经典PSO粒子群算法python实现（面向对象）

以下是笔者调试好的面向对象的PSO算法，已封装好，相关参数在主函数__main__下修改即可。

注：很多同学觉得面向对象很难，其实很简单，网上找几个实例运行看看就能懂了。其中self就是一个初始化自己的过程，相当于一个容器，你告诉机器要放入哪些参数和函数，这些参数和函数叫什么名字。

以下设置可行域为[-10,10]，维度为5维，目标函数为

显然，我们最优解为[0,0,0,0,0]

以下为求解该问题的python程序

import numpy as np
def y(x):
    return np.sum(x**2)
class PSO:
    def __init__(self,y,num_iter,num_particles,num_dimensions,c1,c2,w,bound_low,bound_high,vel_low,vel_high):
        self.fitness_function=y
        self.num_iter=num_iter
        self.num_particles=num_particles
        self.num_dimensions=num_dimensions
        self.c1=c1
        self.c2=c2
        self.w=w
        self.vel_low=vel_low
        self.vel_high=vel_high
        self.bound_low=bound_low
        self.bound_high=bound_high
        # 粒子初始化
        self.pos=np.random.uniform(bound_low,bound_high,(num_particles,num_dimensions))
        self.vel=np.random.uniform(vel_high,vel_high,(num_particles,num_dimensions))
        self.pbest_pos=self.pos.copy()
        self.pnow_val=np.zeros(num_particles)
        self.pbest_val=np.full(num_particles,np.inf)
        self.gbest_pos=np.zeros((1,num_dimensions))
        self.gbest_val=np.inf

    # 边界速度限制处理
    def limit_eachiter(self):
        self.pos[self.pos>self.bound_high]=self.bound_high
        self.pos[self.pos<self.bound_low]=self.bound_low
        self.vel[self.vel>self.vel_high]=self.vel_high
        self.vel[self.vel<self.vel_low]=self.vel_low

    # 计算适应度、更新最优记录
    def fitness_eachiter(self):
        for i in range(self.num_particles):
            self.pnow_val[i]=self.fitness_function(self.pos[i])
            if self.pnow_val[i]<self.pbest_val[i]:
                self.pbest_val[i]=self.pnow_val[i].copy()
                self.pbest_pos[i]=self.pos[i].copy()
            if self.pnow_val[i]<self.gbest_val:
                self.gbest_val=self.pnow_val[i].copy()
                self.gbest_pos=self.pos[i].copy()

    # 粒子速度、位置更新，运用广播机制
    def update_eachiter(self):
        self.vel=self.w*self.vel+self.c1*np.random.rand()*(self.pbest_pos-self.pos)+\
            self.c2*np.random.rand()*(self.gbest_pos-self.pos)
        self.pos=self.pos+self.vel

    # 主程序
    def pso(self):
        for i in range(self.num_iter):
            self.limit_eachiter()
            self.fitness_eachiter()
            self.update_eachiter()
if __name__=='__main__':
    num_iter=600
    num_particles=50
    num_dimensions=5
    c1=2
    c2=2
    w=0.8
    bound_low=-10
    bound_high=10
    vel_low=-1
    vel_high=1
    PSO=PSO(y,num_iter,num_particles,num_dimensions,c1,c2,w,bound_low,bound_high,vel_low,vel_high)
    PSO.pso()
    print('Optimal solution: x = {}, f(x) = {}'.format(PSO.gbest_pos, PSO.gbest_val))

运行一下，结果为：

Optimal solution: x = [-1.17921320e-10 -6.00967243e-11  6.29614150e-11  3.05093263e-10
 -4.88228938e-11], f(x) = 1.1694676763834684e-19

完美！