Performance Analysis of Classical and Quantum Machine Learning Models for Electricity Load Scheduling

Abstract

When talking about rising demand for electricity, shifting to renewable energy sources, and prioritizing grid stability, one thing that is crucial is optimizing the scheduling of electricity load based on the demand. There are various classical machine learning (ML) and meta-heuristic optimization methods (Particle Swarm Optimization, Genetic Algorithms, Differential Evolution, Simulated Annealing) used to forecast load, and then schedule accordingly to minimize cost, reduce the peak load of system, and improve the overall efficiency. However, these methods struggle in handling large-scale combinatorial complexity and multiple constrained systems in real time. This paper presents a comparative study of classical ML-based optimization techniques and introduces a novel quantum-machine-learning approach using the Quantum Approximate Optimization Algorithm (QAOA) for electricity load scheduling. Initially, a survey was conducted and several classical techniques for optimization and forecasting were evaluated on realistic demand data. The paper highlights their strengths and drawbacks in terms of solution quality, computational time, and scalability. Then, the paper proposes and apply QAOA to the same scheduling problem, formulating load scheduling as a quadratic unconstrained binary optimization (QUBO) model suitable for quantum optimization. We assess the performance of QAOA against classical methods under identical constraints. Thus, Simulated Annealing produces a normalized cost of 0.00029 (after 37 seconds) whereas the cost found by Differential Evolution is 0.1314 and by the Genetic Algorithm 0.4379. This tends to confirm the validity of classical heuristics for smaller problems [16]. Our results show that while classical methods perform well for small to medium scale problems, the QAOA approach offers competitive or superior performance in terms of optimization quality and scalability for larger, more complex instances, with potential reductions in computation time under certain conditions. We conclude by discussing the practical challenges of implementing QAOA (e.g. noise, parameter tuning) and suggest directions for future research to advance quantum-assisted optimization in electricity scheduling.