Posted on Leave a comment

Transient stability preventive control based on GCN and DRL

Reference: 

Wang, T., & Tang, Y. (2023). Transient stability preventive control based on graph convolution neural network and transfer deep reinforcement learning. CSEE Journal of Power and Energy Systems. https://doi.org/10.17775/CSEEJPES.2022.05030 

Overview: 

Maintaining transient stability in the dynamic field of power systems continues to be a major concern. Conventional optimization techniques frequently encounter difficulties with convergence, and the training of artificial intelligence solutions may be sluggish. Introducing a novel method that revolutionizes transient-stability preventive control (TSPC) by using Transfer Deep Reinforcement Learning (DRL) and Graph Convolutional Neural Networks (GCNN). This novel approach commences by assessing the present power-flow situation using a transient stability assessor (TSA) that is based on GCNNs. After that, it pinpoints the major factors influencing stability and incorporates this knowledge into an improved Markov decision-making procedure. Through entropy-enhanced TD3 algorithms and knowledge transfer, the method dramatically increases learning efficiency. This approach, which has been validated through simulations on actual power grids using the IEEE 39-bus system, shows improved control effects 

Issues in the existing method: 

  1. Complexity of Modern Power Systems: Growing complexity as a result of power electronics and integration of renewable energy. 
  2. Non-convergence in Conventional Methods: Conventional optimization-based techniques include local optima and non-convergence issues.
  3. High Computational Complexity: Scalability issues prevent large-scale electricity grids from expanding. 
  4. Limited Efficiency of Existing Approaches: The effectiveness of earlier techniques in meeting the demands of real-time control is lacking. 
  5. Long Training Times in AI-based Solutions: Long training times are a problem for deep reinforcement learning (DRL). 
  6. Underutilization of Transfer Learning: The reach and application of current transfer learning techniques in power systems are constrained. 

Methodology: 

GCNN-Based Transient Stability Assessor (TSA): 

The main objective of this study section is to develop a TSA using Graph Convolutional Neural Networks (GCNN). The TSA aims to gain knowledge about the complex connection between transient stability and power-flow conditions in power systems. By precisely predicting the Transient Stability Index (TSI) while considering various input parameters such as load level, fault location, and generator output power, the TSA enhances understanding.  GCNN provides deep knowledge about the durability of the system after a fault and promotes efficient system information learning ,when applied as a graph structure in the power system.

 TSPC Based on Transfer DRL: 

The primary goal of this study section is to construct the Markov Decision Process (MDP) for Transient Stability Preventive Control (TSPC) based on Transfer Deep Reinforcement Learning (DRL). The MDP has state, action, policy, and incentive components to support TSPC decision-making. By connecting DRL-determined actions to variations in generator output power, stability criteria are met. The reward system is carefully designed to promote actions that increase stability, with assessments conducted using N-1 contingency scenarios. By combining GCNN-based TSA with Transfer DRL, the study delivers efficient and effective TSPC, resolving issues with computational complexity and training time while guaranteeing stability and security in power systems.

 

schematic diagram of TSPC based on GCNN and transfer based on the study by Wang, T., & Tang, Y. (2023) 

Conclusions: 

  1. Combined TSA and Transfer DRL: The method integrates TSA with GCNN and transfer DRL for improved control.
  2. Effective Power Flow Control: Demonstrated effective control of power flow in both standard and actual power grid scenarios.
  3. Enhanced Learning Efficiency: Transfer learning accelerates learning and improves effectiveness.
  4. Identification of Influential Generators: Method identifies generators with significant influence for targeted action.
  5. Future Research Directions: Future studies will consider simultaneous rotor angle and transient voltage instability and extend control to multiple regions.

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 *