Systematic Literature Review on AI-Enhanced Dual-Axis Solar Tracking Systems: Techniques and Performance Analysis

Abstract

Recent advancements in artificial intelligence have significantly improved the performance of intelligent dual-axis solar-tracking systems, enabling more efficient photovoltaic (PV) energy harvesting under variable irradiance conditions, transient cloud cover, and mechanical uncertainties. This systematic literature review synthesizes 25 peer-reviewed studies (2018–2024) identified from Scopus and Web of Science using PRISMA 2020 procedures. We examine controller families (fuzzy logic, adaptive neuro-fuzzy interface system, deep reinforcement learning, and hybrid designs), sensing and actuation stacks (ephemeris, light sensors, inertial measurement, and computer-vision-based pose), and reported outcomes for tracking accuracy and energy gain. Across comparable conditions, AI-enabled controllers consistently reduce tracking error by ~10–35% and increase annual energy capture by ~8–25% relative to fixed-tilt or conventional rule-based/PID baselines, with the largest gains observed under partial shading and rapidly varying sky conditions. Validation is dominated by simulation, while prototype and hybrid (simulation plus field) evaluations—though fewer—provide stronger evidence of real-world robustness. Persistent challenges include computational cost on embedded hardware, sample-efficient learning and safety for field deployment, inconsistent reporting of metrics, and limited long-horizon testing. To address these gaps, we recommend (i) standardized benchmarking that reports tracking error, normalized energy yield, control latency and controller power, (ii) low-cost edge-AI implementations (model compression, quantization, and microcontroller-class deployment), and (iii) multi-season field trials with reproducible protocols across climates. The findings indicate a clear shift from static, hand-tuned control toward intelligent, adaptive methods. Hybrid designs emerge as a practical choice for deployment due to their combined stability and adaptability, whereas deep reinforcement learning shows state-of-the-art performance primarily in simulation and calls for careful simulation-to-real transfer. Overall, this review clarifies the evidence base and outlines priorities for fault-tolerant, low-cost, and real-time adaptive dual-axis solar tracking.