-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmesh_DRL_Abstract.m
233 lines (200 loc) · 9.27 KB
/
mesh_DRL_Abstract.m
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
classdef (Abstract) mesh_DRL_Abstract < rl.env.MATLABEnvironment
properties
nNodes = 0;
nFaces = 0;
nCells = 0;
nFronts = 0;
Grid_stack = [];
BC_stack0 = [];
BC_stack = [];
boundaryFile = './grid/boundary_file.cas';
Coord0 = [];
Coord = [];
Xmax = 0;
Xmin = 0;
Ymax = 0;
Ymin = 0;
RANGE = [];
PenaltyForOutOfDomain = -10;
PenaltyForCross = -10;
RewardForFinishDomain = 100;
domainArea = 0;
sumArea = 0;
isPlot = 1;
standardlize = 1;
end
properties
% system state [L, B1, B2, R]'
StateIndex = zeros(4,1)
end
properties(Access = protected)
% Internal flag to store stale env that is finished
IsDone = false
end
methods
function this = mesh_DRL_Abstract(ActionInfo)
%% 状态为输入模板点的x,y坐标
ObservationInfo = rlNumericSpec([4 2]);
ObservationInfo.Name = 'mesh DRL States';
ObservationInfo.Description = 'L, B1, B2, R coordinates';
this = [email protected](ObservationInfo,ActionInfo);
initialize(this);
end
function initialize(this)
[this.BC_stack0, this.Coord0, ~, ~] = read_grid(this.boundaryFile, 0);
this.Xmax = max(this.Coord0(:,1));
this.Ymax = max(this.Coord0(:,2));
this.Xmin = min(this.Coord0(:,1));
this.Ymin = min(this.Coord0(:,2));
this.RANGE = [this.Xmin, this.Xmax, this.Ymin, this.Ymax];
this.domainArea = 400;
end
function initialState = reset(this)
this.Grid_stack = [];
this.BC_stack = this.BC_stack0;
this.Coord = this.Coord0;
if(this.isPlot)
PLOT(this.BC_stack, this.Coord);
end
this.nCells = 0;
this.nNodes = size(this.Coord,1);
this.nFaces = size(this.BC_stack,1);
this.nFronts = size(this.BC_stack,1);
this.sumArea = 0;
BC_stack_sorted = Sort_AFT(this.BC_stack);
node1_base = BC_stack_sorted(1,1);
node2_base = BC_stack_sorted(1,2);
PL = FindOneNeighborNode(node1_base, this.BC_stack, node2_base);
PR = FindOneNeighborNode(node2_base, this.BC_stack, node1_base);
%初始状态为模板点坐标，维度4x2
initialState = [this.Coord(PL,:);this.Coord(node1_base,:);this.Coord(node2_base,:);this.Coord(PR,:)];
if this.standardlize
initialState = Standardlize(initialState);
end
%同时将模板点的标号存起来，后面step时要用
this.StateIndex = [PL;node1_base;node2_base;PR];
end
function [observation,reward,isdone,loggedSignals] = step(this,action)
loggedSignals = [];
isdone = false;
PL = this.StateIndex(1);
node1_base = this.StateIndex(2);
node2_base = this.StateIndex(3);
PR = this.StateIndex(4);
base_length = DISTANCE(node1_base, node2_base, this.Coord(:,1), this.Coord(:,2));
%% 当前观察到的状态
observation =[this.Coord(PL,:);this.Coord(node1_base,:);this.Coord(node2_base,:);this.Coord(PR,:)];
if this.standardlize
observation = Standardlize(observation);
end
loggedSignals.State = observation;
if(this.isPlot)
PLOT_FRONT(this.BC_stack, this.Coord(:,1), this.Coord(:,2), 1);
end
%% 当前状态对应的动作
Pbest = action';% action就是输出的新点坐标
if this.standardlize
Pbest = AntiStandardlize(Pbest, this.Coord(node1_base,:), this.Coord(node2_base,:));
end
if(this.isPlot)
plot(Pbest(1),Pbest(2),'bo');
end
%% 计算动作对环境造成的改变，并评估给出奖励
% 如果Pbest离base front很远，则终止
Vec = Pbest - this.Coord(node1_base,:);
dist = sqrt(Vec*Vec');
if dist > 3 * base_length
isdone = true;
this.IsDone = isdone;
reward = this.PenaltyForOutOfDomain;
return;
end
%% 如果Pbest与现有点很靠近，则选择现有点
PSelect = [];
flag_best = 1;
index = PointClose2ExistedPoints(Pbest,this.Coord,node1_base,node2_base);
if index > 0
node_select = index;
PSelect = this.Coord(node_select,:);
flag_best = 0;
[~, row1] = FrontExist(node1_base,node_select, this.Grid_stack);
[~, row2] = FrontExist(node2_base,node_select, this.Grid_stack);
if(row1 > 0 || row2 > 0)
isdone = true;
this.IsDone = isdone;
reward = this.PenaltyForOutOfDomain;
return;
end
end
%% 如果Pbest在区域外，且不与现有点靠近，则终止
if (OutOfDomain(Pbest, this.RANGE) && index < 0)
isdone = true;
this.IsDone = isdone;
reward = this.PenaltyForOutOfDomain;
return;
end
if flag_best == 1 %如果选择了最佳点，则将最佳点坐标加入坐标点序列
PSelect = Pbest;
this.Coord(end+1,:) = Pbest;
node_select = size(this.Coord, 1);
end
%% 判断是否为左单元
flag_left_cell = IsLeftCell(node1_base, node2_base, node_select, this.Coord(:,1), this.Coord(:,2));
if flag_left_cell == 0
isdone = true;
this.IsDone = isdone;
reward = this.PenaltyForCross;
return;
end
%% 考虑周围的点进行相交性判断，若相交，则终止
nodeCandidate = NodeCandidate(this.BC_stack, node1_base, node2_base, this.Coord(:,1), this.Coord(:,2), PSelect, 3 * base_length);
frontCandidate = FrontCandidate(this.BC_stack, nodeCandidate);
flag_not_cross = IsNotCross(node1_base, node2_base, node_select, frontCandidate, this.BC_stack, this.Coord(:,1), this.Coord(:,2), 0);
if flag_not_cross == 0
isdone = true;
this.IsDone = isdone;
reward = this.PenaltyForCross;
return;
end
%% 若不相交，则更新数据
[this.BC_stack, this.nCells] = UpdateTriCells(this.BC_stack, this.nCells, this.Coord(:,1), this.Coord(:,2), node_select, flag_best);
num_of_new_fronts = size(this.BC_stack,1) - this.nFronts;
if(this.isPlot)
PLOT_NEW_FRONT(this.BC_stack, this.Coord(:,1), this.Coord(:,2), num_of_new_fronts, flag_best)
end
[this.BC_stack, this.Grid_stack, num_deleted_fronts] = DeleteInactiveFront(this.BC_stack, this.Grid_stack);
this.nFronts = this.nFronts + num_of_new_fronts - num_deleted_fronts;
this.nFaces = this.nFaces + num_of_new_fronts;
this.nNodes = size(this.Coord,1);
%% 计算奖励
quality = TriangleQuality(node1_base, node2_base, node_select, this.Coord(:,1), this.Coord(:,2));
if flag_best == 0
quality = quality / 0.8;
end
reward = power(quality, 1);
%% 考虑加上剩余面积判断，填满计算域则给奖励1
elementArea = AreaTriangle(this.Coord(node1_base,:), this.Coord(node2_base,:), this.Coord(node_select,:));
this.sumArea = this.sumArea + elementArea;
leftArea = this.domainArea - this.sumArea;
if(leftArea < 1e-9) || isempty(this.BC_stack)
isdone = true;
this.IsDone = isdone;
reward = this.RewardForFinishDomain;
return;
end
%% 计算下一步的状态
BC_stack_sorted = Sort_AFT(this.BC_stack);
node1_base = BC_stack_sorted(1,1);
node2_base = BC_stack_sorted(1,2);
PL = FindOneNeighborNode(node1_base, this.BC_stack, node2_base);
PR = FindOneNeighborNode(node2_base, this.BC_stack, node1_base);
observation =[this.Coord(PL,:);this.Coord(node1_base,:);this.Coord(node2_base,:);this.Coord(PR,:)];
if this.standardlize
observation = Standardlize(observation);
end
this.StateIndex = [PL; node1_base; node2_base; PR];
loggedSignals.State = observation;
this.IsDone = false;
end
end
end