【回溯算法】【Python实现】TSP旅行售货员问题

问题描述

• 给定一组城市和它们之间的距离矩阵，找到一条距离最短的路径，使得旅行商从一个城市出发，经过所有城市恰好一次，并最终回到出发城市

---

回溯算法

• 旅行售货员问题的解空间是一棵排列树

• 当$i=n$时，算法搜索至叶结点，其相应的路径长度为$cd$，如果$cd<bestd$，则表示当前解优于当前最优解，此时更新$bestd$

• 当$i<n$时，当前扩展结点位于排列树的第$i$层，图$G$中存在从顶点$x[i]$到顶点$x[i+1]$的边时，$x[1:i+1]$构成图$G$的一条路径，且当$x[1:i+1]$的路径长度小于当前最优值时算法进入排列树的第$i+1$层，否则将剪去相应的子树

---

Python实现

import numpy as np


def backtrack_tsp(cities):
    n = len(cities)
    visited = [False] * n  # 记录城市是否已经被访问

    shortest_path = []
    shortest_distance = float('inf')

    def distance(city1, city2):
        x1, y1 = city1
        x2, y2 = city2

        return np.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)

    # 创建距离矩阵
    dist_matrix = np.zeros((n, n))
    for i in range(n):
        for j in range(n):
            dist_matrix[i][j] = distance(cities[i], cities[j])

    def backtrack(path, distance):
        nonlocal shortest_path, shortest_distance

        if len(path) == n:  # 所有城市都已经访问过
            distance += dist_matrix[path[-1]][path[0]]  # 回到起点的距离

            if distance < shortest_distance:  # 更新最短路径和最短距离
                shortest_path = path[:]
                shortest_distance = distance

            return

        last_city = path[-1] if path else 0  # 上一个访问的城市

        for next_city in range(n):
            if not visited[next_city]:
                visited[next_city] = True
                path.append(next_city)
                distance += dist_matrix[last_city][next_city]

                backtrack(path, distance)

                # 恢复回溯前状态
                distance -= dist_matrix[last_city][next_city]
                path.pop()
                visited[next_city] = False

    # 开始回溯搜索
    visited[0] = True
    backtrack([0], 0)

    return shortest_path, shortest_distance


cities = [(0, 0), (1, 5), (2, 3), (5, 2), (6, 4)]
shortest_path, shortest_distance = backtrack_tsp(cities)

print(f'最短路径: {shortest_path}')
print(f'最短距离: {shortest_distance}')

最短路径: [0, 2, 1, 4, 3]
最短距离: 18.56187155119086

---

时间复杂性

• 回溯算法解TSP问题的时间复杂性为$O(n!)$

---