1 Star 1 Fork 3

Secure_DisOpt&Cont/基于交替方向乘子法的VRPTW问题分解模式

加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
克隆/下载
ADMM_based_framework.py 6.81 KB
一键复制 编辑 原始数据 按行查看 历史
Xinli_Shi 提交于 2023-05-11 04:46 . a commit
# coding=gb18030
__author__ = 'Yao'
import DSSR_ng_route
from define_class import *
import numpy as np
import copy
class ADMM_ITER():
def __init__(self):
self.number_of_iteration = 0
self.lower_bound_solution = []
self.upper_bound_solution = []
self.feasible_solution = []
self.lower_bound_value = 0
self.upper_bound_value = float('inf')
self.repeated_nodes = []
self.unserved_nodes = []
self.served_time_dict = {}
class ADMM():
def __init__(self, Iteration_limit):
self.priority_queue = []
self.incumbent = float('inf')
self.incumbent_node_id = -1
self.path_dict = {}
self.number_of_paths = 0
self.iteration_limit = Iteration_limit
self.time_limit_reached = False
self.obj_value = float('inf')
self.Lagrangian_price = {}
self.lower_bound = 0
self.upper_bound = float('inf')
self.admm_itr_list = []
def check_feasibility(self,served_time_dict):
repeated_nodes = []
unserved_nodes = []
for i in served_time_dict:
if served_time_dict[i] > 1:
repeated_nodes.append(i)
if served_time_dict[i] == 0:
unserved_nodes.append(i)
return repeated_nodes,unserved_nodes
def g_served_time_for_each_node(self):
served_time_dict = {}
return served_time_dict
def g_feasible_solution_based_on_upper_bound(self,current_solution):
pass
def update_Lagrangian_price(self,instanceVRP,served_time_dict):
# repeated_nodes,unserved_nodes = self.check_feasibility()
# served_time_dict = self.g_served_time_for_each_node()
for i in instanceVRP.g_node_dict:
if i != 1:
instanceVRP.g_node_dict[i].dual_price += (1-served_time_dict[i]) * instanceVRP.rho
instanceVRP.g_node_dict[i].ADMM_price = instanceVRP.g_node_dict[i].dual_price
def update_rho(self,itr,admm_itr,instanceVRP,served_time_dict):
admm_itr.repeated_nodes, admm_itr.unserved_nodes = self.check_feasibility(served_time_dict)
try:
previous_admm_itr = self.admm_itr_list[itr-1]
if np.square(len(admm_itr.repeated_nodes)+len(admm_itr.unserved_nodes)) > 0.25*np.square(len(previous_admm_itr.repeated_nodes)+len(previous_admm_itr.unserved_nodes)):
instanceVRP.rho=instanceVRP.rho * 2
if np.square(len(admm_itr.repeated_nodes)+len(admm_itr.unserved_nodes))==0:
instanceVRP.rho = instanceVRP.rho / 2
except:
pass
def update_served_time_dict_and_ADMM_price_by_removing_one_block(self,v,instanceVRP,served_time_dict,g_node_dict,path_list_for_last_iteration):
if path_list_for_last_iteration != []:
path_to_remove = path_list_for_last_iteration[v][0]
for i in path_to_remove[1:-1]:
served_time_dict[i] -= 1
for i in served_time_dict:
g_node_dict[i].ADMM_price = g_node_dict[i].dual_price+( 1-2*served_time_dict[i]) *instanceVRP.rho / 2
def update_served_time_dict_and_ADMM_price_by_adding_one_block(self,served_time_dict,path):
for i in path[1:-1]:
served_time_dict[i] += 1
def update_ADMM_price_during_one_iteration(self,v,instanceVRP,served_time_dict,g_node_dict,path_list_for_last_iteration,path):
path_to_add = path
for i in path_to_add[1:-1]:
served_time_dict[i] += 1
if path_list_for_last_iteration != []:
path_to_remove = path_list_for_last_iteration[v][0]
for i in served_time_dict:
if i in path_to_remove:
served_time_dict[i] -= 1
for i in served_time_dict:
g_node_dict[i].ADMM_price += ( 1-2*served_time_dict[i]) *instanceVRP.rho / 2
return served_time_dict
def calculate_upper_bound(self,upper_bound_solution):
upper_bound = 0
for p in upper_bound_solution:
upper_bound += p[2]
return upper_bound
# self.g_feasible_solution_based_on_upper_bound(upper_bound_solution)
def calculate_lower_bound(self,lower_bound_solution,g_node_dict):
path_cost = lower_bound_solution[0][1]
local_lower_bound = path_cost * number_of_used_vehicle + [i.dual_price for i in g_node_dict]
return local_lower_bound
def conductADMM(self,instanceVRP):
path = [[]]
thenetwork = Network(instanceVRP.g_node_dict,instanceVRP.g_link_dict)
thenetwork.g_node_Ngset(len(instanceVRP.g_node_dict),instanceVRP.g_node_dict,instanceVRP.g_link_dict)
served_time_dict = dict.fromkeys(range(2,len(instanceVRP.g_node_dict)+1), 0)
print(served_time_dict)
for itr in range(self.iteration_limit):
print("iteration=", (itr + 1))
admm_itr = ADMM_ITER()
try:
previous_admm_itr = self.admm_itr_list[-1]
except:
previous_admm_itr = ADMM_ITER()
# served_time_dict = {}
# for i in instanceVRP.g_node_dict:
# if i !=thenetwork.origin and i != thenetwork.destination:
# served_time_dict[i] = 0
admm_itr.number_of_iteration = itr
lb_path,lb_path_cost,lb_path_primal_cost = DSSR_ng_route.DSSR_ng_Route_Algorithm(thenetwork,instanceVRP.veh_cap,instanceVRP.g_node_dict,instanceVRP.g_link_dict,completion_bound_flag=1,ADMM_flag=1) #for lower bound
admm_itr.lower_bound_solution.append([lb_path[0],lb_path_cost[0]])
for v in range(instanceVRP.number_of_vehicles):
self.update_served_time_dict_and_ADMM_price_by_removing_one_block(v,instanceVRP,served_time_dict,instanceVRP.g_node_dict,previous_admm_itr.upper_bound_solution)
path, path_dual_cost,path_primal_cost = DSSR_ng_route.DSSR_ng_Route_Algorithm(thenetwork, instanceVRP.veh_cap, instanceVRP.g_node_dict,instanceVRP.g_link_dict, completion_bound_flag=1,ADMM_flag=1)
self.update_served_time_dict_and_ADMM_price_by_adding_one_block(served_time_dict,path[0])
admm_itr.upper_bound_solution.append([path[0], path_dual_cost[0],path_primal_cost[0]])
admm_itr.upper_bound_value = self.calculate_upper_bound(admm_itr.upper_bound_solution)
# admm_itr.lower_bound_value = self.calculate_lower_bound(admm_itr.lower_bound_solution,instanceVRP.g_node_dict)
self.update_Lagrangian_price(instanceVRP,served_time_dict)
# self.update_rho(itr,admm_itr,instanceVRP,served_time_dict)
admm_itr.served_time_dict = copy.copy(served_time_dict)
self.admm_itr_list.append(admm_itr)
print(admm_itr.upper_bound_solution)
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化
Python
1
https://gitee.com/Secure_DisOpt/admm-VRPTW.git
git@gitee.com:Secure_DisOpt/admm-VRPTW.git
Secure_DisOpt
admm-VRPTW
基于交替方向乘子法的VRPTW问题分解模式
master

搜索帮助