关键词: 进度计划, 隧道工程, 多工作面, 软逻辑, 遗传算法

Abstract: With the maturity of the construction technology, the number of extra-long tunnels built in China is gradually increasing. Extra-long tunnels are usually the key activities in a single project due to their long duration, and their duration has a direct impact on the single project. At the same time, the construction cost of long tunnels is high, and balancing the duration and cost is an important goal in the preparation of the construction schedule of long tunnels. In the current situation, the preparation of extra-long tunnel construction schedule is still based on manual experience, and the duration and cost are not yet finely and intelligently managed, which is a pain point that needs to be solved at this time of rapid construction of extra-long tunnels.
The extra-long tunnel is usually divided into several sections for parallel construction to shorten the construction period, and the parallel construction requires additional working surfaces by excavating auxiliary tunnels outside the tunnel. Firstly, the number of working faces directly affects the duration and cost of the project. Secondly, when the number of working faces is constant, where to excavate auxiliary tunnels and where to increase working faces will also have an impact on the construction period and cost. Extra-long tunnels are usually regarded as key activities, and their construction period cannot be delayed. In this case, how to save costs while meeting the construction period requirements is a problem worthy of attention.
When the tunnel realizes multi-face construction by excavating auxiliary tunnels, the sequence of each construction unit is not fixed but can be adjusted. The logical sequence of construction is soft logic, and the construction sequence will affect the duration and cost of the project. The trade-off model of duration and cost is constructed to minimize the project cost under the premise of meeting the duration. The discrete time-cost trade-off problem is NP-hard, and the addition of soft logic increases the difficulty of solving it. Therefore, the operator of genetic algorithm is improved to increase its solution speed and accuracy. Improvements include: (1)Designing adaptive crossover, mutation probability and improved catastrophe operator to avoid the algorithm from falling into local optimum; (2)Optimizing the mutation strategy to enhance the local search ability of the algorithm. Through investigation and research, data of a railroad extra-long tunnel projects are collected, including length, construction time and cost, etc.
Based on the example, the model and algorithm are verified to be effective for the optimization problem. In addition, to test the superiority of the improved algorithm, nine calculation examples with different scales and constraints are designed to compare the proposed algorithm with other algorithms horizontally. For each calculation example, the three comparison algorithms are run 20 times respectively. On the one hand, the average and minimum values of the solution results of each algorithm are counted to compare the stability and optimization ability of the algorithm operation. On the other hand, the average running time of 20 times for each algorithm is counted, and the difference in solution speed is analyzed.
The experimental results show that: (1)The proposed model and algorithm can be effectively applied to the extra-long tunnel construction schedule optimization problem. The compiler only needs to input the information of construction unit length, construction speed and cost, as well as the time and cost of auxiliary tunnels, and then can get the number and locations of auxiliary tunnels, the construction mode of working faces, and the total duration and cost and so on. (2)By comparing the proposed improved genetic algorithm with the standard genetic algorithm and particle swarm algorithm, it can be found that the other two algorithms can easily fall into local optimum when solving the problem as the size of the problem increases. Moreover, the solution obtained by the particle swarm algorithm is more volatile, while the solution obtained by the improved genetic algorithm is superior and more stable. In terms of running speed, the improved genetic algorithm is faster than the standard genetic algorithm. This is due to the fact that the improved genetic algorithm optimizes the search mechanism. The above results demonstrate that the improved genetic algorithm has better global search capability, stability and faster running speed when dealing with large-scale tunnel schedule optimization problems.
This study can help enterprises to realize the intelligent compilation of extra-long tunnel construction schedule, assist in enterprises to make scientific decisions on the number and location of auxiliary tunnels. It helps to reduce unfavorable decisions due to human experience, and helps to control the project duration and cost. This study is applicable to the case where the auxiliary tunnel is an inclined shaft, vertical shaft or horizontal cavern. When the auxiliary tunnel is in other forms, the model needs to be constructed according to its construction characteristics, which will be improved through further research.

Key words: scheduling, tunnel engineering, multiple working surfaces, soft logic, genetic algorithm