Research On Particle Swarm Optimization Algorithm Based On Agent Behavior Decision

Agent is the concrete realization of artificial intelligence.In swarm intelligence,each individual in the population can be regarded as an agent.These agents decide the next search trajectory according to certain rules to approximate the global optimal solution of the optimization problem.Nowadays,the Particle Swarm Optimization(PSO)algorithm and its variants have proven to be useful methods to solve complex optimization problems.In the past20 years,PSO has attracted widespread attention in the academic community.However,the PSO algorithm has the problems of particle position oscillation,insufficient diversity and easy to fall into local traps in the search process.In response to such problems,this paper proposes three PSO variants algorithms.The main research contents of this paper are as follows:1.Aiming at the problems of particle position oscillation and partial dimensional degradation in particle swarm algorithm,this paper proposes Particle Swarm Optimization with tanh Function and Dimensional Learning(TDPSO).Specifically,the algorithm adaptively adjusts its own learning factor according to the distance between the particle and its learning object,so that the movement of the particle is larger when it is far from the learning object,and smaller when it is close to the learning object,which is reduced to a certain extent.The phenomenon of oscillation when particles are in motion.On the other hand,the particles will learn the information of each dimension of the global optimal particle,and the particles will move to the area where the global optimal particle is located.This further reduces the phenomenon of particle regression in some dimensions during the search process and improves the convergence speed of the algorithm.Through simulation experiments on 20 commonly used benchmark test functions,it can be seen that the TDPSO algorithm can find better solutions.2.Aiming at the problem that a single learning exemplar will lead to insufficient population diversity,this paper proposes the Particle Swarm Optimization with Social Influence(PSOSI).Specifically,each particle in PSOSI will choose the global best particle and the best companion particle as its learning model.To further describe the influence of different exemplars,we define two gravity coefficients inspired by mechanics.These gravity coefficients not only ensure that the current best experience of the population is shared,but also enrich the diversity of the population.In addition,each particle will further perform a variable scale search according to the distance between itself and its best companion particle in each dimension,which improves the overall convergence ability of the algorithm.Through simulation experiments on all the 28 functions of CEC2013 test suit,it can be seen that PSOSI has good performance in convergence speed and convergence accuracy.3.In view of the problem that the global optimal particle may not be near the actual global optimal solution during the search process,which makes the swarm easy to fall into local optimality,this paper proposes a new method named Particle Swarm Optimization based on progress rate(PRPSO).The PRPSO algorithm adopts fitness and progress rate as two indicators to evaluate the performance of each particle.The particles own the top three fitness values in the population are recorded as high-quality particles,representing the optimal experience among current swarm,on the other hand particles with the progress rate of the first 10% of the population are viewed as upstart particles,and they are most likely to be located near the global optimal solution.At the same time,three different learning models are designed for different particles and different search stages to enrich the wisdom of the population when facing different complex optimization problems.Finally,the concept of elastic collision is proposed to avoid the sudden loss of velocity when particles collide with the boundary of the solution space.Through simulation experiments on the whole 30 functions of CEC2017 test suit,it can be seen that PRPSO can effectively maintain the diversity of the population during the search process and increase the possibility of the population to find the actual global optimal solution.In addition,the practical application value of the PRPSO algorithm is further verified through the practical application of short-term power load forecasting.