Abstract:Alternative current optimal power flow (ACOPF) problems have been studied for over fifty years, and yet the development of an optimal algorithm to solve them remains a hot and challenging topic for researchers because of their nonlinear and nonconvex nature. A number of methods based on linearization and convexification have been proposed to solve to ACOPF problems, which result in near-optimal or local solutions, not optimal solutions. Nowadays, with the prevalence of machine learning, some researchers have begun to utilize this technology to solve ACOPF problems using the historical data generated by the grid operators. The present paper reviews the research on solving ACOPF problems using machine learning and neural networks and proposes future studies. This body of research is at the beginning of this area, and further exploration can be undertaken into the possibilities of solving ACOPF problems using machine learning.
What problem does this paper attempt to address?
### Problems Addressed by the Paper
This paper primarily explores how to use Machine Learning (ML) techniques to solve the Alternative Current Optimal Power Flow (ACOPF) problem. Specifically, the paper attempts to address the following issues:
1. **Limitations of Traditional Methods**:
- The ACOPF problem is nonlinear and non-convex, making it difficult for traditional linearization and convex optimization methods to find a global optimal solution.
- Direct Current Optimal Power Flow (DCOPF), although simple and suitable for large-scale networks, may cause faults and instability in actual power grids and is inadequate in addressing climate change.
2. **Application of Machine Learning**:
- Using historical data or simulation data to train Neural Networks (NN) to solve the ACOPF problem.
- Exploring different machine learning algorithms, such as Shallow Neural Network (SNN), Deep Neural Network (DNN), and Physics-Informed Neural Network (PINN), to improve computational efficiency and prediction accuracy.
3. **Specific Directions of Algorithms**:
- **Offline Training Models**: Training models offline using historical data to directly predict the mapping relationship between system load and optimal generation set.
- **Initialization Acceleration**: Using machine learning algorithms to predict the initial values for ACOPF solvers (e.g., MATPOWER Interior Point Solver, MIPS) to accelerate the convergence process.
- **Direct Solving**: Combining deep learning and other algorithms to directly solve the ACOPF problem while ensuring that the solutions meet physical constraints.
4. **Performance Evaluation**:
- Evaluating the computational efficiency, feasibility, and optimality of different machine learning models through simulation tests on the IEEE 118-bus system.
- Using specific metrics for evaluation, including speed-up factor, feasibility, optimality, and prediction error.
### Summary
This paper aims to use machine learning techniques to solve the ACOPF problem, overcoming the limitations of traditional methods, and proposes a series of neural network-based methods to improve computational efficiency and prediction accuracy. By comparing the performance of different algorithms in actual power grids, the effectiveness of these methods is validated.