"""评估训练好的模型并绘制速度时空图""" import os import yaml import numpy as np import matplotlib.pyplot as plt import matplotlib matplotlib.use("Agg") from sumo_edge_vsl_environment import SUMOEdgeVSLEnvironment from ppo_agent import PPOAgent from appo_sumo_agent import APPOSUMOAgent from dqn_agent import DQNAgent import traci def evaluate_model(model_type, checkpoint_dir, config): """评估单个模型""" env = SUMOEdgeVSLEnvironment(config) if model_type == "PPO": agent = PPOAgent( state_dim=env.state_dim, action_dims=[env.action_dim] * env.num_edges, hidden_layers=[256, 256, 128], device="cuda" ) agent.load(os.path.join(checkpoint_dir, "model_best.pt")) elif model_type == "APPO": agent = APPOSUMOAgent( state_dim=env.state_dim, action_dims=[env.action_dim] * env.num_edges, hidden_dim=128, device="cuda", total_episodes=500 ) agent.load(os.path.join(checkpoint_dir, "model_best.pt")) elif model_type == "DQN": agents = [] for i in range(env.num_edges): agent_i = DQNAgent( state_dim=env.state_dim, num_actions=5, hidden_dim=256, device="cuda" ) agent_i.load(os.path.join(checkpoint_dir, f"model_best_edge{i}.pt")) agents.append(agent_i) agent = agents # 收集所有车检器组位置（只包含有检测器的位置） detector_groups = [] for edge_id in env.control_edges: edge_info = env.parser.edge_info[edge_id] for pos in edge_info.detector_positions: pos_idx = edge_info.detector_positions.index(pos) # 检查该位置是否有检测器 has_detector = False for lane_idx in edge_info.traffic_lane_indices: if edge_info.detectors.get((lane_idx, pos_idx)): has_detector = True break if has_detector: detector_groups.append((edge_id, pos)) # 运行仿真 state = env.reset(seed=42) speed_data = [] density_data = [] action_data = [] brake_counts = [] speed_variances = [] throughputs = [] done = False step = 0 while not done: if model_type == "DQN": action = np.array([ag.select_action(state) for ag in agent]) else: action, _, _ = agent.select_action(state, deterministic=True) action_data.append(action.copy()) state, reward, done, info = env.step(action) if not done: step_speeds = [] step_densities = [] for edge_id, pos in detector_groups: edge_info = env.parser.edge_info[edge_id] pos_idx = edge_info.detector_positions.index(pos) # 获取该位置所有车道的速度和占有率并取平均 speeds = [] occupancies = [] for lane_idx in edge_info.traffic_lane_indices: det_id = edge_info.detectors.get((lane_idx, pos_idx)) if det_id: spd = traci.inductionloop.getLastIntervalMeanSpeed(det_id) occ = traci.inductionloop.getLastIntervalOccupancy(det_id) if spd >= 0: speeds.append(spd * 3.6) if occ >= 0: occupancies.append(occ) # 如果没有检测器数据，使用edge的实时速度作为备用 if speeds: avg_speed = np.mean(speeds) else: edge_speed = traci.edge.getLastStepMeanSpeed(edge_id) avg_speed = edge_speed * 3.6 if edge_speed >= 0 else 0 step_speeds.append(avg_speed) avg_occupancy = np.mean(occupancies) if occupancies else 0 step_densities.append(avg_occupancy) speed_data.append(step_speeds) density_data.append(step_densities) total_brakes = 0 all_speeds = [] for edge_id, edge_info in env.parser.edge_info.items(): for det_id in edge_info.detectors.values(): total_brakes += traci.inductionloop.getLastIntervalVehicleNumber(det_id) spd = traci.inductionloop.getLastIntervalMeanSpeed(det_id) if spd >= 0: all_speeds.append(spd * 3.6) brake_counts.append(total_brakes) speed_variances.append(np.var(all_speeds) if all_speeds else 0) throughputs.append(info["throughput"]) step += 1 env.close() return { "speed_data": np.array(speed_data), "density_data": np.array(density_data), "action_data": np.array(action_data), "brake_counts": brake_counts, "speed_variances": speed_variances, "throughputs": throughputs } def plot_results(results_dict, output_dir): """绘制对比图并保存CSV""" os.makedirs(output_dir, exist_ok=True) # 保存CSV数据 import pandas as pd for name, results in results_dict.items(): speed_df = pd.DataFrame(results["speed_data"]) speed_df.to_csv(os.path.join(output_dir, f"{name}_speed.csv"), index=False) density_df = pd.DataFrame(results["density_data"]) density_df.to_csv(os.path.join(output_dir, f"{name}_density.csv"), index=False) action_df = pd.DataFrame(results["action_data"]) action_df.to_csv(os.path.join(output_dir, f"{name}_action.csv"), index=False) metrics_df = pd.DataFrame({ "brake_counts": results["brake_counts"], "speed_variances": results["speed_variances"], "throughputs": results["throughputs"] }) metrics_df.to_csv(os.path.join(output_dir, f"{name}_metrics.csv"), index=False) # 速度时空图 - 2x2布局 fig, axes = plt.subplots(2, 2, figsize=(16, 12)) axes = axes.flatten() for idx, (name, results) in enumerate(results_dict.items()): im = axes[idx].imshow(results["speed_data"].T, aspect="auto", cmap="RdYlGn", vmin=0, vmax=120, origin="lower") axes[idx].set_title(f"{name} Speed Heatmap") axes[idx].set_xlabel("Time Step") axes[idx].set_ylabel("Detector Group Index") plt.colorbar(im, ax=axes[idx], label="Speed (km/h)") plt.tight_layout() plt.savefig(os.path.join(output_dir, "speed_heatmaps.png"), dpi=150) plt.close() # 密度时空图 - 2x2布局 fig, axes = plt.subplots(2, 2, figsize=(16, 12)) axes = axes.flatten() for idx, (name, results) in enumerate(results_dict.items()): im = axes[idx].imshow(results["density_data"].T, aspect="auto", cmap="YlOrRd", vmin=0, vmax=100, origin="lower") axes[idx].set_title(f"{name} Density Heatmap") axes[idx].set_xlabel("Time Step") axes[idx].set_ylabel("Detector Group Index") plt.colorbar(im, ax=axes[idx], label="Occupancy (%)") plt.tight_layout() plt.savefig(os.path.join(output_dir, "density_heatmaps.png"), dpi=150) plt.close() # 关键指标曲线 fig, axes = plt.subplots(2, 2, figsize=(15, 10)) for name, results in results_dict.items(): axes[0, 0].plot(results["brake_counts"], label=name, alpha=0.7) axes[0, 1].plot(results["speed_variances"], label=name, alpha=0.7) axes[1, 0].plot(results["throughputs"], label=name, alpha=0.7) axes[0, 0].set_title("Brake Counts Over Time") axes[0, 0].set_xlabel("Time Step") axes[0, 0].set_ylabel("Brake Count") axes[0, 0].legend() axes[0, 0].grid(True, alpha=0.3) axes[0, 1].set_title("Speed Variance Over Time") axes[0, 1].set_xlabel("Time Step") axes[0, 1].set_ylabel("Variance (km/h)²") axes[0, 1].legend() axes[0, 1].grid(True, alpha=0.3) axes[1, 0].set_title("Throughput Over Time") axes[1, 0].set_xlabel("Time Step") axes[1, 0].set_ylabel("Throughput (veh/h)") axes[1, 0].legend() axes[1, 0].grid(True, alpha=0.3) # 统计摘要 summary_text = "" for name, results in results_dict.items(): summary_text += f"{name}:\n" summary_text += f" Avg Brake: {np.mean(results['brake_counts']):.1f}\n" summary_text += f" Avg Variance: {np.mean(results['speed_variances']):.2f}\n" summary_text += f" Avg Throughput: {np.mean(results['throughputs']):.1f}\n\n" axes[1, 1].text(0.1, 0.5, summary_text, fontsize=10, family="monospace", verticalalignment="center", transform=axes[1, 1].transAxes) axes[1, 1].axis("off") plt.tight_layout() plt.savefig(os.path.join(output_dir, "metrics_comparison.png"), dpi=150) plt.close() def evaluate_no_control(config): """评估无管控基准""" env = SUMOEdgeVSLEnvironment(config) # 收集所有车检器组位置（只包含有检测器的位置） detector_groups = [] for edge_id in env.control_edges: edge_info = env.parser.edge_info[edge_id] for pos in edge_info.detector_positions: pos_idx = edge_info.detector_positions.index(pos) has_detector = False for lane_idx in edge_info.traffic_lane_indices: if edge_info.detectors.get((lane_idx, pos_idx)): has_detector = True break if has_detector: detector_groups.append((edge_id, pos)) state = env.reset(seed=42) speed_data = [] density_data = [] action_data = [] brake_counts = [] speed_variances = [] throughputs = [] done = False while not done: action = np.array([4] * env.num_edges) action_data.append(action.copy()) state, reward, done, info = env.step(action) if not done: step_speeds = [] step_densities = [] for edge_id, pos in detector_groups: edge_info = env.parser.edge_info[edge_id] pos_idx = edge_info.detector_positions.index(pos) speeds = [] occupancies = [] for lane_idx in edge_info.traffic_lane_indices: det_id = edge_info.detectors.get((lane_idx, pos_idx)) if det_id: spd = traci.inductionloop.getLastIntervalMeanSpeed(det_id) occ = traci.inductionloop.getLastIntervalOccupancy(det_id) if spd >= 0: speeds.append(spd * 3.6) if occ >= 0: occupancies.append(occ) if speeds: avg_speed = np.mean(speeds) else: edge_speed = traci.edge.getLastStepMeanSpeed(edge_id) avg_speed = edge_speed * 3.6 if edge_speed >= 0 else 0 step_speeds.append(avg_speed) avg_occupancy = np.mean(occupancies) if occupancies else 0 step_densities.append(avg_occupancy) speed_data.append(step_speeds) density_data.append(step_densities) total_brakes = 0 all_speeds = [] for edge_id, edge_info in env.parser.edge_info.items(): for det_id in edge_info.detectors.values(): total_brakes += traci.inductionloop.getLastIntervalVehicleNumber(det_id) spd = traci.inductionloop.getLastIntervalMeanSpeed(det_id) if spd >= 0: all_speeds.append(spd * 3.6) brake_counts.append(total_brakes) speed_variances.append(np.var(all_speeds) if all_speeds else 0) throughputs.append(info["throughput"]) env.close() return { "speed_data": np.array(speed_data), "density_data": np.array(density_data), "action_data": np.array(action_data), "brake_counts": brake_counts, "speed_variances": speed_variances, "throughputs": throughputs } if __name__ == "__main__": with open("config_sumo_vsl.yaml", "r", encoding="utf-8") as f: config = yaml.safe_load(f) models = { "PPO": "checkpoints_sumo_vsl/final", "APPO": "checkpoints_sumo_appo/final", "DQN": "checkpoints_sumo_dqn/final" # "PPO": "checkpoints_sumo_vsl/20260324_202539", # "APPO": "checkpoints_sumo_appo/20260324_202514", # "DQN": "checkpoints_sumo_dqn/20260324_202548" } results = {} for name, ckpt in models.items(): print(f"Evaluating {name}...") results[name] = evaluate_model(name, ckpt, config) print("Evaluating No Control...") results["No Control"] = evaluate_no_control(config) plot_results(results, "evaluation_results") print("Evaluation complete! Results saved to evaluation_results/")