Abstract:

The production of hazardous chemicals poses a complex and severe safety risk， with maintenance frequently resulting in accidents. The high complexity of maintenance tasks， including various sources of hazards and overlapping risk factors， such as “human-machine-environment，” increases the likelihood of accidents. To reduce risks during maintenance of hazardous chemical facilities， it focuses on optimizing the “personnel-task” assignment scheme， specifically studying the method for minimizing comprehensive risks in a typical maintenance scenario where one person is assigned multiple tasks.A mathematical model and intelligent algorithm are designed to provide an assignment scheme for this problem. The objective function focused on the “personnel-task” assignment and directly affected human risk. Furthermore， the different operating times of different personnel on different tasks resulted in varying total maintenance periods， which impacts equipment failure and environmental accident risks during the maintenance project's existence period. The overall safety risk loss function factors in human， machine， and environmental risks and considers the comprehensive index of task complexity， as well as hard constraints such as one person multiple task assignments with non-overlapping time and qualification matching. From an algorithmic perspective， the solution difficulties of the model， such as non-linearity， non-convexity， double MAX function， 0-1 matrix matching discrimination， and NP-hard， are addressed through an improved genetic algorithm based on greedy rules and adaptive learning mechanisms. The feasibility of this algorithm is verified in two scenarios， random and practical， the calculation results show that the Greedy Algorithm has a fast computation speed， high solution accuracy， and an acceptable error rate. AGA uses a greedy solution as the initial solution， which greatly improves the initial solving performance. However， due to algorithm limitations， its local search capabilities are weak， and it cannot search for better solutions based on the initial solution. HLGA， relative to AGA， has increased local search capabilities due to the addition of learning operators. Overall， the calculation results show that HLGA is indeed more suitable for solving this type of problem than AGA. It demonstrates its advantages on convergence speed， solution efficiency， and quality. A framework for modeling one-to-many personnel assignment problems with non-overlapping is provided. It offers a new reference for solving models with MAX function and matching matrix. The proposed algorithm also provides inspiration for designing algorithms that combine heuristic algorithms and intelligent algorithms. By focusing on personnel assignment， a practical solution is offerred for safety management in the hazardous chemical industry.

Key words: task complexity, one-to-many assignment, matching qualifications, exact algorithm, Genetic algorithm that combines greedy rules and adaptive learning