Photovoltaic (PV) system performance can be degraded by different factors such as shadowing, soiling, aging of modules, and faults in the components. These factors can lead to a significant loss of energy production if not detected and corrected in time.
Performance monitoring can provide an accurate estimation of power generation of a PV array under normal operating conditions and produces an alarm when the PV array produces significantly less energy generation than it should under the given operating conditions. It would remind operators to take timely corrective actions in order to reduce the periods of abnormal performance degradation and maximize PV system lifetime and yield. Development of models used to estimate the expected PV array energy generation under normal operating conditions is the most essential step for the implementation of performance monitoring.
In our recent study titled “Performance assessment of photovoltaic modules based on daily energy generation estimation,” published in the journal Energy, we have proposed two solutions to develop models for more accurate estimation of PV output based on which PV performance monitoring has been significantly improved. The two proposed solutions are, respectively, a new data preprocessing method and development of sub-models in different weather condition.
Models used to estimate the expected PV power generations
The model for PV output estimation included in the process of PV performance monitoring should have not only high accuracy but also low complexity. To this end, linear regression models whose coefficients are fitted using the least squares method based on historical measurements are commonly selected.
Generally, linear regression models of PV efficiency can be developed based on the negative linear relationship between PV efficiency (i.e. PV power divided by area of PV arrays), and PV cell temperature. Therefore, the accuracy of PV power measurements directly affects the accuracy of the fitted regression models. Moreover, since measuring of PV cell temperature is seldom available or easily accessible in most PV plants, estimating PV cell temperature is used instead to develop the model. Generally, PV module temperature is estimated based on several major effect factors, i.e., ambient air temperature, the plane of array irradiance, mounting type of PV arrays, and wind speed. Accordingly, the accuracies of measurements of these factors have indirect impacts on the PV model accuracy.
Data preprocessing method aiming to be robust against erroneous measurements in normal operation conditions
Regarding both meteorological measurements and PV power data, several common data quality issues exist. For this purpose, data preprocessing, which detects and handles outliers, becomes one of the widely-used ways for enhancing the accuracy of energy estimation models and further improves PV performance monitoring.
The outliers in the measurements can be divided into two categories, i.e., i) outliers caused by faults or malfunctions, and indicating significant energy generation degradations; ii) erroneous measurements due to inaccurate instruments but in normal operation conditions. For examples of the second category, the maximum power point tracking (MPPT) capabilities of inverters are frequently poor to track the maximum power point (MPP) when the irradiance changes rapidly. Thus, considerable uncertainty is introduced to the measurements of PV output power. Besides, in relatively large PV plants, the local pyranometers are aligned and maintained regularly. Nevertheless, for residential or commercial PV systems, the pyranometers may not be aligned with the PV modules, and the analyst may be unaware of the inaccurate measurements. Consequently, irradiance data measured by the misaligned pyranometer deviates from the real incident solar irradiance on PV arrays.
Different from the existing data preprocessing methods eliminating all the detected outliers, we propose to keep the second type of outliers for model development. Theoretically speaking, this strategy can increase the robustness of the performance monitoring algorithm, as erroneous measurements in normal operation may not be detected as anomalies in further steps.
To this end, an outlier detection method comparing outputs of different inverters measured simultaneously is applied to be robust against the erroneous measurements in normal operation. It is noteworthy that the erroneous measurements of irradiance have an equal effect on the overall PV plant, not certain inverters. Moreover, the analysis shows that the difference in the effects of the fast-changing irradiance on the MPPT techniques of different inverters is negligible. Therefore, the erroneous measurements of irradiance and MPP power will not be detected by comparing the outputs of different inverters, and the outliers due to faults or malfunctions of a certain PV string will be discovered by the proposed method.
Sub-models developed in different weather conditions
Data analysis indicates that there exists a time difference between the peaks of the irradiance values measured by an azimuthal misaligned pyranometer and power of PV modules. The time difference is caused by the difference between the azimuth angles of the pyranometer and the PV array. Interestingly, this phenomenon occurs only on sunny days rather than cloudy days. This is because solar modules receive more direct solar radiation in clear days and more diffuse components during cloudy days. Direct solar radiation is easily affected by the changes of the azimuth angle, while diffuse solar radiation is barely affected by it. As a consequence, the linear relationship between solar irradiance and power of PV modules turns to be weakened on sunny days, not cloudy days. Readers can refer to the published paper for figures and a more detailed explanation on this issue.
On the basis of the above, the second strategy is proposed to develop sub-models in different weather conditions (i.e., sunny days and cloudy days) to better capture the relationship between the measurements of solar irradiance and PV power.
Performance monitoring of PV modules using the proposed strategies
Our method gives performance monitoring results with daily reporting intervals. We set normal operation thresholds on the residual between estimated and measured daily energy production to detect anomalies. When it exceeds the thresholds, an alarm of anomaly occurs.
When daily records on a new day come, we first apply a classification algorithm consists of feature extraction using Principal Component Analysis (PCA) and classification using Support Vector Machine method to classify the weather conditions. Then the potential daily energy generations in the given normal operating conditions are estimated using the corresponding sub-model of the classified weather. Finally, the residual between the expected and the real daily energy generation is compared to the defined thresholds.
The results of performance monitoring confirmed that when compared to the existing methods, the proposed strategies slightly increase the estimation accuracy and significantly reduce false alarms, i.e., an anomaly detected while the system operating normally.
These findings are described in the article entitled Performance assessment of photovoltaic modules based on daily energy generation estimation, recently published in the journal Energy. This work was conducted by Jing-Yi Wang and Zheng Qian from Beihang University, and Hamidreza Zareipour and David Wood from the University of Calgary.