本文以 TSP 问题为例,通过具体代码,说明模拟退火算法的迭代过程。
关于 TSP 问题的介绍从略,算法模块的代码(algorithm.py)如下,注释中已说明算法的迭代过程:
from datetime import datetime
from typing import List
import math
import random
class SimulatedAnnealing(object):
"""
模拟退火
"""
def __init__(self, num_point: int, mat_dist: List[List[float]],
t_start: int = 5000, t_end: int = 10 ** (-8), coe_anneal: float = 0.98, num_iter: int = 1000):
"""
模拟退火,数据初始化
:param num_point: TSP 节点数量
:param mat_dist: 距离矩阵
:param t_start: 初始温度
:param t_end: 结束温度
:param coe_anneal: 降温系数
:param num_iter: 每个温度时的迭代次数
"""
# 问题参数
self.num_point = num_point
self.mat_dist = mat_dist
# 算法参数
self.t_start = t_start
self.t_end = t_end
self.coe_anneal = coe_anneal
self.num_iter = num_iter
print("初始温度: {}".format(self.t_start))
print("结束温度: {}".format(self.t_end))
print("降温系数: {}".format(self.coe_anneal))
print("每个温度时的迭代次数: {}".format(self.num_iter), '\n')
# 结果
self.route_opt, self.distance_opt = [], None # 最优路径、最优路径距离
self.route_res, self.distance_res = [], None # 结果路径、结果路径距离
def run(self):
"""
算法运行
:return: 无
"""
dts = datetime.now()
random.seed(1024)
# 初始解
route = self._init_solution()
obj = self._get_distance(route=route)
self.route_opt, self.distance_opt = route, obj
# loop 1: 降温
t = self.t_start
count_anneal = 0
while t > self.t_end:
print("当前温度: {}".format(t))
print("降温次数: {}".format(count_anneal), '\n')
# loop 2: 每个温度的迭代
for i in range(self.num_iter):
route_old = route.copy()
route = self._create_new_solution(route=route)
obj, obj_old = self._get_distance(route=route), self._get_distance(route=route_old)
dis_obj = obj - obj_old
print("原有路径: {}".format(route_old))
print("原有路径距离: {}".format(obj_old))
print("当前路径: {}".format(route))
print("当前路径距离: {}".format(obj), '\n')
# Metropolis 准则
if dis_obj >= 0:
r = random.random()
if r >= math.exp(-dis_obj / t):
route = route_old.copy()
print("随机数 {0} 过大,当前路径距离 {1} 较差,恢复原有解".format(r, obj))
# 更新全局解
if obj < self.distance_opt:
self.route_opt, self.distance_opt = route, obj
print("全局最优路径更新为: {}".format(self.route_opt))
print("距离: {}".format(self.distance_opt), '\n')
t *= self.coe_anneal
count_anneal += 1
# 运行结果
self.route_res = route.copy()
self.distance_res = obj
print("结果路径: {}".format(self.route_res))
print("结果路径距离: {}".format(self.distance_res), '\n')
print("最优路径: {}".format(self.route_opt))
print("最优路径距离: {}".format(self.distance_opt), '\n')
dte = datetime.now()
tm = round((dte - dts).seconds + (dte - dts).microseconds / (10 ** 6), 3)
print("算法运行时间: {} s".format(tm), '\n')
def _init_solution(self):
"""
初始解
:return: route: 初始路径
"""
route = [i for i in range(self.num_point)]
return route
def _get_distance(self, route: List[int]) -> float:
"""
计算路径距离
:param route: 路径
:return: 距离
"""
distance = sum(self.mat_dist[route[i]][route[i + 1]] for i in range(len(route) - 1))
return distance
def _create_new_solution(self, route: List[int]) -> List[int]:
"""
产生一个新解
:param route: 当前解
:return: route_: 生成的新解
"""
route_ = route.copy()
# 通过随机交换两个位置的方式产生新解
pos1, pos2 = random.randint(0, self.num_point - 1), random.randint(0, self.num_point - 1)
tmp, route_[pos1] = route_[pos1], route_[pos2]
route_[pos2] = tmp
return route_
生成随机算例,并调用算法模块进行求解的主程序代码(main.py)如下:
from datetime import datetime
import math
import random
from algorithm import SimulatedAnnealing
dts = datetime.now()
""" 参数 """
# 地点数量
num_point = 20
# 坐标范围、边界宽度
ran_coo = (0, 100)
edge = 1
# 坐标列表、距离矩阵
random.seed(1024)
list_coo = [(random.randint(ran_coo[0] + edge, ran_coo[1] - edge),
random.randint(ran_coo[0] + edge, ran_coo[1] - edge)) for _ in range(num_point)]
mat_dist = [[math.sqrt((list_coo[i][0] - list_coo[j][0]) ** 2 + (list_coo[i][1] - list_coo[j][1]) ** 2)
for j in range(num_point)] for i in range(num_point)]
""" 算法 """
t_start, t_end, coe_anneal, num_iter = 1500, 1000, 0.95, 1000
simulated_annealing = SimulatedAnnealing(num_point=num_point, mat_dist=mat_dist,
t_start=t_start, t_end=t_end, coe_anneal=coe_anneal, num_iter=num_iter)
simulated_annealing.run()
dte = datetime.now()
tm = round((dte - dts).seconds + (dte - dts).microseconds / (10 ** 6), 3)
print("程序运行总时间: {} s".format(tm), '\n')
汪定伟《智能优化方法》