Distribution Network Topology Reconfiguration via Graph Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing distribution network reconfiguration methods face challenges with increasing computational complexity, inefficiency, high communication costs, and inability to securely reduce network losses as the number of nodes grows, particularly due to mutual current influences and non-independent topologies, and they fail to effectively address these issues.
Innovation Solution
The proposed solution involves a multi-stage topology reconfiguration method based on graph computing, which divides the network into grid regions, applies an improved optimal flow pattern and branch exchange method, and uses a synchronous alternating direction method of multipliers for iterative training to optimize switch states, ensuring minimal power loss and maintaining network safety and privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional centralized computing is used for distribution network reconfiguration, then the system can handle small-scale networks, but computational complexity and communication costs increase significantly as the number of nodes grows
Solution Approach 1:
The distribution network is divided into multiple grid regions, and each region is further segmented into basic units consisting of nodes and their connected branches. This hierarchical segmentation allows parallel processing of different regions and units, reducing the overall computational complexity from O(n) to O(n/p) where p is the number of parallel processors, while maintaining solution optimality through coordinated optimization across segments.
2Quantity of substance
If the number of nodes in the distribution network increases, then the network scale expands and can serve more users, but the workload grows and computational efficiency decreases
Solution Approach 1:
The patent transforms the traditional single-dimension sequential optimization approach into a multi-dimensional parallel computing framework by introducing spatial dimension (grid regions) and hierarchical dimension (basic units within regions). This allows the computational problem to be solved simultaneously across multiple dimensions, achieving near-linear scalability where computational efficiency decreases only logarithmically with the number of nodes rather than linearly.
3Measurement precision
If mathematical optimization methods are used for reconfiguration, then optimal solutions can be found quickly for convex problems, but the methods are not directly applicable to non-convex nonlinear problems and are overly dependent on initial values
Solution Approach 1:
The patent introduces a current transfer factor as an intermediary variable that linearizes the non-convex power flow equations. By transforming the original non-linear optimization problem into a linearized form using this intermediary, the method enables direct application of efficient linear programming techniques while maintaining accuracy, and eliminates dependence on initial values as the linearized system has a unique solution.
4Measurement precision
If heuristic methods are used for reconfiguration, then simple algorithms can achieve high-quality searches for optimal solutions, but as the number of nodes increases, the workload grows and computational efficiency decreases
Solution Approach 1:
The patent replaces the mechanical iterative search process of heuristic methods with an analytical linearized optimization model. Instead of repeatedly evaluating objective functions and gradient directions as in traditional heuristic approaches, the linearized model provides a direct computational path that yields optimal solutions in a single calculation step, reducing computational complexity from polynomial to linear time while maintaining solution quality.
5Measurement precision
If intelligent optimization algorithms are used, then effective solutions can be obtained for small-scale systems, but these algorithms struggle to converge to optimal solutions in large-scale node systems
Solution Approach 1:
The patent fundamentally changes the mathematical parameters of the optimization problem by transforming non-linear power flow equations into linearized form. This parameter transformation converts the optimization landscape from a complex non-convex surface with multiple local optima (where intelligent algorithms struggle to converge) into a linear surface with a unique global optimum, ensuring convergence performance that scales efficiently with system size.
6Device complexity
If distributed computing methods are used to reduce communication costs, then scalability improves, but user privacy protection becomes more challenging
Solution Approach 1:
The patent implements local quality by allowing each basic unit and grid region to perform optimization calculations using only local network data and parameters. User privacy information remains localized and is never transmitted to central controllers or other regions. The global optimal solution emerges from coordinated local optimizations, achieving both reduced communication costs and enhanced privacy protection through this distributed local-computation architecture.
Data Source
AI summary
Provided are a method for multi-stage topology reconfiguration of a distribution network based on graph computing and a system for multi-stage topology reconfiguration of a distribution network based on graph computing. The method includes providing a distribution network reconfiguration method based on the optimal flow pattern method and the branch exchange method; constructing a multi-stage reconfiguration model for the distribution network with an optimal power loss model; providing a multi-stage reconfiguration calculation method for the distribution network based on graph computing; and solving the optimal power loss model based on graph computing to obtain the distribution network reconfiguration technique.