Application of Bayesian regression method in performance assessment of PV modules

Threshold-based methods are widely applied for performance assessment of photovoltaic (PV) modules because of their unique advantages. Traditional threshold-based methods apply deterministic estimation models to calculate the residuals between the estimated and the measured PV power, and then compare the residuals to the predefined normal thresholds. When a residual exceeds the thresholds, the corresponding point is identified as an outlier. These methods have a limitation, i.e., they do not take the uncertainties in the model parameters into consideration. In this paper, a new PV performance assessment method is proposed to improve the traditional methods by using the Bayesian Regression (BR) method. The method consists of: (i) developing the BR model of PV efficiency, (ii) constructing confidence intervals based on the developed model, and (iii) using the constructed confidence intervals as the normal thresholds for performance assessment of PV modules. To verify the effectiveness of the proposed methodology, the methods based on the deterministic estimation models are implemented for comparison. Results show that the proposed method is greatly superior to the traditional methods when used to detect power losses of 10%.