DDPG结合GCN做预测的代码

DDPG（Deep Deterministic Policy Gradient）结合 GCN（Graph Convolutional Network）可以用于处理图数据的预测任务。以下是一个简单的示例代码，展示了如何使用 PyTorch 实现 DDPG-GCN 模型：

import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
import random
from collections import deque
from torch_geometric.nn import GCNConv

# 创建 DDGP-GCN 的 Actor 类
class Actor(nn.Module):
    def __init__(self, state_dim, action_dim, hidden_dim):
        super(Actor, self).__init__()
        self.fc1 = nn.Linear(state_dim, hidden_dim)
        self.gcn = GCNConv(hidden_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, action_dim)

    def forward(self, state, adj):
        x = F.relu(self.fc1(state))
        x = self.gcn(x, adj)
        x = F.relu(x)
        x = self.fc2(x)
        x = torch.tanh(x)
        return x

# 创建 DDPG-GCN 的 Critic 类
class Critic(nn.Module):
    def __init__(self, state_dim, action_dim, hidden_dim):
        super(Critic, self).__init__()
        self.fc1 = nn.Linear(state_dim + action_dim, hidden_dim)
        self.gcn1 = GCNConv(hidden_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, 1)

    def forward(self, state, action, adj):
        x = torch.cat([state, action], 1)
        x = F.relu(self.fc1(x))
        x = self.gcn1(x, adj)
        x = F.relu(x)
        x = self.fc2(x)
        return x

# 创建 DDPG-GCN 的智能体类
class Agent:
    def __init__(self, state_dim, action_dim, hidden_dim, gamma=0.99, tau=1e-2, lr_actor=1e-3, lr_critic=1e-3, buffer_size=100000, batch_size=64):
        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

        self.actor = Actor(state_dim, action_dim, hidden_dim).to(self.device)
        self.actor_target = Actor(state_dim, action_dim, hidden_dim).to(self.device)
        self.critic = Critic(state_dim, action_dim, hidden_dim).to(self.device)
        self.critic_target = Critic(state_dim, action_dim, hidden_dim).to(self.device)

        self.actor_optimizer = torch.optim.Adam(self.actor.parameters(), lr=lr_actor)
        self.critic_optimizer = torch.optim.Adam(self.critic.parameters(), lr=lr_critic)

        self.buffer = deque(maxlen=buffer_size)
        self.batch_size = batch_size
        self.gamma = gamma
        self.tau = tau

    # 策略网络（Actor）选择动作
    def select_action(self, state, adj):
        state = torch.FloatTensor(state).to(self.device)
        adj = torch.FloatTensor(adj).to(self.device)
        self.actor.eval()
        with torch.no_grad():
            action = self.actor(state, adj).cpu().data.numpy()
        self.actor.train()
        return action

    # 存储（状态，动作，奖励，下一个状态）元组到缓存中
    def remember(self, state, action, reward, next_state, adj):
        state = torch.FloatTensor(state).to(self.device)
        action = torch.FloatTensor(action).to(self.device)
        reward = torch.FloatTensor([reward]).to(self.device)
        next_state = torch.FloatTensor(next_state).to(self.device)
        adj = torch.FloatTensor(adj).to(self.device)
        self.buffer.append((state, action, reward, next_state, adj))

    # 从缓存中随机抽样，进行训练
    def train(self):
        if len(self.buffer) < self.batch_size:
            return

        # 从缓存中随机抽样
        batch = random.sample(self.buffer, self.batch_size)
        state, action, reward, next_state, adj = zip(*batch)

        state = torch.cat(state)
        action = torch.cat(action)
        reward = torch.cat(reward)
        next_state = torch.cat(next_state)
        adj = torch.cat(adj)

        # 计算 Q 目标值
        next_action = self.actor_target(next_state, adj)
        q_target = reward + self.gamma * self.critic_target(next_state, next_action, adj).detach()
        q_target = q_target.to(self.device)

        # 更新 Critic 网络
        q_value = self.critic(state, action, adj)
        critic_loss = F.mse_loss(q_value, q_target)
        self.critic_optimizer.zero_grad()
        critic_loss.backward()
        self.critic_optimizer.step()

        # 更新 Actor 网络
        actor_loss = -self.critic(state, self.actor(state, adj), adj).mean()
        self.actor_optimizer.zero_grad()
        actor_loss.backward()
        self.actor_optimizer.step()

        # 更新目标网络（Target Network）
        for target_param, param in zip(self.actor_target.parameters(), self.actor.parameters()):
            target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data)

        for target_param, param in zip(self.critic_target.parameters(), self.critic.parameters()):
            target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data)

    # 保存模型
    def save(self, filename):
        torch.save({
            'actor_state_dict': self.actor.state_dict(),
            'critic_state_dict': self.critic.state_dict(),
            'actor_optimizer_state_dict': self.actor_optimizer.state_dict(),
            'critic_optimizer_state_dict': self.critic_optimizer.state_dict(),
            }, filename)

    # 加载模型
    def load(self, filename):
        checkpoint = torch.load(filename)
        self.actor.load_state_dict(checkpoint['actor_state_dict'])
        self.critic.load_state_dict(checkpoint['critic_state_dict'])
        self.actor_optimizer.load_state_dict(checkpoint['actor_optimizer_state_dict'])
        self.critic_optimizer.load_state_dict(checkpoint['critic_optimizer_state_dict'])

在上述代码中，我们首先定义了一个 GCN 网络，然后将其嵌入到 DDPG 智能体的 Actor 和 Critic 网络中。我们还定义了智能体的存储缓存、训练函数以及保存/加载函数。最后，我们可以使用以下代码来训练 DDPG-GCN 模型：

agent = Agent(state_dim, action_dim, hidden_dim)
for episode in range(num_episodes):
    state = env.reset()
    for step in range(num_steps):
        action = agent.select_action(state, adj)
        next_state, reward, done, _ = env.step(action)
        agent.remember(state, action, reward, next_state, adj)
        agent.train()
        if done:
            break
        state = next_state
    agent.save('ddpg_gcn.pt')

在训练过程中，我们首先使用 `select_action` 函数从智能体的 Actor 网络中选择动作。我们使用 `remember` 函数将（状态，动作，奖励，下一个状态）元组存储到智能体的缓存中。然后，我们使用 `train` 函数从缓存中随机抽样并进行训练。最后，我们使用 `save` 函数保存模型。