XAI based Energy Platform Development

1. Deep learning platform

    Real-time analysis of power monitoring data streams is required to detect electrical equipment failure and response to that failure; failure to respond immediately could lead to major accidents such as blackout and fire.

Figure 1. XAI-based Energy Platform Architecture for Detecting Abnormalities of Power Facilities

  • For more accurate analysis on stream data, a platform development is required to apply deep learning train/inference stream. To solve the problem of processing large AMI data on a deep running basis, we study source technology for deep running-based processing platform. Furthermore, we develop online learning scheme for processing AMI data to accelerate train using in-memory processing which causes faster processing than hard disk-based file system.
  • we build a deep learning platform for processing stream data, which performs collecting stream by using Kafka and pre-processing the data stream before forwarding it into deep learning model which use Pytorch or TensorFlow.

2. Online learning

    Continual Learning or Lifelong Learning is a new training strategy based on replaying memory and allowing beneficial effects to previous tasks. To preserve old information, current continual learning scheme accumulates observed examples into limited buffer or repeatedly trains generative model. This idea of learning scheme is effective to reduce catastrophic forgetting which is deterioration in overall performance when training sequentially.

Figure 2. Continuously incoming unbounded data stream and Continual Learning with K-means selection

  • Through multi-functional utility function optimization for AMI learning acceleration, an adaptive incremental scheduling of the batch size and the number of iterative learning in a temporary mini-position is performed for Concept Drift.
  • The proposed scheme ensures queue stability through data queue management and data instance conversion for mini batch processing for unbound stream data.

  • To train multi-AMI data causing dynamic change of data distribution, we develop the hybrid deep learning scheduling scheme in online learning.
  • The proposed scheme is stably recognizing current data stream and preserves various skewed data distribution into historical memory and selectively trains deep learning model to ensure robustness against dynamic change of data distribution.