UniV-Forecasting

An open source univariate time series forecasting framework that provides following features:

A general framework intergrated with data preprocess, hyper-parameters setting, hyper-parameters tuning, model training, model evaluation, and experiment logging.
An easy user-replaced model coding paradigm compatible with both statistical, stochastic, and training models.
Ready-to-use forecasting models, supported with both GPU acceleration or CPU only.

Main Dependence

python >= 3.6
pytorch = 1.9.1
CUDA (as required as pytorch, if using GPU)
ray = 1.6.0 (as requried by the specific optimizaiton algorithm, if using TaskTuner)
scikit-learn = 1.0.2

The training models we implemented are referred to these papers.

Model	Paper
DeepAR	DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks
ConvRNN	Autoregressive Convolutional Recurrent Neural Network for Univariate and Multivariate Time
RNN (Elman, GRU, LSTM)	Recurrent neural networks for time series forecasting: current status and future directions
CNN	Convolutional neural networks for energy time series forecasting
MLP	PSO-MISMO modeling strategy for MultiStep-ahead time series prediction

The stochastic models we implemented are referred to these papers.

Model	Paper
RVFL	A review on neural networks with random weights
IELM	Extreme learning machine: theory and applications
SCN	Stochastic configuration networks: fundamentals and algorithms
ESN	Optimization and applications of echo state networks with leaky-integrator neurons
GESN	Growing echo-state network with multiple subreservoirs
DESN	Design of deep echo state networks
PSO-GESN	PSO-based growing echo state network

Our proposed models are corresponding to these papers.

Model	Paper
MSVR	Multi-step-ahead time series prediction using multiple-output support vector regression
ESM-CNN	Error-feedback stochastic modeling strategy for time series forecasting with convolutional neural networks
ETO-SDNN	Growing stochastic deep neural network for time series forecasting with error-feedback triple-phase optimization

This framework is created by Xinze Zhang, Qi Sima, and Siyue Yang, supervised by Prof. Yukun Bao, in the school of Management, Huazhong university of Science and Technology (HUST).

Notice

The DeepAR provided in this repository is modified based on the work of TimeSeries. Yunkai Zhang, Qiao Jianga, and Xueying Ma are original authors of TimeSeries.
The ConvRNN provided in this repository is modified based on the work of ConvRNN. KurochkinAlexey, Fess13 are original authors of ConvRNN.
The PSO-GESN provided in this repository is modified based on the source code created by Qi Sima.