Search

As a population-based stochastic optimization algorithm, Gravitational Search Algorithm (GSA) has attracted numerous interests and has been applied in various applications. However, GSA has drawbacks such as uneven search and premature convergence in practical applications. This paper specifically explains the inherent characteristic of GSA in prioritizing the center position. Correspondingly, an improvement strategy of fitness normalization with mass shift is proposed, creating a situation where gravity and repulsion are mixed. Then, the global best mechanism with weights is incorporated into the particle's velocity update formula, which compensates for the difficulties in the later exploitation stage. Finally, an empirical formula for the initial gravitational constant related to the size of the solution space is proposed, which enhances the global search ability together with the former strategy. 12 shifted benchmark functions are used to construct 20 optimization problems ranging from 2 to 120 dimensions. The average performance of the proposed algorithm, other GSA and well-known algorithms are compared under the same budget. The results demonstrate that the proposed GSA not only effectively addresses the drawbacks of GSA and maintains good performance, but also exhibits strong competitiveness compared to various similar algorithms. [ABSTRACT FROM AUTHOR]

• A combination of Deep Q-Learning algorithm and metaheuristic GSA is offered. • GSA initializes the weights and the biases of the neural networks. • A comparison with classical random, metaheuristic PSO and GWO is carried out. • The validation is done on real-time nonlinear servo system position control. • The drawbacks of randomly initialized neural networks are mitigated. This paper presents a novel Reinforcement Learning (RL)-based control approach that uses a combination of a Deep Q-Learning (DQL) algorithm and a metaheuristic Gravitational Search Algorithm (GSA). The GSA is employed to initialize the weights and the biases of the Neural Network (NN) involved in DQL in order to avoid the instability, which is the main drawback of the traditional randomly initialized NNs. The quality of a particular set of weights and biases is measured at each iteration of the GSA-based initialization using a fitness function aiming to achieve the predefined optimal control or learning objective. The data generated during the RL process is used in training a NN-based controller that will be able to autonomously achieve the optimal reference tracking control objective. The proposed approach is compared with other similar techniques which use different algorithms in the initialization step, namely the traditional random algorithm, the Grey Wolf Optimizer algorithm, and the Particle Swarm Optimization algorithm. The NN-based controllers based on each of these techniques are compared using performance indices specific to optimal control as settling time, rise time, peak time, overshoot, and minimum cost function value. Real-time experiments are conducted in order to validate and test the proposed new approach in the framework of the optimal reference tracking control of a nonlinear position servo system. The experimental results show the superiority of this approach versus the other three competing approaches. [ABSTRACT FROM AUTHOR]