along/HPCC2025
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

修改蒙特卡洛采样法

Browse Source
This commit is contained in:
weixin_46229132 2025-03-14 11:01:02 +08:00
parent b3b5e597b8
commit dfec68e122
2 changed files with 30 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
env.py
View File

@ -38,9 +38,9 @@ class PartitionMazeEnv(gym.Env):
self.phase = 0 # 阶段控制，0：区域划分阶段，1：迷宫初始化阶段，2：走迷宫阶段 self.phase = 0 # 阶段控制，0：区域划分阶段，1：迷宫初始化阶段，2：走迷宫阶段
self.partition_step = 0 # 区域划分阶段步数，范围 0~4 self.partition_step = 0 # 区域划分阶段步数，范围 0~4
# TODO 切的刀数现在固定为4（2+2） self.CUT_NUM = 6 # 横切一半，竖切一半
self.partition_values = np.zeros( self.partition_values = np.zeros(
4, dtype=np.float32) # 存储 c₁, c₂, r₁, r₂ self.CUT_NUM, dtype=np.float32) # 存储 c₁, c₂, r₁, r₂
# 定义动作空间：全部动作均为 1 维连续 [0,1] # 定义动作空间：全部动作均为 1 维连续 [0,1]
self.action_space = spaces.Box( self.action_space = spaces.Box(
@ -51,7 +51,7 @@ class PartitionMazeEnv(gym.Env):
# 阶段 0 状态：前 4 维表示已决策的切分值（未决策部分为 0） # 阶段 0 状态：前 4 维表示已决策的切分值（未决策部分为 0）
# 阶段 1 状态：车辆位置 (2D) # 阶段 1 状态：车辆位置 (2D)
self.observation_space = spaces.Box( self.observation_space = spaces.Box(
low=0.0, high=1.0, shape=(4 + 2 * self.num_cars,), dtype=np.float32) low=0.0, high=1.0, shape=(self.CUT_NUM + 2 * self.num_cars,), dtype=np.float32)
# 切分阶段相关变量 # 切分阶段相关变量
self.col_cuts = [] # 存储竖切位置（c₁, c₂），当值为0时表示不切 self.col_cuts = [] # 存储竖切位置（c₁, c₂），当值为0时表示不切
@ -72,7 +72,7 @@ class PartitionMazeEnv(gym.Env):
# 重置所有变量，回到切分阶段（phase 0） # 重置所有变量，回到切分阶段（phase 0）
self.phase = 0 self.phase = 0
self.partition_step = 0 self.partition_step = 0
self.partition_values = np.zeros(4, dtype=np.float32) self.partition_values = np.zeros(self.CUT_NUM, dtype=np.float32)
self.col_cuts = [] self.col_cuts = []
self.row_cuts = [] self.row_cuts = []
self.init_maze_step = 0 self.init_maze_step = 0
@ -103,7 +103,7 @@ class PartitionMazeEnv(gym.Env):
[self.partition_values, np.zeros(np.array(self.car_pos).flatten().shape[0], dtype=np.float32)]) [self.partition_values, np.zeros(np.array(self.car_pos).flatten().shape[0], dtype=np.float32)])
# 如果未完成 4 步，则仍处于切分阶段，不发奖励，done 为 False # 如果未完成 4 步，则仍处于切分阶段，不发奖励，done 为 False
if self.partition_step < 4: if self.partition_step < self.CUT_NUM:
return state, 0.0, False, False, {} return state, 0.0, False, False, {}
else: else:
# 完成 4 步后，计算切分边界 # 完成 4 步后，计算切分边界

37
mtkl_sovler.py
View File

@ -2,11 +2,12 @@ import random
import math import math
import yaml import yaml
import json import json
import numpy as np
# 固定随机种子，便于复现 # 固定随机种子，便于复现
random.seed(42) random.seed(42)
num_iterations = 1000000 num_iterations = 10000
# --------------------------- # ---------------------------
# 参数设置 # 参数设置
@ -37,14 +38,21 @@ best_solution = None
for iteration in range(num_iterations): for iteration in range(num_iterations):
# 随机生成分区的行分段数与列分段数 # 随机生成分区的行分段数与列分段数
R = random.randint(1, 5) # 行分段数 R = random.randint(0, 5) # 行分段数
C = random.randint(1, 5) # 列分段数 C = random.randint(0, 5) # 列分段数
# 生成随机的行、列分割边界 # 生成随机的行、列分割边界
row_boundaries = sorted(random.sample(range(1, H), R - 1)) horiz = [np.clip(np.floor(random.random() * 10) /10, 0.0, 0.9) for _ in range(R)]
row_boundaries = [0] + row_boundaries + [H] horiz = sorted(set(horiz))
col_boundaries = sorted(random.sample(range(1, W), C - 1)) horiz = horiz if horiz else []
col_boundaries = [0] + col_boundaries + [W] row_boundaries = [0] + horiz + [1]
row_boundaries = [boundary * H for boundary in row_boundaries]
vert = [np.clip(np.floor(random.random() * 10) /10, 0.0, 0.9) for _ in range(C)]
vert = sorted(set(vert))
vert = vert if vert else []
col_boundaries = [0] + vert + [1]
col_boundaries = [boundary * W for boundary in col_boundaries]
# --------------------------- # ---------------------------
# 根据分割边界生成所有矩形任务 # 根据分割边界生成所有矩形任务
@ -116,12 +124,14 @@ for iteration in range(num_iterations):
total_flight_time = sum(task['flight_time'] for task in tasks) total_flight_time = sum(task['flight_time'] for task in tasks)
if tasks: if tasks:
# 车辆从区域中心到第一个任务中心 # 车辆从区域中心到第一个任务中心
car_time = math.dist(tasks[0]['center'], region_center) * car_time_factor car_time = math.dist(tasks[0]['center'],
region_center) * car_time_factor
# 依次经过任务中心 # 依次经过任务中心
for j in range(len(tasks) - 1): for j in range(len(tasks) - 1):
prev_center = tasks[j]['center'] prev_center = tasks[j]['center']
curr_center = tasks[j + 1]['center'] curr_center = tasks[j + 1]['center']
car_time += math.dist(curr_center, prev_center) * car_time_factor car_time += math.dist(curr_center,
prev_center) * car_time_factor
# 回到区域中心 # 回到区域中心
car_time += math.dist(region_center, curr_center) * car_time_factor car_time += math.dist(region_center, curr_center) * car_time_factor
else: else:
@ -167,18 +177,21 @@ if best_solution is not None:
car_paths = {} car_paths = {}
for i in range(k): for i in range(k):
num_tasks = len(best_solution['system_tasks'][i]) num_tasks = len(best_solution['system_tasks'][i])
print(f"系统 {i}: 完成时间 T = {best_solution['T_k_list'][i]}, 飞行任务数量: {num_tasks}") print(
f"系统 {i}: 完成时间 T = {best_solution['T_k_list'][i]}, 飞行任务数量: {num_tasks}")
tasks = best_solution['system_tasks'][i] tasks = best_solution['system_tasks'][i]
tasks.sort(key=lambda r: math.hypot(r['center'][0] - region_center[0], tasks.sort(key=lambda r: math.hypot(r['center'][0] - region_center[0],
r['center'][1] - region_center[1])) r['center'][1] - region_center[1]))
if tasks: if tasks:
print(f"轨迹路线: 区域中心({region_center[0]:.1f}, {region_center[1]:.1f})", end="") print(
f"轨迹路线: 区域中心({region_center[0]:.1f}, {region_center[1]:.1f})", end="")
current_pos = region_center current_pos = region_center
car_path = [] car_path = []
for j, task in enumerate(tasks, 1): for j, task in enumerate(tasks, 1):
current_pos = task['center'] current_pos = task['center']
car_path.append(current_pos) car_path.append(current_pos)
print(f" -> 任务{j}({current_pos[0]:.1f}, {current_pos[1]:.1f})", end="") print(
f" -> 任务{j}({current_pos[0]:.1f}, {current_pos[1]:.1f})", end="")
print(" -> 区域中心") print(" -> 区域中心")
car_paths[i] = car_path car_paths[i] = car_path