Adaptive Human Learning Optimization Algorithm Based On Feedback Control And Its Application

Human Learning Optimization(HLO)algorithm is a meta-heuristic optimization algorithm with simple implementation and excellent performance,which performs the search for the optimal solution of the problem by simulating the random learning,individual learning and social learning behaviors in the process of human learning.Inspired by the idea of feedback control,this thesis proposes the adaptive human learning optimization algorithm based on the feedback control strategy,which achieves a better tradeoff between exploration and exploitation of the algorithm by dynamically adjust the operation rate of the random learning operator.On this basis,an improved selective evolutionary random network is designed based on the adaptive HLO and random feedforward neural networks,which is further applied to fault diagnosis.The main research work of this thesis is as follows:(1)The functions and characteristics of the random learning operator of HLO are deeply studied.According to the requirements of exploration and exploitation in the search process of HLO,the adaptive human learning optimization algorithm based on incremental tuning rule(AHLOIT)and the adaptive human learning optimization algorithm based on proportional tuning rule(AHLOPT)are proposed by introducing the mechanism of feedback control.The dynamic tuning of the functions of the random learning operator is realized by the proposed adaptive pr strategies,which effectively improves the search performance of the algorithm.The results of CEC15 benchmark functions and multi-dimensional knapsack problems show that the proposed AHLOPT and AHLOIT have better comprehensive performance than previous adaptive HLO variants and other 4 meta-heuristic algorithms.(2)The adaptive human learning optimization algorithm based on intelligent PID control(AHLOPID)is proposed by introducing the PID controllers,which are used to adjust the functions of the random learning operator by tuning pr based on the feedback control mechanism.The parameters involved in the reference and the PID controllers are defined as hyper-parameters,and then an optimization and tuning method of hyperparameters is designed based on HLO,in which the basic HLO algorithm is used to automatically optimize the hyper-parameters,and therefore the adaptive strategy based on PID control can be easily and efficiently implemented.Then four different PID controller structures are studied and the influences on the performance of HLO are investigated.The simulation results of the CEC17 benchmark functions and multidimensional knapsack problems show that the global optimization ability of AHLOPID is further improved.(3)Aiming at the unstable performance of random feedforward neural network,an improved selective evolutionary random network(ISERN)method is proposed.In the proposed ISERN,the AHLOPID algorithm is used to optimize the structure of the original random feedforward neural network to obtain a flexible and well-structured primitive neural network(PNN).Then,the real working network for the specific problems is optimized based on the PNN and the feature selection is carried out cooperatively and simultaneously as a wrapper.Finally,the ISERN is applied to the fault diagnosis applications of the ocean-going ship desalination system.The simulation results of UCI data-sets and fault diagnosis show the ISERN method has a satisfactory classification ability.