Rethinking Neural Width for Alternating Current Optimal Power Flow Proxies
For power grid operators, this provides a principled way to design minimal neural networks that are easier to formally verify, crucial for safety-critical operations.
The paper addresses the lack of systematic methods for determining neural network width in ACOPF proxies, introducing a Loss-Guided Neural Densification algorithm that achieves performance parity with baselines using up to ten times fewer neurons per layer.
Deep learning proxies for Alternating Current Optimal Power Flow (ACOPF) lack systematic methods for determining architectural size. This paper conducts a constructive thought experiment to answer a fundamental inquiry: how wide must a neural network be to almost accurately approximate the ACOPF manifold? We introduce a Loss-Guided Neural Densification (LG-ND) algorithm that incrementally discovers necessary capacity by expanding only when the current deep neural network topology fails to improve further. Empirical results across various IEEE systems show that LG-ND achieves performance parity with literature baselines using up to ten times fewer neurons per layer. Such architectural minimalism is critical for the formal verification required in safety-critical grid operations.