A Comparative Study on Neural Network Architectures for DC Optimal Power Flow

Neural networks (NNs) are increasingly used to accelerate Optimal Power Flow (OPF) calculations, especially for large-scale or real-time applications. This paper presents a comparative analysis of three feedforward NN architectures: a) simple, b) Deep, and c) Wide. All architectures are evaluated on IEEE 5-bus, 57-bus, and 300-bus test systems for the DC OPF problem. We evaluate the architectures in terms of optimality gap and constraints violations. Reported violations include power balance, generator limits, and line flow limits.Results show that smaller architectures often generalize better and maintain feasibility more reliably, especially in small- to mid-scale systems. In contrast, deeper and wider networks can introduce overfitting, without proportional gains in large-scale cases. This study highlights the trade-off between model size and constraint satisfaction, advocating for leaner neural designs in physically constrained power system environments.