Abstract:This paper proposes an input convex neural network (ICNN)-Assisted optimal power flow (OPF) in distribution networks. Instead of relying purely on optimization or machine learning, the ICNN-Assisted OPF is a combination of optimization and machine learning. It utilizes ICNN to learn the nonlinear but convex mapping from control variables to system state variables, followed by embedding into constrained optimization problems as convex constraints. Utilizing a designed ICNN structure, a fast primal-dual gradient method is developed to solve the ICNN-Assisted OPF, with the chain rule of deep learning applied to accelerate the algorithmic implementation. Convergence and optimally properties of the algorithm design are further established. Finally, different distribution network applications are discussed and proposed by means of the ICNN-Assisted OPF.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is an effective method for achieving optimal power flow (OPF) in distribution networks. Traditional methods rely on accurate distribution network models and use different optimization algorithms and techniques to solve them, such as Newton's method, interior - point method, etc. However, these methods face challenges such as difficulty in accurate modeling, high computational complexity, and difficulty in adapting to different network structures. With the development of the intelligence of modern distribution networks, data - driven and machine - learning methods have begun to be applied to solve the OPF problem to improve computational speed and efficiency. But these methods also have problems such as insufficient generalization ability and difficulty in ensuring that network constraints are satisfied.
For this reason, this paper proposes an optimal power flow (ICNN - Assisted OPF) strategy assisted by input convex neural network (ICNN). This strategy combines the advantages of optimization and machine learning, uses ICNN to learn the nonlinear but convex mapping from control variables to system state variables, and embeds it into the constrained optimization problem as a convex constraint. By designing a specific ICNN structure, a fast primal - dual gradient method is developed to solve the ICNN - Assisted OPF problem, and the chain rule of deep learning is used to accelerate the algorithm implementation. In addition, the paper also establishes the convergence and optimality properties of the algorithm and discusses the applicability in different distribution network applications.
Specifically, ICNN - Assisted OPF has the following three important advantages:
1. **Universality**: It is applicable to distribution networks with different structures. Whether it is an unbalanced or balanced network, whether it is a meshed or radial network, it can be processed in a unified manner without relying on a specific network structure.
2. **Reduced dependence on models**: By using ICNN to learn the mapping from control variables to network states and constructing global convex constraints, the dependence on accurate distribution network models is reduced, and at the same time, the credibility of data - driven decision - making is enhanced.
3. **Efficient algorithm design**: A neural network structure including weights and Softplus activation functions is designed, a fast primal - dual gradient algorithm is developed, the chain rule of deep learning is used to accelerate the algorithm implementation, performance guarantees are provided, and the convergence and optimality properties of the algorithm are established.
In conclusion, this paper aims to propose an efficient, reliable optimal power flow solution applicable to multiple distribution network structures by combining optimization and machine - learning methods.