Cost-optimal photovoltaic soiling monitoring to improve energy yield: DOE-driven machine learning and MAUT with outdoor validation

• Field-validated, cost-optimal PV soiling monitoring is proposed to protect energy yield and conversion performance. • Low-cost and high-cost monitoring systems are benchmarked side-by-side on identical 100 W PV panels outdoors. • A DOE-driven workflow screens 128 sensing configurations and trains supervised regression to predict soiling-induced power loss. • MAUT-based decision layer selects the sensing design that maximizes accuracy per dollar for deployable energy management. • A three-sensor setup costing under USD 22 achieves near-benchmark performance (R² = 0.99, RMSE = 0.0022784). Soiling is a major source of photovoltaic (PV) energy-yield loss, reducing effective irradiance and causing voltage and current degradation that directly lowers conversion efficiency and increases operating cost. Despite extensive research on sensors and imaging for soiling detection, practitioners still lack structured guidance for selecting sensing configurations that are both accurate and economically deployable at scale. This study develops and validates a cost-optimal, AI-enabled soiling monitoring approach using an experimental–analytical workflow. Two monitoring systems, one low-cost and one high-cost, were deployed on identical 100 W PV panels for ten days under real outdoor conditions to collect synchronized electrical and environmental measurements. Natural soiling produced approximately 12% power loss. A two-level factorial Design of Experiments generated 128 virtual sensing configurations, and supervised regression models were trained to predict soiling-induced power loss. Multi-Attribute Utility Theory then integrated prediction accuracy and hardware cost to identify the optimal configuration. Results show that three low-cost sensors, a 0–25 V voltage sensor, an ACS712 30 A current sensor, and a GP2Y1010AU0F dust sensor, achieve R² = 0.99 with RMSE = 0.0022784 at a total hardware cost below USD 22. The findings provide a scalable pathway for PV soiling monitoring that supports energy management by optimizing accuracy per dollar with field-verified performance.