AC Optimal Power Flow Using Dual DNN Voltage Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for solving the AC-OPF problem in electrical power systems are inefficient, particularly for large-scale systems, as they fail to achieve significant computational speedup while ensuring constraint satisfaction, especially in handling uncertainties from intermittent renewables and stochastic loads.
Innovation Solution
The method employs two separate deep neural networks (DNNs) to predict voltage magnitudes and phase angles independently, reducing computation complexity and enabling fast computation of active and reactive generations, with a post-processing step for adjusting voltages to satisfy inequality constraints, thereby ensuring system stability and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional physics-based solvers (e.g., MIPS) are used to solve the AC-OPF problem, then all operational constraints are considered and feasible solutions are obtained, but the computational speedup is limited to less than one order of magnitude
Solution Approach 1:
The patent segments the solution process into two independent DNNs: one predicting voltage magnitudes and another predicting voltage phase angles. This segmentation allows each network to focus on specific constraints, improving both computational efficiency and constraint satisfaction independently
Solution Approach 2:
The patent introduces an intermediary post-processing step that adjusts the DNN predictions to satisfy inequality constraints. This intermediary process ensures constraint satisfaction while maintaining the speed advantages of the DNN approach, achieving over one order of magnitude speedup compared to conventional solvers
2Device complexity
If a single DNN is used to predict both voltage magnitudes and phase angles, then the model is simpler, but the computation complexity increases
Solution Approach 1:
The patent divides the single DNN model into two separate DNNs, one for voltage magnitude prediction and another for phase angle prediction. This segmentation reduces the computational complexity of each individual network while maintaining the overall system's predictive capability
Solution Approach 2:
Each DNN is specialized to predict specific variables (magnitude or phase angle) with appropriate local constraints. This local specialization allows each network to be optimized for its specific function, reducing overall computation complexity while improving prediction accuracy
Data Source
AI summary
DeepOPF-V, a deep neural network (DNN)-based voltage-constrained approach for solving an alternating-current optimal power flow (AC-OPF) problem, is used to determine an operating point of an AC electrical power system. DeepOPE-V advantageously uses two DNNs to separately determine voltage magnitudes and voltage phase angles of buses in the system without cross-over operations between the two DNNs. A computation complexity is reduced when compared to using a single DNN for generating both the magnitudes and phase angles, allowing high computation efficiency achieved by DeepOPE-V. Remaining variables of the system are computed based on the determined magnitudes and phase angles. A solution for the operating condition is predicted. A fast post-processing (PP) method is developed to improve the feasibility of the predicted solution. The PP method uses linear adjustment to adjust the predicted solution to improve the solution feasibility while enabling fast execution of the PP method.