Search

8th World Conference on Photovoltaic Energy Conversion; 1482-1485, Photovoltaic (PV) inverters convert DC power to AC power. Inverters typically employ maximum power point tracking (MPPT) algorithms to maximize power production. Many modern inverters support several independent MPPT inputs to maximize energy production from arrays with different configurations or orientations. There is no consensus test procedure for evaluating the DC-to-AC conversion efficiency for multi-input inverters. Herein, we propose a test procedure based on the open-source System Validation Platform (SVP) software. We apply the procedure to a commercial inverter with six MPPT inputs to demonstrate that the resulting measurements can be used to fit a model that predicts inverter power at all conditions with reasonable accuracy.

38th European Photovoltaic Solar Energy Conference and Exhibition; 973-978, Failure diagnosis (detection and classification) in photovoltaic (PV) systems through data diagnostic algorithms is a fundamental task that ensures quality of operation and significantly improves the performance and reliability of operating PV systems. The scope of this work is to present the development of Failure Diagnosis Routines (FDRs) and Trend-based performance Losses Routines (TLRs) for diagnosing PV underperformance issues due to failure occurrences and performance loss events. The proposed routines complement the developed Data Quality Routines (DQRs) and operate exclusively on recorded electrical and meteorological measurements. The proposed routines were experimentally validated on a large-scale PV system installed in Larissa, Greece. The results demonstrated the effectiveness of the routines for detecting system underperformance issues and accurately classifying the detected issues into zero power production incidents, degradation, soiling and snow losses. The failure detection stage of the FDRs achieved a detection accuracy of 97.3% for zero power production incidents during daylight hours. A precision accuracy of 96.32% was obtained by the FDRs when classifying zero power production due to fault incidents, while the TLRs achieved 91.66% classification accuracy.

38th European Photovoltaic Solar Energy Conference and Exhibition; 1134-1137, The high penetration level of photovoltaic (PV) systems to the utility grid and the intermittent nature of the generated power, introduces new challenges for the stability of electricity grids. The scope of this study is to present a novel intraday (up to 5-hours ahead forecasting) PV power production forecasting algorithm that is fully data-driven and based on machine learning ensemble principles. The methodology followed to develop the forecasting ensemble comprised of a cluster of Bayesian neural networks (BNN) that were trained with different exogenous variables, lag structures and estimation window duration. Preliminary obtained results demonstrated that the resulting hour-ahead forecasting ensemble model showed accuracies beyond the state-of-the-art (

A. Livera, M. Theristis, G. Makrides, J. Surrerlueti, S. Ransome and G. E. Georghiou

37th European Photovoltaic Solar Energy Conference and Exhibition; 1256-1260, Accurate photovoltaic (PV) production forecasting is an important feature that can assist utilities and plant operators in the direction of energy management and dispatchability planning. Although numerous forecasting models have been reported in literature, the challenge of improved accuracy remains unsolved. In this work, a day-ahead PV power model utilising a hybrid approach is derived to feed into an Artificial Neural Network (ANN) and a linear regression model trained for PV power forecasting. The study focuses on improving the forecasting accuracy by employing machine learning and linear regression models that could record the behaviour of the PV system. The performance of the hybrid model was assessed against a single ANN model using a historical test set. The results showed that the hybrid model outperformed the single ANN model exhibiting a normalised mean square error (nRMSE) of 7.05%.

37th European Photovoltaic Solar Energy Conference and Exhibition; 1401-1404, Different data pipelines and statistical methods are applied to photovoltaic (PV) performance datasets to quantify the PV module degradation rate. Since the real value of degradation rate is unknown, a variety of unvalidated values has been reported in the literature. As such, the PV industry commonly treats this metric in an assumptive manner based on a statistically extracted range from the literature. However, the accuracy and uncertainty of degradation rate depends on a number of parameters including seasonality in respect to the local climatic conditions and also the response of a particular PV technology. In addition, the selection of data pipeline and statistical method may compound on the accuracy and uncertainty. In order to provide insights, a framework of bulk simulations of PV performance datasets using data from different climates is under development. Known degradation rates are emulated and large parametric studies are conducted in order to observe the convergence time on different PV module types based on several selection criteria such as performance metric, statistical method, etc. The preliminary results that are presented in this paper confirm that, indeed, climates and PV module types with typically lower seasonality can provide accurate degradation rate results in a shorter time period, compared to locations and PV module types that exhibit higher seasonality. As expected, the selection of data pipeline (e.g. metric, temperature correction, etc.) and statistical method also has a strong influence and therefore, introduces additional challenges.

37th European Photovoltaic Solar Energy Conference and Exhibition; 1352-1356, High-quality datasets are crucial for the performance and reliability analysis of photovoltaic (PV) systems. With respect to data integrity, invalid data are a common problem exhibited in PV monitoring systems. A data pipeline approach was recently introduced aiming to support reproducible results in PV performance. The methodology is expanded in this study by examining further outlier observations in respect to detection techniques, impact and treatment methods. The outlier detection results demonstrated that the standard boxplot rule yielded the highest detection rate of 95.3% (by taking a moving data window) at 40% of outlying data points and the effect of random outlying data points was mitigated by listwise deletion. The comparative analysis of outlying data treatment demonstrated that back-filling with the Sandia Array Performance Model (SAPM) yielded more accurate degradation rate (RD) estimates (absolute percentage error, APE, of up to 0.36% at 40% of outlying data) compared to filtering out the outlying data points (APE of up to 2.53% with listwise deletion).

35th European Photovoltaic Solar Energy Conference and Exhibition; 1700-1705, The increased penetration of photovoltaic (PV) systems introduce new challenges for the stability of electricity grids. In this scope, a machine learning technique utilising artificial neural networks (ANN) was implemented to forecast the hour-ahead (HA) PV power production without the utilisation of numerical weather prediction (NWP) models. Instead, historical PV operational and meteorological data-sets were used for the training and validation stages of the model to calculate HA PV power generation data-sets for time t + 1h while the model input parameters (weather measurements) were applied for time t. The best-performing model comprised of four input parameters (in-plane global irradiance (GI), ambient temperature (Tamb), elevation angle (AlS), azimuth angle (AzS)), a single hidden layer and 22 hidden neurons. Additionally, the results obtained over a test set period of 55 days demonstrated that accurate HA forecasts could be achieved without incorporating NWP since a daily-normalised root mean square error (nRMSE) of 3.63% was achieved. Finally, the model performed very well during clear sky days with a nRMSE close to 1% whereas 57% of all days demonstrated a nRMSE below 5%.

This paper summarizes the current status of the ALCHEMI project, a European collaborative undertaking under the Solar-Era.Net co-fund scheme. The project’s aim is to develop novel, low cost, HCPV modules that operate at ∼1000 suns, and demonstrate a DC module efficiency of >37% under Concentrator Standard Test Conditions (CSTC), and a manufacturing process that will achieve costs

36th European Photovoltaic Solar Energy Conference and Exhibition; 1087-1090, Cracks in wafer-based silicon solar cells are a well-known problem in the photovoltaic (PV) industry. Their formation is inevitable during either the manufacturing or the service life of a module and up to now it is not clear how to quantify their impact on PV performance. In order to improve the reliability and thus the lifetime of crystalline silicon (c-Si) PV modules, it is crucial to examine the environmental effects on the performance of modules and to also understand the exact causes of module degradation. Currently, commercial c-Si PV modules offer a 25-year warranty and their reliability and durability over this time period is considered as proven when a module successfully qualifies the tests according to IEC 61215. In practice though, there is currently no available parameter for the mechanical strength of cells and wafers, which guarantees the reliability of an individual module. In this study, an advanced experimental testing and a finite element analysis (FEA) model are conducted in order to identify the effect of clamping and frame on the durability of PV modules and hence to quantify the electromechanical degradation due to cell cracks. The above techniques could be used to optimize the material properties and frame structure for improved durability and reliability of PV modules.

36th European Photovoltaic Solar Energy Conference and Exhibition; 1272-1277, In this work, the prediction performance of the mechanistic performance model (MPM) and a machine learning Feed Forward Neural Network (FFNN), was evaluated using yearly datasets, containing sixty-minute average and instantaneous field measurements obtained from the outdoor test sites in Nicosia, Cyprus and in Arizona, USA, respectively. The model exhibiting the lowest prediction error was derived based on different model training conditions. The obtained results demonstrated that both models provide good predictive quality using both instantaneous and average measurements. The performance of the models was strongly dependent on the duration of the train set, since for a random 70:30 % train and test set split using the yearly dataset from the GI OTF, a mean absolute percentage error (MAPE) of 1.95 % and 1.55 % was obtained for the MPM and FFNN, respectively. Alternatively, for a random 30:30 % train and test set data partition, the MPM and the FFNN achieved a MAPE of 2.03 % and 1.67 %, respectively. Moreover, by applying a medium irradiance condition filter (GI>0.3 kW/m2) to the dataset during a random 70:30 % train and test set approach, the MPM and the FFNN achieved a MAPE of 1.77 % and 1.37 %, respectively, demonstrating that the predictive accuracy of the models was enhanced by data filtering. Additionally, the MPM and the FFNN achieved the lowest MAPE of 2.05 % and 2.11 % when the train set (10 % of the entire dataset) contained 80 % of clear and 20 % of variable and diffuse measurements (same amount of weather type measurements as the amount of the irradiance profile classes of the location). Finally, for accurate predictions a random train and test approach should be utilized, the training process should be performed using the greatest possible amount of train data samples and by simultaneously applying an irradiance condition filter.

36th European Photovoltaic Solar Energy Conference and Exhibition; 1433-1438, Accurate PV production forecasting is an important feature that can assist utilities and plant operators in the direction of energy management and dispatchability planning. In this work, intra-day (1 to 3 hour-ahead) solar irradiance forecasting utilising Support Vector Machines for Regression (SVR) is derived in order to feed to an Artificial Neural Network (ANN) trained for PV power generation forecasting (1 to 3 hours ahead). This study focused on the improvement of intra-day PV power forecasting through improved solar irradiance forecasting by leveraging data-driven machine learning models that could record the solar irradiance profile and the behaviour of the PV system. The bestperforming models comprised of 3 parameters for the solar irradiance forecasting in-plane global irradiance (GI), elevation angle () and azimuth angle (s)) and 4 parameter the PV power forecasting model. (GI, ambient temperature (Tamb), and s). In addition, the results obtained over the test set period demonstrated that the intra-day PV power forecasting demonstrated a daily-normalised root mean square error (nRMSE) of 3.52% to 7.84% (solar irradiance forecasting nRMSE was 2.93% to 6.52%) indicating that both models have recorded the behaviour of their respective parameters.

Energy yield measurement and radiation yield determination in the field of photovoltaics (PV) are subject to fast development regarding estimation uncertainties and error in prediction [1]. Both are determined mainly by constraints given by equipment development, calibration schemes and operation routines. Further, an increasing range of PV technologies is available on the market showing rather different spectral responsivities. These require precise PV device calibrations, either outdoor or indoor, with accurate measurement of the light-source. Under these boundary conditions accurate spectrally resolved solar irradiance measurements are gaining higher importance compared to recent years. Finally also PV energy yield estimations (predictions) may benefit from more accurate information on the solar spectrum. The International Spectroradiometer and Broadband intercomparison (ISRC) is evaluating measurement devices, measurement routine and equivalence in measurement results. Last year edition involved 9 scientific institutions and 5 commercial partners of 8 countries, testingmeasurement capabilities and best practices in spectrally resolved solar irradiance between 300 nm and 1700 nm. This work compares results and best practice approaches during the recent years of intercomparison. Capability of precision improvements in measurement as well as deviation in measurement approaches, instruments and institutes are highlighted. The analysis aims to conclude on effects of harmonization efforts, spreading of best-practice measurement routines and discussions on certain aspects such as temperature control or traceability of calibration.

33rd European Photovoltaic Solar Energy Conference and Exhibition; 1298-1301, This paper presents the collaboration initiative by the Universities of Cyprus and Jaén, Spain named as “DEGRADE-CPV”. This effort aims in quantifying the degradation rates and modes of different systems based on the concentrating photovoltaic (CPV) technology. Knowledge of degradation is crucial for the reduction of investment risk and also to increase the bankability. The methodology and experimental campaigns conducted at both locations are described and the results of the indoor characterisation using the Helios 3198 solar simulator are presented in this work. The preliminary outdoor characterisation highlights the importance of incorporating corrections on the quantification of the degradation rate.

33rd European Photovoltaic Solar Energy Conference and Exhibition; 1825-1830, Energy performance improvements of photovoltaic (PV) modules can be achieved by using innovative packaging materials, specifically designed to decrease the operating temperature and increase power output. This is important, especially since the aesthetic appeal of black-coloured PV modules is becoming a global market driving feature. The scope of this work is to analyse the performance of different crystalline silicon (c-Si) PV modules manufactured with different packaging components (backsheet types) capable of reducing the operating cell temperatures and in parallel, to assess their thermal behaviour and reliability under real operating conditions in a warm climate. The comparative analysis of the thermal behaviour, based on acquired cell temperature measurements over a five-year period, verified that backsheets can be designed to influence the cell temperature and enhance performance by operating at lower temperatures in some cases exceeding 10 °C. The systems with the backsheets that retained lower operating temperatures produced higher annual energy yield, under the warm conditions exposed due to reduced thermal losses of approximately 1 %. In particular, the annual energy yield results showed that the systems equipped with the white control and black colour thermal management backsheet produced consistently the highest annual energy yield over the evaluation period. Finally, the results of the indoor and outdoor degradation rate analysis showed that, over the five-year period, there was no significant difference in the estimated degradation rate amongst the installed systems, since the results are within the uncertainty range.

The spectral sensitivity of a concentrating triple junction (3J) solar cell has been investigated. The atmospheric parameters such as the air mass (AM), aerosol optical depth (AOD) and precipitable water (PW) change the distribution of the solar spectrum in a way that the spectral, electrical and thermal performance of a 3J solar cell is affected. In this paper, the influence of the spectral changes on the performance of each subcell and whole cell has been analysed. It has been shown that increasing the AM and AOD have a negative impact on the spectral and electrical performance of 3J solar cells while increasing the PW has a positive effect, although, to a lesser degree. A three-dimensional finite element analysis model is used to quantify the effect of each atmospheric parameter on the thermal performance for a range of heat transfer coefficients from the back-plate to the ambient air and also ambient temperature. It is shown that a heat transfer coefficient greater than 1300 W/(m(2) K) is required to keep the solar cell under 100 degrees C at all times. In order to get a more realistic assessment and also to investigate the effect of heat transfer coefficient on the annual energy yield, the methodology is applied for four US locations using data from a typical meteorological year (TMY3).

Photovoltaic (PV) devices are spectrally selective, and their performance is influenced by unavoidable spectral variations. In addition, multijunction-based concentrating photovoltaic (CPV) devices show a strong spectral dependence due to the series connection of various junctions with different absorption bands, and also due to the use of concentrator optics. In this work, the accuracy of a new set of analytical equations that quantify the spectral impact caused by the changes in air mass (AM), aerosol optical depth (AOD) and precipitable water (PW) is discussed. Four different CPV devices based on lattice-matched and metamorphic triple-junction solar cells and a poly(methyl methacrylate) (PMMA) and silicon-on-glass (SoG) Fresnel lenses are considered. A long-term outdoor experimental campaign was carried out at the Centre for Advanced Studies on Energy and Environment (CEAEMA) of the University of Jaén, Spain. Results show a high accuracy in the estimations of the spectral factor (SF), with an average mean absolute percentage error (MAPE) within 0.91% and a mean relative error (MRE) within −0.32%.

29th European Photovoltaic Solar Energy Conference and Exhibition; 3833-3836, The operation of a stand-alone photovoltaic (PV) system is compared with a hybrid PV/Wind for a 100 m2 household in Larnaca, Cyprus in respect to costs and Loss of Load Probability (LOLP). A model was developed to simulate the operation of both systems over a period of a year using real hourly meteorogical data and a hypothetical load demand profile. Optimum sizing combinations of the number of system components were selected in order to achieve a 2±0.1% of LOLP with the lowest possible cost. The most economic design in terms of capital and replacement cost and lowest unsatisfied energy was found to be the PV/Wind with 30.8% lower cost compared to a PV only system..

27th European Photovoltaic Solar Energy Conference and Exhibition; 260-265, A thermal model has been developed to predict the heat output of a PV cell, in order to examine the most efficient and cost effective cooling system for a 500x concentrating PV cell. The worst case scenario (i.e. zero forced convection from a flat surface) was selected in order to identify and measure the scale of the challenge. Different geometries and materials of heat sinks were then designed and tested for passively cooling purposes of the High Concentration Photovoltaic system. It is shown that passive cooling of a CPV system with CR of 500x is insufficient to maintain the cells below 80°C, especially for high ambient temperatures. Numerical analysis and simulations in MatLAB and COMSOL Multiphysics report on the level of cell’s surface temperature and convective heat transfer coefficient that is required for the PV cell to operate below safe temperature limits.

27th European Photovoltaic Solar Energy Conference and Exhibition; 4245-4248, A residential standalone photovoltaic system is evaluated in terms of its optimum configuration taking into consideration cost and reliability aspects. Given a specific electricity demand, the optimum combination of PV arrays and batteries is assessed using the Loss of Load Probability (LOLP) as a criterion. Both the sizing of the system and the reliability of its components are also taken into consideration for assessing the LOLP. The supply performance of the system is simulated based on hourly demand and meteorological data while the stochastic behaviour of the system, with regards to its reliability, is simulated by a Markov model. Literature based failure data are used to quantify the reliability model and, thus, estimate the mean operation and maintenance cost of the system.

Energy yield measurement and radiation yield determination in the field of photovoltaics (PV) are subject to fast development regarding estimation uncertainties and error in prediction [1]. Both are determined mainly by constraints given by equipment development, calibration schemes and operation routines. Further, an increasing range of PV technologies is available on the market showing rather different spectral responsivities. These require precise PV device calibrations, either outdoor or indoor, with accurate measurement of the light-source. Under these boundary conditions accurate spectrally resolved solar irradiance measurements are gaining higher importance compared to recent years. Finally also PV energy yield estimations (predictions) may benefit from more accurate information on the solar spectrum. The International Spectroradiometer and Broadband intercomparison (ISRC) is evaluating measurement devices, measurement routine and equivalence in measurement results. Last year edition involved 9 scientific institutions and 5 commercial partners of 8 countries, testingmeasurement capabilities and best practices in spectrally resolved solar irradiance between 300 nm and 1700 nm. This work compares results and best practice approaches during the recent years of intercomparison. Capability of precision improvements in measurement as well as deviation in measurement approaches, instruments and institutes are highlighted. The analysis aims to conclude on effects of harmonization efforts, spreading of best-practice measurement routines and discussions on certain aspects such as temperature control or traceability of calibration.

Energy yield measurement and radiation yield determination in the field of photovoltaics (PV) are subject to fast development regarding estimation uncertainties and error in prediction [1]. Both are determined mainly by constraints given by equipment development, calibration schemes and operation routines. Further, an increasing range of PV technologies is available on the market showing rather different spectral responsivities. These require precise PV device calibrations, either outdoor or indoor, with accurate measurement of the light-source. Under these boundary conditions accurate spectrally resolved solar irradiance measurements are gaining higher importance compared to recent years. Finally also PV energy yield estimations (predictions) may benefit from more accurate information on the solar spectrum. The International Spectroradiometer and Broadband intercomparison (ISRC) is evaluating measurement devices, measurement routine and equivalence in measurement results. Last year edition involved 9 scientific institutions and 5 commercial partners of 8 countries, testingmeasurement capabilities and best practices in spectrally resolved solar irradiance between 300 nm and 1700 nm. This work compares results and best practice approaches during the recent years of intercomparison. Capability of precision improvements in measurement as well as deviation in measurement approaches, instruments and institutes are highlighted. The analysis aims to conclude on effects of harmonization efforts, spreading of best-practice measurement routines and discussions on certain aspects such as temperature control or traceability of calibration.

The varying shape of the direct normal irradiance (DNI) spectrum is mainly determined by air mass (AM), aerosol optical depth (AOD) and precipitable water (PW). Unlike some previous studies that aimed at modelling the spectral impact on photovoltaics (PV), a recently published method takes these parameters into account when modelling spectral effects on concentrating PV. A short review of this method is provided initially in this paper. Then, this work presents the results of an empirical validation for a typical lattice-matched 3J GaInP/GaInAs/Ge solar cell during four specific days selected from a wider 3-month experimental campaign. During this period, spectral DNI measurements were recorded at 5-minute intervals and combined with the spectral response of the CPV solar cell considered to calculate measured values of the spectral factor (SF). Results show how predicted values of SF are in close agreement with measured ones as root mean square error (RMSE) values do not exceed 2% for all the days analysed. Further, negligible values of mean bias error (MBE) are obtained. The best results are obtained in days with moderate values of AOD and PW -RMSE around 0.5%- while modelled values of SF get worse -RMSE slightly less than 2%- in days with extreme values of such parameters. Last, the method investigated here yielded a value of RMSE of 0.8%, which is far below 2.3% obtained by applying the other methods for the whole 3-month period under study.