A-QAOA: IMPROVED CONVERGENCE VIA ADAPTIVE PARAMETER SCHEDULING IN PRECISION AGRICULTURE

Abstract

Optimization of precision agriculture, a combinatorial optimization problem that has both dependent and independent binary variables associated with selecting the one-to-one mapping of crop planting, irrigation, fertilization, and land management practices will scale poorly as farm size increases. The recently proposed standard Quantum Approximate Optimization Algorithm (QAOA) falls into the category of hybrid quantum classically based optimization but has slow convergence rates and has been shown to become trapped in local minima through two p independent, static optimization variables with no run-time guidance. To address this, we propose an Adaptive Quantum Approximate Optimization Algorithm (A-QAOA) that implements a feedback-driven, layer-wise parameter scheduling approach based upon two adaptive runtime metrics of cost function change (C) and solution variance (Var(C)). A-QAOA improves convergence rates without compromising the optimum solution by reducing the classical optimizer's search space from O(p) — representing 2p independent variables — to just 2 adaptive base variables γ₀ and β₀. A-QAOA will provide a faster convergence rate (72.5%) and shorter runtime (72.1%) compared to the standard QAOA when tested within Qiskit Aer using the COBYLA optimization algorithm across five independent graph instances (10 trials/graph for 50 total trials per algorithm). Specifically, the A-QAOA converged 72.5% faster compared with the standard QAOA (95.4→26.2 iterations; t=78.53, p<0.0001, Cohen's d=15.7) and exhibited a run-time reduction of 72.1% (30.0s→8.4s; t=28.05, p<0.0001, d=5.6).To our knowledge, this is the first work utilizing runtime feedback based adaptive QAOA parameter scheduling methods for resource allocation in the field of precision agriculture and for demonstrating benchmark quantum-enhanced precision agriculture site optimization using NISQ (Noisy Intermediate-Scale Quantum) technology.