Optimal pyranometer placement in bifacial PV plants on complex terrain

• Seven pyranometer placement algorithms are benchmarked in utility-scale PV plants. • Simulated irradiance includes shading, tracking geometry, and bifacial effects. • Multi-objective methods outperform geometric heuristics in accuracy and generalization. • Validation is performed using real SCADA data from three mountainous PV sites. • A hybrid two-stage strategy is proposed for scalable sensor deployment. Accurate placement of irradiance sensors is critical for performance monitoring in utility-scale photovoltaic (PV) plants, particularly those featuring single-axis trackers, bifacial modules, and non-uniform terrain. Installing pyranometers on every tracker row is infeasible, making optimal sensor selection a key design challenge. In this work, we benchmark seven pyranometer placement algorithms—including geometric heuristics, unsupervised clustering, metaheuristics, and multi-objective optimization using a detailed simulation framework that incorporates clear-sky irradiance modeling, terrain-induced shading, backtracking dynamics, and bifacial rear-side contribution. The methods are evaluated across three operational PV plants in Spain, ranging from 30 to 70 MWp, each characterized by complex topography. We assess each algorithm using multiple performance metrics: mean absolute error (MAE), mean relative error (MRE), temporal correlation (R 2 ), and inter-sensor redundancy. Results show that multi-objective algorithms, particularly those incorporating simulated irradiance, consistently outperform geometry-only approaches in both accuracy and robustness. Notably, a geometry-based multi-objective method achieves comparable performance. We find that simple dispersion heuristics fail to generalize under steep terrain, while simulated annealing offers a strong trade-off between accuracy and runtime. These findings support a hybrid sensor placement strategy combining fast geometric pre-selection with energy-based refinement. The proposed methodology is scalable and applicable to modern PV systems, offering a reproducible framework for data-driven sensor deployment in heterogeneous landscapes.