-
Notifications
You must be signed in to change notification settings - Fork 0
/
env_6CAV_cooperation_20230319_benchmarks.py
389 lines (273 loc) · 17.5 KB
/
env_6CAV_cooperation_20230319_benchmarks.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
# -*- coding: utf-8 -*-
"""
Created on Tue Jan 17 13:32:12 2023
@author: Kaige
"""
# for 20230131-190919 and 20230201-002841
import os
import matplotlib.pyplot as plt
import numpy as np
import scipy.io as sio
import random
import math
import pandas as pd
import sympy as sp
from gym import spaces
##########################################################################################################################
def cal_delta(a,b,c,d):
p = (3*a*c-b*b)/(3*a*a)
q = (27*a*a*d - 9*a*b*c + 2*b*b*b)/(27*a*a*a)
delta = (q/2)*(q/2)+(p/3)*(p/3)*(p/3)
return delta
##################################################################################
class Network():
def __init__(self):
#########################################################################################################################
self.infeasible_penalty = -10
##################################### DNN model parameters ###############################################################
self.delta_1 = 120268800/30 # computing demnad (in cycles) for feature extraction
self.delta_2 = 1000/30 # computing demnad (in cycles) for feature fusion
self.delta_3 = 9255020/30 # computing demnad (in cycles) for fast inference
self.delta_4 = 2.3000e+09/30 # computing demnad (in cycles) for full inference
self.rho = 0.3; # Average early exit ratio in the default DNN model
self.rho_tilde = 0.6; # Average early exit ratio in the data-fusion DNN model
self.w = 295936; # feature data size (in bit)
self.Delta = 0.1; # Delay requirement in s, i.e., 100ms
##
self.kappa = 1e-28
self.delta = self.delta_1 + self.delta_3 + (1-self.rho) * self.delta_4
self.delta_tilde = 2*self.delta_1 + self.delta_2 + self.delta_3 + (1-self.rho_tilde) * self.delta_4
self.delta_hat = self.delta_1 + self.delta_2 + self.delta_3 + (1-self.rho_tilde) * self.delta_4
##
self.w_new = self.w/1e6*500
self.delta_hat_new = self.delta_hat/1e9*500
self.Delta_new = self.Delta*500
self.varphi = np.sqrt(2*np.power(self.delta,3)/(np.power(self.delta_hat,2)*self.delta_tilde))
############################################### Communication parameters #################################################
self.noise_power = np.power(10, -104/10)/1000 # -104dBm
self.Transmit_power = np.power(10, 23/10)/1000 # 23dBm
self.Interference_constant = 0
self.bandwidth_basis = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
7,7,7,7,7,7,7,7,7,7,7,7,7,7,7] #*1000000
self.f_center = 6 # Ghz
self.episode_length = 75
self.switch_coeff = 0.4
############################################### optimal resource allocation ###################################
self.obj_opt_6CAV = np.load('KKT_opt/KKT_6CAV_opt_data.npy')
print(self.obj_opt_6CAV.shape)
self.obj_opt_5CAV = np.squeeze(sio.loadmat('KKT_opt/KKT_5CAV_opt_data.mat').get('KKT_obj_5CAV_all'))
print(self.obj_opt_5CAV.shape)
self.obj_opt_4CAV = np.squeeze(sio.loadmat('KKT_opt/KKT_4CAV_opt_data.mat').get('KKT_obj_4CAV'))
print(self.obj_opt_4CAV.shape)
self.obj_opt_3CAV = np.squeeze(sio.loadmat('KKT_opt/KKT_3CAV_opt_data_all.mat').get('KKT_obj_3CAV_all'))
print(self.obj_opt_3CAV.shape)
self.obj_opt_2CAV = np.squeeze(sio.loadmat('KKT_opt/KKT_2CAV_opt_data.mat').get('KKT_obj_2CAV'))
self.obj_opt_1CAV = np.squeeze(sio.loadmat('KKT_opt/KKT_1CAV_opt_data.mat').get('KKT_obj_1CAV'))
############################################### CAV environment ###############################################
self.CAV_pair_num = 6
self.n_agents = self.CAV_pair_num
self.n = self.n_agents
self.n_adversaries = 0
self.action_space = [spaces.Discrete(2) for i in range(self.n)]
# self.observation_space = [spaces.Box(5,) for i in range(self.n)]
self.observation_space = [(6,) for i in range(self.n)] # bandwidth, own workload, own distance, avg workload of others, average distance of others
self.distances = np.arange(6,36,6) # All candidate states for transmitter-receiver distances
# state transition prob. matrix for distance states
self.trans_matrix = np.array([[0.35, 0.30, 0.20, 0.10, 0.05],
[0.25, 0.30, 0.25, 0.15, 0.05],
[0.10, 0.25, 0.30, 0.25, 0.10],
[0.05, 0.15, 0.25, 0.30, 0.25],
[0.05, 0.10, 0.20, 0.30, 0.35]])
# self.trans_matrix = np.array([[0.9, 0.1, 0, 0, 0],
# [0.05, 0.9, 0.05, 0, 0],
# [0, 0.05, 0.9, 0.05, 0],
# [0, 0, 0.05, 0.9, 0.05],
# [0, 0, 0, 0.1, 0.9]])
self.workloads = np.array([4,5,6,7,8])
self.load_trans = np.array([[0.35, 0.30, 0.20, 0.10, 0.05],
[0.25, 0.30, 0.25, 0.15, 0.05],
[0.10, 0.25, 0.30, 0.25, 0.10],
[0.05, 0.15, 0.25, 0.30, 0.25],
[0.05, 0.10, 0.20, 0.30, 0.35]])
# self.workloads = np.array([4,5,6,7,8])
# self.load_trans = np.array([[0.55, 0.24, 0.12, 0.06, 0.03],
# [0.16, 0.575, 0.16, 0.07, 0.035],
# [0.07, 0.14, 0.58, 0.14, 0.07],
# [0.035, 0.07, 0.16, 0.575, 0.16],
# [0.03, 0.06, 0.12, 0.24, 0.55]])
# print(np.sum(self.load_trans,axis=1))
# self.load_trans = np.array([[0.9, 0.1, 0, 0, 0],
# [0.05, 0.9, 0.05, 0, 0],
# [0, 0.05, 0.9, 0.05, 0],
# [0, 0, 0.05, 0.9, 0.05],
# [0, 0, 0, 0.1, 0.9]])
self.O_init = 8
def reset(self):
self.step_count = 0
####################### Available bandwidth (depending on HDV status) #####################################################
band_basis = self.bandwidth_basis[self.step_count]
if band_basis == 7:
self.Bandwidth_available = np.random.default_rng().choice([6,7],1, [0.5, 0.5])
elif band_basis == 2:
self.Bandwidth_available = np.random.default_rng().choice([2,3],1, [0.5, 0.5])
else:
self.Bandwidth_available = np.random.default_rng().choice([band_basis-1, band_basis, band_basis+1], 1, [0.25, 0.5, 0.25])
Bandwidth_norm = (self.Bandwidth_available - 2)/5
########################### CAV Workload (number of objects for detection) ##################################################################
# Dynmaic workload: number of objects for detection by the DNN model
self.O_vehs_all = self.O_init * np.ones(self.CAV_pair_num)
O_norm = (self.O_vehs_all - np.array([4,4,4,4,4,4]))/4
############################# Distance and channel gain ###############################################
self.Distance_all = 6*np.ones(self.CAV_pair_num)
Distance_norm = (self.Distance_all - np.array([6,6,6,6,6,6]))/24
# Path_loss_dB_all = 32.4 + 20*np.log10(self.Distance_all) + 20*math.log10(self.f_center) # NR-V2X 37.885 highway case, d in meter, f_center in GhZ, Path_loss_dB in dB
# self.Channel_gain_all = 1/np.power(10, Path_loss_dB_all/10)
########################################## State ######################################################
self.curr_state = [0]*self.n_agents
self.pre_action = np.array([1]*self.n_agents)
# self.pre_action_BFoptimal = np.array([1]*self.n_agents)
# self.pre_action_random = np.array([1]*self.n_agents)
# print("self.pre_action_BFoptimal:",self.pre_action_BFoptimal)
for i_agent in range(0, self.n_agents):
self.curr_state[i_agent] = np.r_[Bandwidth_norm, O_norm[i_agent], Distance_norm[i_agent], (sum(O_norm)-O_norm[i_agent])/(self.n_agents-1),(sum(Distance_norm)-Distance_norm[i_agent])/(self.n_agents-1),self.pre_action[i_agent]]
# self.curr_state[i_agent] = np.r_[Bandwidth_norm, O_norm[i_agent], Distance_norm[i_agent], (sum(O_norm)-O_norm[i_agent])/(self.n_agents-1),(sum(Distance_norm)-Distance_norm[i_agent])/(self.n_agents-1)]
# print("self.curr_state:", self.curr_state)
state = self.curr_state
self.step_count = 1
return state
def step(self, curr_action):
# switch_coeff = 0
# switch_coeff = 1
########################################################################### MADDPG ######################################################################################################
actions_array = np.array(curr_action)
# print("actions_array:", actions_array)
actions = np.array([np.argmax(actions_array[0]), np.argmax(actions_array[1]), np.argmax(actions_array[2]),np.argmax(actions_array[3]),np.argmax(actions_array[4]),np.argmax(actions_array[5])])
# print("actions:", actions)
################################### Action processing #################################################
# Number of CAV pairs in cooperation mode
Activated_CAV_pair_num = sum(actions)
Activated_index = np.where(actions == 1)
# Activated_index = np.array(Activated_index)[0].tolist()
# print("Activated_index:", Activated_index)
################################### Given state and action ############################################
obj_opt = 0
if Activated_CAV_pair_num > 0:
O_vehs = self.O_vehs_all[Activated_index]
Distance = self.Distance_all[Activated_index]
# print("O_vehs:", O_vehs)
# print("Distance:", Distance)
if Activated_CAV_pair_num > 1:
sort_index = np.array(Distance).argsort()
# print("sort_index:", sort_index)
O_vehs_ordered = O_vehs[sort_index]
Distance_ordered = Distance[sort_index]
if Activated_CAV_pair_num == 6:
obj_opt = self.obj_opt_6CAV[int(self.Bandwidth_available-2), int(O_vehs_ordered[0]-4), int(O_vehs_ordered[1]-4), int(O_vehs_ordered[2]-4), int(O_vehs_ordered[3]-4), int(O_vehs_ordered[4]-4), int(O_vehs_ordered[5]-4), int(Distance_ordered[0]/6-1), int(Distance_ordered[1]/6-1), int(Distance_ordered[2]/6-1), int(Distance_ordered[3]/6-1), int(Distance_ordered[4]/6-1), int(Distance_ordered[5]/6-1)]
elif Activated_CAV_pair_num == 5:
obj_opt = self.obj_opt_5CAV[int(self.Bandwidth_available-2), int(O_vehs_ordered[0]-4), int(O_vehs_ordered[1]-4), int(O_vehs_ordered[2]-4), int(O_vehs_ordered[3]-4), int(O_vehs_ordered[4]-4), int(Distance_ordered[0]/6-1), int(Distance_ordered[1]/6-1), int(Distance_ordered[2]/6-1), int(Distance_ordered[3]/6-1), int(Distance_ordered[4]/6-1)]
elif Activated_CAV_pair_num == 4:
obj_opt = self.obj_opt_4CAV[int(self.Bandwidth_available-2), int(O_vehs_ordered[0]-4), int(O_vehs_ordered[1]-4), int(O_vehs_ordered[2]-4), int(O_vehs_ordered[3]-4), int(Distance_ordered[0]/6-1), int(Distance_ordered[1]/6-1), int(Distance_ordered[2]/6-1), int(Distance_ordered[3]/6-1)]
elif Activated_CAV_pair_num == 3:
obj_opt = self.obj_opt_3CAV[int(self.Bandwidth_available-2), int(O_vehs_ordered[0]-4), int(O_vehs_ordered[1]-4), int(O_vehs_ordered[2]-4), int(Distance_ordered[0]/3-1), int(Distance_ordered[1]/3-1), int(Distance_ordered[2]/3-1)]
elif Activated_CAV_pair_num == 2:
obj_opt = self.obj_opt_2CAV[int(self.Bandwidth_available-2), int(O_vehs_ordered[0]-4), int(O_vehs_ordered[1]-4), int(Distance_ordered[0]/3-1), int(Distance_ordered[1]/3-1)]
elif Activated_CAV_pair_num == 1:
obj_opt = self.obj_opt_1CAV[int(self.Bandwidth_available-2), int(O_vehs[0]-4), int(Distance[0]/3-1)]
####################################### Calculate the reward ##########################################
reward = 0
reward_modified = 0
if obj_opt == -1:
obj_modified = 0
actions_modified = np.array([0,0,0,0,0,0])
switch_cost = self.CAV_pair_num - np.count_nonzero(actions_modified == self.pre_action)
reward_modified = obj_modified - switch_cost * self.switch_coeff
reward = reward_modified + self.infeasible_penalty
else:
obj_modified = obj_opt
actions_modified = actions
switch_cost = self.CAV_pair_num - np.count_nonzero(actions_modified == self.pre_action)
reward = obj_opt - switch_cost * self.switch_coeff
reward_modified = reward
# print("reward_modified:",reward_modified)
# if obj_opt != -1:
# reward = obj_opt - switch_cost/10
# print("switch_cost:", switch_cost)
self.pre_action = actions_modified
# print("self.pre_action:", self.pre_action)
####################################### Get the next state ##########################################
####################### Available bandwidth (depending on HDV status) #####################################################
band_basis = self.bandwidth_basis[self.step_count]
if band_basis == 7:
self.Bandwidth_available = np.random.default_rng().choice([6,7],1, [0.5, 0.5])
elif band_basis == 2:
self.Bandwidth_available = np.random.default_rng().choice([2,3],1, [0.5, 0.5])
else:
self.Bandwidth_available = np.random.default_rng().choice([band_basis-1, band_basis, band_basis+1], 1, [0.25, 0.5, 0.25])
Bandwidth_norm = (self.Bandwidth_available - 2)/5
########################### CAV Workload (number of objects for detection) ##################################################################
O_next = np.zeros([self.n_agents])
for i_agent in range(0, self.n_agents):
p_O = self.load_trans[int(self.O_vehs_all[i_agent]-4),:]
O_next[i_agent] = np.random.default_rng().choice(self.workloads, 1, p=p_O)
self.O_vehs_all = O_next
O_norm = (self.O_vehs_all - np.array([4,4,4,4,4,4]))/4
############################# Distance and channel gain ###############################################
distance_next = np.zeros([self.n_agents])
for i_agent in range(0, self.n_agents):
p_D = self.trans_matrix[int(self.Distance_all[i_agent]/6-1),:]
distance_next[i_agent] = np.random.default_rng().choice(self.distances, 1, p=p_D)
self.Distance_all = distance_next
Distance_norm = (self.Distance_all - np.array([6,6,6,6,6,6]))/24
# print("self.O_vehs_all:", self.O_vehs_all)
# print("self.Distance_all:", self.Distance_all)
##################################### State ###########################################################
state_next = [0]*self.n_agents # This is a list
rew_n = [0]*self.n_agents
done_n = [0]*self.n_agents
obj_n = [0]*self.n_agents
switch_n = [0]*self.n_agents
self.step_count += 1
done = False
if self.step_count == self.episode_length:
done = True
# self.episode += 1
for i_agent in range(0, self.n_agents):
state_next[i_agent] = np.r_[Bandwidth_norm, O_norm[i_agent], Distance_norm[i_agent], (sum(O_norm)-O_norm[i_agent])/(self.n_agents-1),(sum(Distance_norm)-Distance_norm[i_agent])/(self.n_agents-1),self.pre_action[i_agent]]
# state_next[i_agent] = np.r_[Bandwidth_norm, O_norm[i_agent], Distance_norm[i_agent], (sum(O_norm)-O_norm[i_agent])/(self.n_agents-1),(sum(Distance_norm)-Distance_norm[i_agent])/(self.n_agents-1)]
rew_n[i_agent] = reward
done_n[i_agent] = done
obj_n[i_agent] = obj_modified
switch_n[i_agent] = switch_cost
self.curr_state = state_next # This is a list of one dimensional (4,) np arrays
return state_next, rew_n, done_n, obj_n, switch_n, reward_modified
# ###-------------------------------
# env = Network()
# x = env.reset()
# print("Initial state:", x)
# curr_action = np.random.randint(0,2,env.CAV_pair_num)
# print("actions",curr_action)
# w, y, z = env.step(curr_action)
# print("reward:", y)
# print("state_next:", w)
# print("done:", z)
# curr_action = np.random.randint(0,2,env.CAV_pair_num)
# print("actions",curr_action)
# w, y, z = env.step(curr_action)
# print("reward:", y)
# print("state_next:", w)
# print("done:", z)
# curr_action = np.random.randint(0,2,env.CAV_pair_num)
# print("actions",curr_action)
# w, y, z = env.step(curr_action)
# print("reward:", y)
# print("state_next:", w)
# print("done:", z)
# curr_action = np.random.randint(0,2,env.CAV_pair_num)
# print("actions",curr_action)
# w, y, z = env.step(curr_action)
# print("reward:", y)
# print("state_next:", w)
# print("done:", z)