python多目标遗传算法2 NSGA2

""" 快速非支配排序 """ import random import numpy as np def nonDominationSort(pops, fits): """快速非支配排序算法 Params: pops: 种群，nPop * nChr 数组 fits: 适应度， nPop * nF 数组 Return: ranks: 每个个体所对应的等级，一维数组 """ nPop = pops.shape[0] nF = fits.shape[1] # 目标函数的个数 ranks = np.zeros(nPop, dtype=np.int32) nPs = np.zeros(nPop) # 每个个体p被支配解的个数 sPs = [] # 每个个体支配的解的集合，把索引放进去 for i in range(nPop): iSet = [] # 解i的支配解集 for j in range(nPop): if i == j: continue isDom1 = fits[i] <= fits[j] isDom2 = fits[i] < fits[j] # 是否支配该解-> i支配j if sum(isDom1) == nF and sum(isDom2) >= 1: iSet.append(j) # 是否被支配-> i被j支配 if sum(~isDom2) == nF and sum(~isDom1) >= 1: nPs[i] += 1 sPs.append(iSet) # 添加i支配的解的索引 r = 0 # 当前等级为 0， 等级越低越好 indices = np.arange(nPop) while sum(nPs==0) != 0: rIdices = indices[nPs==0] # 当前被支配数为0的索引 ranks[rIdices] = r for rIdx in rIdices: iSet = sPs[rIdx] nPs[iSet] -= 1 nPs[rIdices] = -1 # 当前等级的被支配数设置为负数 r += 1 return ranks # 拥挤度排序算法 def crowdingDistanceSort(pops, fits, ranks): """拥挤度排序算法 Params: pops: 种群，nPop * nChr 数组 fits: 适应度， nPop * nF 数组 ranks：每个个体对应的等级，一维数组 Return： dis: 每个个体的拥挤度，一维数组 """ nPop = pops.shape[0] nF = fits.shape[1] # 目标个数 dis = np.zeros(nPop) nR = ranks.max() # 最大等级 indices = np.arange(nPop) for r in range(nR+1): rIdices = indices[ranks==r] # 当前等级种群的索引 rPops = pops[ranks==r] # 当前等级的种群 rFits = fits[ranks==r] # 当前等级种群的适应度 rSortIdices = np.argsort(rFits, axis=0) # 对纵向排序的索引 rSortFits = np.sort(rFits,axis=0) fMax = np.max(rFits,axis=0) fMin = np.min(rFits,axis=0) n = len(rIdices) for i in range(nF): orIdices = rIdices[rSortIdices[:,i]] # 当前操作元素的原始位置 j = 1 while n > 2 and j < n-1: if fMax[i] != fMin[i]: dis[orIdices[j]] += (rSortFits[j+1,i] - rSortFits[j-1,i]) / \ (fMax[i] - fMin[i]) else: dis[orIdices[j]] = np.inf j += 1 dis[orIdices[0]] = np.inf dis[orIdices[n-1]] = np.inf return dis if __name__ == "__main__": y1 = np.arange(1,5).reshape(4,1) y2 = 5 - y1 fit1 = np.concatenate((y1,y2),axis=1) y3 = 6 - y1 fit2 = np.concatenate((y1,y3),axis=1) y4 = 7 - y1 fit3 = np.concatenate((y1,y4),axis=1) fit3 = fit3[:2] fits = np.concatenate((fit1,fit2,fit3), axis=0) pops = np.arange(fits.shape[0]).reshape(fits.shape[0],1) random.seed(123) # 打乱数组 indices = np.arange(fits.shape[0]) random.shuffle(indices) fits = fits[indices] pops = pops[indices] print(indices) # 首先测试非支配排序算法 ranks = nonDominationSort(pops, fits) print('ranks:', ranks) # 测试拥挤度排序算法 dis = crowdingDistanceSort(pops,fits,ranks) print("dis:", dis)