Hierarchical Grid Response for DER Uncertainty and Real-Time Restoration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional distribution service restoration methods face challenges in managing the dynamics and uncertainties of distributed energy resources (DERs) for optimal scheduling and control, particularly in large-scale systems with high computational complexity, limiting their feasibility for real-time applications during extreme events like hurricanes or earthquakes.
Innovation Solution
A reinforcement learning-based collaborative resilience grid response framework is introduced, featuring a hierarchical management structure with flexible multi-agent placement for distribution system control entities, reducing model complexity through node cell segmentation and observational data construction to generate control actions for controllable components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional centralized model-based optimization methods are used for distribution service restoration, then system-wide coordination and optimal scheduling can be achieved, but computational complexity and computational burden increase significantly with system size, making real-time application infeasible
Solution Approach 1:
The patent divides the distribution system into multiple feeder systems and further segments each feeder into node cells (groups of nodes). This hierarchical segmentation allows the restoration problem to be solved in a distributed manner across multiple agents, reducing the computational burden on any single agent while maintaining system-wide coordination through inter-agent communication.
Solution Approach 2:
The patent introduces a hierarchical dimension to the restoration framework, with a high-level coordinator managing inter-feeder coordination and low-level agents handling local node cell restoration. This dimensional transformation from a flat centralized approach to a hierarchical distributed approach enables real-time decision-making by reducing the problem size at each level.
2Measurement precision
If detailed system models are used for precise optimization in centralized model-based restoration, then accurate control can be achieved, but computational speed decreases and scalability is constrained by system size
Solution Approach 1:
Each agent in the patent performs partial optimization for its local node cell rather than solving the complete system-wide optimization problem. The low-level agents focus on local restoration actions while the high-level coordinator handles global coordination, allowing parallel computation that significantly improves computational speed while maintaining sufficient precision for real-time application.
3Adaptability or versatility
If a large number of controllable devices and DERs are integrated into the distribution system, then system functionality and resilience are improved, but modeling complexity and computational burden increase greatly
Solution Approach 1:
The patent groups multiple nodes into node cells and aggregates controllable devices within each cell. This merging reduces the number of individual elements that need to be modeled and controlled separately, simplifying the overall system model while preserving the functional capabilities of the integrated devices through coordinated control at the cell level.
Data Source
AI summary
Systems and methods described herein can involve reducing a feeder system model by node cell segmentation on feeder system according to system topology information and system operation characteristics to generate a node cell segmented distribution grid; constructing observational data from systemwide status information aggregated from nodes identified from the node cell segmented distribution grid to meet a reinforcement learning (RL) policy network input requirement; training an RL policy framework to generate control actions for controllable components of the system; and executing the RL policy framework to generate control actions for controllable nodes in the node cell segmented distribution grid.