INTELLIGENT AVOIDANCE METHOD FOR LOCAL OPTIMAL SOLUTIONS IN NONLINEAR PROGRAMMING BASED ON REINFORCEMENT LEARNING

Guodong Wang

Keywords

Reinforcement learning, nonlinear programming, local optimalsolution avoidance, long short-term memory (LSTM) network, graphattention model

Abstract

To maintain a dynamic balance between efficient exploration and exploitation and avoid falling into local optima, this study investigates an intelligent avoidance method for local optimal solutions in nonlinear programming based on reinforcement learning. By transforming the nonlinear programming problem into a deterministic optimisation problem through deterministic descriptions; using long short-term memory networks as the encoder for the policy network and value network in reinforcement learning, extracting temporal features of the solution space within the policy network, estimating state value within the value network, and adjusting the policy network; employing a graph attention model that fuses node and edge features as the decoder for the policy network, combining temporal features to generate a probability distribution for breaking the solution space, and adjusting nonlinear programming decisions; designing a local optimal solution intelligent avoidance reward function for the value network based on the deterministic optimisation problem of nonlinear programming; aiming to maximise the reward function, training the reinforcement learning network by combining policy gradient optimisation algorithms with adaptive Q-learning algorithms, updating the parameters of the policy and value networks to ensure a dynamic balance between exploration and exploitation, thereby achieving intelligent avoidance of local optima in nonlinear programming. Experimental results demonstrate that this method exhibits superior reinforcement learning network training performance, with the optimal learning rate being 1×10-3; in the nonlinear programming problem of vehicle network task offloading, applying this method effectively reduces vehicle energy consumption and resource leasing costs, with maximum energy consumption approximately 100 J and maximum resource leasing cost around 2 Yuan.

Important Links:

Go Back