python 实现粒子群算法（基本型）

最近的学习python，尝试了python实现粒子群算法的基本过程，定义了变量的上下界，利用pso 求解简单约束的问题，Spere 函数进行了检验。

感觉利用MATLAB的矩阵形式编写的pso 较快。

适应值函数通过对象的方式传递给pso class。

#-*- coding:utf-8 -*-
# 作者：gongxu
# 时间：20181111
# 基本的粒子群算法，能够解决约束问题，约束问题仅限上下边界。

import numpy as np   
import matplotlib.pyplot as plt

class pso:
	"""
	w: 惯性系数
	c1：学习因子1
	c2：学习因子2
	max_iter：最大迭代次数
	var_num：变量的个数
	bound：变量的边界
	pop:种群的个数
	fitness_fun:适应值函数
	"""
	def __init__(self,w,c1,c2,max_iter,var_num,bound,pop,fitness_fun):
		self.w =w
		self.c1 =c1
		self.c2 =c2
		self.max_iter=max_iter
		self.var_num=var_num
		self.bound=bound
		self.pop=pop
		self.fitness_fun=fitness_fun
		self.best_fit=[]
		self.fitness=np.zeros(pop)
		self.positions=bound[1,:]+np.random.rand(pop,var_num)*(bound[0,:]-bound[1,:])
		self.velecitys=0.1*self.positions.copy()
		self.iter_fit=np.zeros(max_iter)
		self.fitness=self.fitness_fun(self.positions)
		self.best_fit=self.fitness.copy()
		self.best_pos=self.positions.copy()
		self.global_fit=np.min(self.fitness)
		self.global_pos=self.positions[np.argmin(self.fitness)].copy()
	def fly_finding(self):
		self.velecitys=self.w*self.velecitys+self.c1*np.random.rand(self.pop,self.var_num)*(self.best_pos-self.positions) \
		+self.c2*np.random.rand(self.pop,self.var_num)*(self.global_pos-self.positions)
		for i in range(self.pop):
			index_up=self.velecitys[i]>self.bound[0,:]
			self.velecitys[i,index_up]=self.bound[0,index_up].copy()
			index_down=self.velecitys[i]<self.bound[1,:]
			self.velecitys[i,index_down]=self.bound[1,index_down].copy()
		self.positions=self.positions.copy()+self.velecitys.copy()
		for i in range(self.pop):
			index_up=self.positions[i]>self.bound[0,:]
			self.positions[i,index_up]=self.bound[0,index_up].copy()
			index_down=self.positions[i]<self.bound[1,:]
			self.positions[i,index_down]=self.bound[1,index_down].copy()
	def update_position(self):
		self.fitness=self.fitness_fun(self.positions)
		index=self.fitness<self.best_fit
		self.best_fit[index]=self.fitness[index].copy()
		self.best_pos[index,:]=self.positions[index,:].copy()
		self.global_fit=np.min(self.best_fit)
		self.global_pos=self.best_pos[np.argmin(self.best_fit),:].copy()
	def run(self):
		for i in range(0,self.max_iter):
			self.fly_finding()
			self.update_position()
			self.iter_fit[i]=self.global_fit.copy()
			print('gloabal best fitness:',self.iter_fit[i],end='\n')
	def show_result(self):
		plt.plot(range(self.max_iter),self.iter_fit,'-b')
		plt.title('converage of process in pso')
		plt.xlabel('iter')
		plt.ylabel('fitness')
		plt.show()
	def print_result(self):
		print('the result is showing...')
		print('best_fitness for min problem:',self.global_fit)
		print('best solution for result:',self.global_pos)


def fitness_fun(x):
	result=np.sum(x*x,axis=1)
	return result

def main():
	w=1
	c1=2
	c2=2
	max_iter=1000
	var_num=10
	pop=50
	bound=np.zeros((2,var_num))
	bound[0,:]=np.ones(var_num)*10    #上边界，每个变量的上边界可以不同
	bound[1,:]=-np.ones(var_num)*10	  #下边界，每个变量的下边界可以不同
	pso_new=pso(w,c1,c2,max_iter,var_num,bound,pop,fitness_fun)
	pso_new.run()
	pso_new.print_result()
	pso_new.show_result()
if __name__=='__main__':
	main()