Posted on Leave a comment

Proactive Routing Strategy in Smart Power Grids Using GRL

Reference:

Islam, M. A., Ismail, M., Atat, R., Boyaci, O., & Shannigrahi, S. (2023). Software-Defined Network-Based Proactive Routing Strategy in Smart Power Grids Using Graph Neural Network and Reinforcement Learning. E-Prime-Advances in Electrical Engineering, Electronics and Energy, 5, 100187.

https://doi.org/10.1016/j.prime.2023.100187

Overview:

Different QoS support is needed for periodic fixed scheduling (FS) and emergency-driven (ED) packets generated by sensors and actuators in smart power grids. Current routing algorithms lack QoS distinction and flexibility. Our suggested SDN proactive routing approach employs a graph-neural-network (GNN) model to predict traffic circumstances and prioritizes ED packets with distinct queues, guaranteeing precise congestion forecasts. Additionally, a reinforcement learning (RL) based method dynamically chooses the best routes and adjusts queue service rates according to actual and anticipated network states. We tested our framework on IEEE 14-bus and IEEE 39-bus systems. It significantly outperforms current benchmarks, improving network efficiency and QoS support. This novel strategy transforms the performance and reliability of smart grids.

Limitations in the existing methods:

  • The traditional centralized power grids have limited adaptability.
  • Analog control and manual data collection are unreliable.
  • The shift to smart grids is prompted by the integration of renewable energy sources and growing demand.
  • Smart grids incorporate sophisticated automation and monitoring.
  • Data packets are classified into two categories: event-driven (ED) and fixed scheduling (FS).
  • Both packet types require high dependability and low latency.
  • Conventional routing methods lack the dynamism needed for smart grids.
  • Modeling ED traffic accurately remains difficult.
  • Current AI-based routing ignores potential network changes.
  • The proposed proactive framework combines RL techniques with GNN prediction.

Methodology:

SDN Layers and System Model

The system model includes OpenFlow switches, communication links, and a control center in the cyber layer. It also features power nodes, field devices, and transmission lines in the physical layer. Using an SDN paradigm, it enables packet routing in smart grids. The SDN consists of three layers: intelligence, control, and data forwarding. For adaptive queue service rates and optimal routing, the intelligence layer employs RL agents and a GNN-based prediction model. By recording temporal and spatial information, the model leverages RL agents to identify sub-optimal paths. Consequently, it modifies congestion circumstances to forecast future network conditions effectively.

 

Illustration of system model under consideration from the study by Islam, M. A., Ismail, M., Atat, R., Boyaci, O., & Shannigrahi, S. (2023)

Routing Strategy and Implementation that is Proactive

A proactive routing method uses emergency occurrences from the DoE’s Electric Emergency and Disturbance Report and a practical dataset from the IEEE 14 and 39 bus test systems. It pre-configures queue service rates based on anticipated congestion events. By transforming sparse queue length data into a non-sparse timer indication signal, the GNN model is effectively trained to forecast future congestion conditions. This strategy seeks to reduce packet loss and latency, ensuring effective routing choices in the smart grid network.

Conclusion:

  • For smart power grids, a proactive SDN-based routing framework is suggested to enhance the routing of Emergency Data (ED) and Fault Sensing (FS) packets.
  • ED and FS packets are placed in different queues to minimize interference and maintain the intended Quality of Service (QoS).
  • By training a Graph Neural Network (GNN) model to anticipate future route congestion, forwarding rules can be updated ahead of time to reduce delays.
  • The framework adjusts queue service rates based on anticipated ED traffic quantities.
  • We evaluated IEEE 14-bus and IEEE 39-bus test systems using a dataset of actual emergency situations in American power networks.
  • The proactive routing architecture performed better than the Q-learning (QL) and Bellman-Ford benchmarks.
  • The findings revealed no packet loss for either ED or FS packets, less than 1% of packets with unfulfilled latency thresholds, and an average delay of under 60 milliseconds.
  • Delays result from the need to retrain GNN and Reinforcement Learning (RL) models for topological changes

Sakthivel R

I am a First-year M.Sc., AIML student enrolled at SASTRA University

I possess a high level of proficiency in a variety of programming languages and frameworks including Python. I have experience with cloud and database technologies, including SQL, Excel, Pandas, Scikit, TensorFlow, Git, and Power BI

Leave a Reply

Your email address will not be published. Required fields are marked *