添加 DDPG、DPAC、Off-PAC 算法实现及连续动作空间训练代码

RL_Algothrithms/networks_cont.py

@@ -0,0 +1,32 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+# 连续动作策略网络 (Deterministic Actor)
+class ContActor(nn.Module):
+    def __init__(self, state_dim, action_dim, max_action):
+        super(ContActor, self).__init__()
+        self.fc1 = nn.Linear(state_dim, 128)
+        self.fc2 = nn.Linear(128, action_dim)
+        self.max_action = max_action
+        
+    def forward(self, state):
+        x = F.relu(self.fc1(state))
+        # 使用 tanh 将输出限制在 [-1, 1]，然后映射到物理实际的边界 (例如 [-2.0, 2.0])
+        action = torch.tanh(self.fc2(x)) * self.max_action
+        return action
+
+# 连续动作 Q 值网络 (用于评价 (State, Action) 对)
+class ContQCritic(nn.Module):
+    def __init__(self, state_dim, action_dim):
+        super(ContQCritic, self).__init__()
+        # 将状态和动作拼接后输入网络
+        self.fc1 = nn.Linear(state_dim + action_dim, 128)
+        self.fc2 = nn.Linear(128, 1)
+        
+    def forward(self, state, action):
+        # 拼接维度
+        xu = torch.cat([state, action], 1)
+        x = F.relu(self.fc1(xu))
+        q_value = self.fc2(x)
+        return q_value