms such as particle swarm optimization (PSO) and genetic algorithms (GA), NOM provides a differentiable optimization framework and allows the objective function to have an arbitrary architecture and activation function, is not limited to a specific optimization problem, and can be embedded into a differentiable optimization procedure for updating design variables through backpropagation. Finally, the proposed method was validated by a tunnel project. The original scheme exceeded the contractual duration and the construction progress was seriously delayed. After optimization, the construction durations were reduced by 10.1%, 11.6%, and 13.0%, respectively, and all satisfied the contract requirements. A shorter duration corresponded to a larger number of construction machines, and the construction cost increased accordingly.