/Novel-Seq2Seq-Module

Novel-Seq2Seq model

Primary LanguageJupyter Notebook

Novel-Seq2Seq

A simplified demo of the paper "A Novel Sequence-to-Sequence based Deep Learning Model for Multi-step Load Forecasting".

Environment requirement

The code is implemented on the Win10 system using Python 3.8. The libraries used are as follows

torch 1.10.1
numpy 1.20.2
pandas 1.2.4
matplotlib 3.3.4
alive_progress 1.6.2

Prepare_data:

Consist of two class, Datasets and DataPrepare.

The function of DataPrepare as follows:

  • Data normalization.
  • Tranform series data into input and target pairs, which can be uesd train supervised model.
  • Split sample into train datasets, valid datasets and test datasets.

The class of Datasets is a generator, which is inherit by torch.utils.data.Dataset

The parameters of DataPrepare as follows:

  1. datapath: 数据集文件路径
  2. dataflie: 数据集名称
  3. input_steps: [int] 样本的输入步数
  4. pred_horizion: [int] 样本的预测步数
  5. Split ratio: [Tuple[float]] 样本划分比例，依次为 训练集、验证集、测试集

After preprocess, the return value of Dataprepare is a tuple, which are train_ip, train_op, valid_ip, valid_op, test_ip, test_op and the shape of ip is [sample_num, input_steps, features], and the shape of op is [pred_horizion]