Multi-dimensional continuous optimization variable global optimization method based on reinforcement learning

Abstract

The invention discloses a multi-dimensional continuous optimization variable global optimization method based on reinforcement learning, wherein the method comprises the steps of establishing a reinforcement learning environment; selecting a specified number of optimization variables in the specified optimization variable set by using a reinforcement learning method, and then performing optimization on values of the optimization variables by using a continuous optimization variable optimization algorithm in a sequence optimization strategy; and optimizing an overall process and a constraint introduction method. According to the method, for the global optimization problem of the multi-dimensional continuous optimization variables, the purpose of intelligent optimization is achieved; the limitation of a traditional global optimization method on the number of the optimization variables can be broken through; and wide application of the artificial intelligence technology in the aspect of optimization becomes possible. The method can be applied to industrial design, manufacturing and processing, control optimization, investment decision, system engineering and other occasions with large-scale design variables; benefited from the strong intelligent combination optimization capability of deep reinforcement learning, the method also has a good global optimization effect on a system with a complex coupling relationship among variables.

Description

Technical field [0001] The present invention belongs to the field of optimization algorithms, and particularly relates to a global optimization method for large-scale continuous value optimization variables. Background technique [0002] Optimization methods can be basically divided into two categories: gradient-based optimization methods and global optimization methods. Gradient-based methods are highly efficient in optimizing single-extreme problems, but most complex multi-extreme problems need to be dealt with in engineering practice and other occasions, and gradient-based methods are prone to fall into local optimality and cannot meet the optimization requirements well. Traditional global optimization methods mainly include genetic algorithms, particle swarm optimization, etc. These methods have good global optimization capabilities and can be applied to complex multi-extreme optimization problems. However, traditional global optimization algorithms have limitations on the n...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem that the existing global optimization algorithm cannot support large-scale optimization variables and the convergence speed is slow, the present invention proposes a multi-dimensional continuous optimization variable global optimization method based on reinforcement learning, which can be realized in the actual optimization process A higher degree of intelligent optimization, without or with less manual intervention, further improves the efficiency of optimization

Images