Identifying local PV shading in urban areas based on AC power and regional irradiance measurement

As the price of photovoltaic applications is decreasing, more and more PV is installed in an urban environment. Driven by the need for aesthetics and integration in building constructions, new BIPV products are developed, which opens the possibility to extend the usable area to almost any built surface. As a consequence of this, in the new era of BIPV applications, it is only natural that PV is installed on a surface that is regularly shaded. Furthermore, because of the increased capacity, it is expected, that the dependency of future buildings on on-site electricity generation will increase. These factors drive the need for accurate performance monitoring and fault detection methods, of which shade detection of PV applications is an essential element. A new method is developed, that relies on locally measured AC power and regionally measured irradiance data. By calculating the apparent Performance Ratio (aPR) and applying machine-learning algorithms on the measured AC power time series, locally shaded time periods can be identified, without local irradiance measurement. The method consists of 5 steps: Step1. Creating analemma graph of aPR  Step2. Eliminate data points, with cloud-shading.  Step3. Binarizing the remaining measurement points with an aPR-treshold.  Step4. Train Support Vector Machine (SVM) with the previously binarized dataset.  Step5. Use trained SVM to perform a soft-margin nonlinear classification of all data points recorded in the first step. With this method it is possible to distinguish between local and cloud-shading and - in case of a mono-pane installation - to plot the shading contour of the nearby objects, which is useful input for fault detection and monitoring of PV installations in an urban area. The next step is the validation of the method by on-site shading measurements