This is a Pytorch implementation of KD-ST: Distillation Knowledge-Based Space-Time Data Prediction on Industrial IoT Edge Devices.
Currently, IIoT tends to offer significant efficiency and productivity gains to industrial operations. Compared to traditional IoT consumer-oriented devices, IIoT is large-scale and faces more life-threatening or high-risk situations due to system failures and downtime. IIoT imperatively requires energy-saving equipment and secure operation to overcome the existences of high energy consumption, limited battery capacity, substantial safety hazard, and complex data processing. Outstandingly, edge computing has emerged as a promising technology to support IIoT systems by allocating computation and storage resources at the network edge. The proposed KD-ST can efficiently slove the problem that the existing deep learning algorithms are too high complexity to deploy on edge devices, achieved a good tradeoff between execution cost and model accuracy in model inference.
- pytorch
- numpy
- matplotlib
- pandas
- math
You can get the dataset at https://www.kaggle.com/javi2270784/gas-sensor-array-temperature-modulation.
Our teacher model is stored in model <model.pt>.
You can run the following four files:
- LSTM-based student network <KD_ST_LSTM.py>
- LSTM-based transfer student network <Transfer_LSTM.py>
- 1DCNN-based student network <KD_ST_1DCNN.py>
- 1DCNN-based transfer student network <Transfer_1DCNN.py>