A novel dynamic/adaptive K-nearest neighbor model for the prediction of solar photovoltaic systems’ performance

The inherent variability in solar Photovoltaic (PV) systems motivates developing dynamic and adaptive data-driven models capable of accurately predicting the systems' performance. A dynamic/adaptive K-Nearest Neighbor (K-NN) model to estimate the conversion efficiency of a solar PV system is proposed in this work. The proposed model estimates the conversion efficiency of a test pattern based on three steps: (i) identifying the similar historical operational conditions experienced by the system, (ii) assigning a weight to each condition based on its similarity to the test pattern at hand, and (iii) computing the weighted average conversion efficiency of the test pattern. Thus, the proposed model is adaptive and dynamic as the nearest neighbors and the associated weights vary from one test pattern to another. The proposed model is applied to a 7.98 kWp rooftop grid-connected solar PV system at the Hashemite University in Jordan. Its effectiveness is evaluated by resorting to the Mean Square Error (MSE), Accuracy, Coefficient of Determination (R2), and Adjusted R2 (R2adjusted), and comparing the results with the traditional K-NN and other techniques adequately developed in the same context, such as Artificial Neural Network (ANN), Extreme Learning Machine (ELM), and Multiple Linear Regression (MLR). Results show that the proposed model exhibited superior performance across all metrics, with a performance gain reaching up to 45.2% and 6.8% for the MSE and R2adjusted, respectively, compared to the base MLR. Further, the simplicity and computational efficiency of the proposed model validate its practical deployment in predicting PV system performance.