Add Model STAEformer

[hczs]

任务执行配置信息

数据集配置文件（STAEformerDataset.json）

{
  "batch_size": 16,
  "cache_dataset": true,
  "num_workers": 0,
  "pad_with_last_sample": true,
  "train_rate": 0.7,
  "eval_rate": 0.1,
  "scaler": "standard",
  "input_window": 12,
  "output_window": 12,

  "load_external": true,
  "normal_external": false,
  "ext_scaler": "none",
  "add_time_in_day": true,
  "add_day_in_week": true
}

模型配置文件（STAEformer.json）

• MERT_LA 数据集

  {
    "max_epoch": 200,
  
    "learner": "adam",
    "learning_rate": 0.001,
    "lr_epsilon": 1e-8,
    "weight_decay": 0.0003,
  
    "lr_decay": true,
    "lr_scheduler": "multisteplr",
    "lr_decay_ratio": 0.1,
    "steps": [20,30],
  
    "use_early_stop": true,
    "patience": 30,
  
    "steps_per_day": 288,
    "input_dim": 3,
    "output_dim": 1,
    "input_embedding_dim": 24,
    "tod_embedding_dim": 24,
    "dow_embedding_dim": 24,
    "spatial_embedding_dim": 0,
    "adaptive_embedding_dim": 80,
    "feed_forward_dim": 256,
    "num_heads": 4,
    "num_layers": 3,
    "dropout": 0.1
  }

• PEMS_BAY 数据集

  {
  "max_epoch": 300,
  
  "learner": "adam",
  "learning_rate": 0.001,
  "lr_epsilon": 1e-8,
  "weight_decay": 0.0001,
  
  "lr_decay": true,
  "lr_scheduler": "multisteplr",
  "lr_decay_ratio": 0.1,
  "steps": [10,30],
  
  "use_early_stop": true,
  "patience": 20,
  
  "steps_per_day": 288,
  "input_dim": 3,
  "output_dim": 1,
  "input_embedding_dim": 24,
  "tod_embedding_dim": 24,
  "dow_embedding_dim": 24,
  "spatial_embedding_dim": 0,
  "adaptive_embedding_dim": 80,
  "feed_forward_dim": 256,
  "num_heads": 4,
  "num_layers": 3,
  "dropout": 0.1
  }

任务执行结果

模型评价指标对比如下

• METR_LA	METR_LA	LibCity	STAEformer
  MAE	3.32	3.34
  RMSE	6.98	7.02
  MAPE	9.69%	9.70%

• PEMS_BAY	PEMS_BAY	LibCity	STAEformer
  MAE	1.86	1.88
  RMSE	4.31	4.34
  MAPE	4.35%	4.41%