Add model SSTBAN

[hczs]

数据集使用：PEMSD8 数据集配置文件需修改info的data_col字段改为 traffic_flow，只取流量特征值

数据集参数：

{
  "batch_size": 4,
  "cache_dataset": false,
  "num_workers": 0,
  "pad_with_last_sample": true,
  "train_rate": 0.6,
  "eval_rate": 0.2,
  "scaler": "standard",
  "load_external": false,
  "normal_external": false,
  "ext_scaler": "none",
  "input_window": 36,
  "output_window": 36,
  "add_time_in_day": false,
  "add_day_in_week": false,
  "len_closeness": 2,
  "len_period": 1,
  "len_trend": 2,
  "interval_period": 1,
  "interval_trend": 7,

  "time_slice_size": 5,
  "num_of_vertices": 170,
  "num_his": 36,
  "num_pred": 36
}

模型训练参数：

{
  "max_epoch": 100,
  "learner": "adam",
  "lr_epsilon": 1e-8,
  "weight_decay": 0.0001,

  "step_size": 10,
  "lr_scheduler": "steplr",
  "lr_decay_ratio": 0.9,

  "use_early_stop": true,
  "patience": 5,

  "gpu": 1,
  "L": 3,
  "K": 16,
  "d": 8,
  "mode": "train",
  "learning_rate": 0.001,
  "in_channels": 1,
  "out_channels": 1,
  "T_miss_len": 12,
  "node_miss_rate": 0.5,
  "self_weight_dis": 0.8,
  "reference": 3,
  "bn_decay": 0.1
}

[Kazeya27]

• 性能与原论文差距过大
• 以后pr麻烦附上原论文对应实验结果

[hczs]

• 性能与原论文差距过大
• 以后pr麻烦附上原论文对应实验结果

参数已经更新到最上方，实验结果如下：	PEMSD8	Libcity	SSTBAN
MAE	18.75	18.45
RMSE	31.92	28.44
MAPE	0.29	0.18