【PVPMC】Fraunhofer-Bor-光伏系统先进性评价的智能数据质量控制
© Fraunhofer ISE Intelligent Data Quality Control for Evaluation of PV Monitoring Data FRAUNHOFER INSTITUTE FOR SOLAR ENERGY SYSTEMS ISE Jefferson Bor 薄中南, Elke Lorenz, Anna Dittmann, Christian Braun, Nicolas Holland, Steffen Karalus, Wiebke Herzberg, Wolfgang Heydenreich Fraunhofer Institute for Solar Energy Systems PVPMC modelling workshop Kunshan 10.12.2019 www.ise.fraunhofer.de © Fraunhofer ISE AGENDA Motivation Our Data Set Automated Orientation Detection Automated Shadow Detection © Fraunhofer ISE The Initiative of Intelligent Data Quality Control Remote Inspection of monitoring station Filtering invalid data Imperfectness of initial Installation Changes during operation due to: Ground situation Strong wind, snow Incautious movement, e.g. cleaning Clamping loose … Threshold values are not constant Some effects has complex correlation or no correlation, e.g. Shadow Frost, snow Curtailment … © Fraunhofer ISE The Initiative of Intelligent Data Quality Control Remote Inspection of monitoring station Filtering invalid data Imperfectness of initial Installation Changes during operation due to: Ground situation Strong wind, snow Incautious movement, e.g. cleaning Clamping loose … Threshold values are not constant Some effects has complex correlation or no correlation, e.g. Shadow Snow Curtailment … We need automated detection for the collected data from monitoring and PV systems © Fraunhofer ISE Our Data Set for the Concept Development PV Live measurement station 3 silicon sensors (tilted) Tilt 25° East/ South/ West - simulated „standard PV plant“ Pyranometer (horizontal) high accuracy/straight-forward combination with satellite-based irradiance values Real-time transmission every minute fast access to data © Fraunhofer ISE PV Live Data Ground-measurements and satellite-based irradiance data Ground-measurements 40 stations in Baden Württemberg* 1min resolution, Satellite-based irradiance data High spatial resolution: ~1km x 2km Time resolution: 15min *~36 km2, like Hainan Province © Fraunhofer ISE AGENDA Motivation Our Data Set Automated Orientation Detection Automated Shadow Detection © Fraunhofer ISE POA Irradiance Different azimuth angles of 3 Si sensors Tilt angle 25°, East/South/West Summer West Clear sky East South Measurement Station, PV-Live Station Freiburg © Fraunhofer ISE Detection of Orientation From GHI to POA irradiance Source: https://ecosmartsun.com PV System Performance: GHI to POA Global Horizontal Irradiance (GHI) Plane of Array (POA) © Fraunhofer ISE Flowchart showing how to model POA irradiance from measured GHI. Figure from „Evaluation of Global Horizontal Irradiance to Plane of Array Irradiance Models at Locations across the United States“ by Matthew Lave, Member, IEEE, William Hayes, Andrew Pohl, and Clifford W. Hansen. Model Chain of Orientation Detection Decomposition Models + Transposition Models © Fraunhofer ISE Let 𝑀𝑀 𝑡𝑡 𝑡𝑡 𝑡𝑡𝑡𝑡 , 𝑎𝑎 𝑎𝑎 𝑡𝑡 𝑎𝑎 𝑎𝑎𝑡𝑡 𝑎 : 𝐺𝐺 𝐺𝐺𝐺𝐺 → 𝑃𝑃𝑃𝑃 𝑃𝑃 be the model chain that maps GHI to POA. Find Parameters through optimization: 𝑚𝑚 𝑚𝑚𝑚𝑚 𝑡𝑡 𝑡𝑡 𝑡𝑡𝑡𝑡 , 𝑎𝑎 𝑎𝑎 𝑡𝑡 𝑎𝑎 𝑎𝑎𝑡𝑡 𝑎 | 𝑀𝑀 𝑡𝑡 𝑡𝑡 𝑡𝑡𝑡𝑡 , 𝑎𝑎 𝑎𝑎 𝑡𝑡 𝑎𝑎 𝑎𝑎𝑡𝑡 𝑎 𝐺𝐺 𝐺𝐺𝐺𝐺 − 𝑃𝑃𝑃𝑃 𝑃𝑃 | Optimization computed with least square minimization Analysis here Optimization for the three senors together to find azimuth ( tilt 25° fixed here) We aim to find out Influence of dataset (weather, season, number of days) Model Chain of Orientation Detection Find optimized fit of tilt/azimuth angles © Fraunhofer ISE Choice of Data Set for Detection of Orientation Influence of weather conditions Clear sky day : Great impact from orientation Cloudy day : little impact from orientation Clear sky days are suitable for detecting orientation Irradiance [W/m²] © Fraunhofer ISE For May, 2018 Fit result of everyday Varying weathers cause scattering of result Choice of Clearsky days are important Opt. Tilt angle 25°, East Si Sensor Choice of Data Set for Detection of Orientation Influence of weather conditions © Fraunhofer ISE Weather (30 days mean value): Use data of all weather situations vs. Use only clear sky days (detection of clear sky day is done separately) Amount of Sensor (30 days mean value): Use all three sensor data together vs. Use only one sensor data Choice of Data Set for Detection of Orientation Comparison of weather and amount of sensor © Fraunhofer ISE Results of Comparison of Orientation Detection All Weather Conditions vs. Only Clear Sky Period: 18.May - 04.Sep.2018 All weather conditions 178,7 Clearsky only 178,8 Period: 05.Sep. - 31.Aug.2019 All weather conditions 183,47 Clearsky only 183,53 Very similar results when using all weather conditions or only cleasky data Detected azimuth angle of south sensor, Station Freiburg © Fraunhofer ISE Results of Comparison of Orientation Detection All Weather Conditions vs. Only Clear Sky Period: 18.May - 04.Sep.2018 All weather conditions 178,7 Clearsky only 178,8 Period: 05.Sep. - 31.Aug.2019 All weather conditions 183,47 Clearsky only 183,53 Very similar results when using all weather conditions or only cleasky data Too less clear sky days Detected azimuth angle of south sensor, Station Freiburg © Fraunhofer ISE Period: 18.May - 04.Sep.2018 All weather conditions 178,7 Clearsky only 178,8 Period: 05.Sep. - 31.Aug.2019 All weather conditions 183,47 Clearsky only 183,53 Variations in a range of +/- 2° when using 30 days of data for orientation detection 178,7 183,47 Results of Comparison of Orientation Detection All Weather Conditions vs. Only Clear Sky Detected azimuth angle of south sensor, Station Freiburg © Fraunhofer ISE Huge variation is possible based on 1 sensor Decent in summer months with suffcient clear sky days Not reliable during German winter with a lot of overcast days Results of Comparison of Orientation Detection All Sensors vs. Only One Sensor Detected azimuth angle of south sensor, Station Freiburg © Fraunhofer ISE Huge variation is possible based on 1 sensor Decent in summer months with suffcient clear sky days Not reliable during German winter with a lot of overcast days Results of Comparison of Orientation Detection All Sensors vs. Only One Sensor Detected azimuth angle of south sensor, Station Freiburg © Fraunhofer ISE Conclusion and Outlook Model to detect orientation available and tested with 40 PV Live stations Clear sky days are important for detection Stable detection by using one month of data only possible, if Enough clear sky days happen Three sensors are used together Applying on power measuring data from PV system (distributed system) Determining configuration without documents/inspection Control of tracking system © Fraunhofer ISE AGENDA Motivation Our Data Set Automated Orientation Detection Automated Shadow Detection © Fraunhofer ISE Detection of Shadow Suspicious behaviour under clear sky condition Measurement from PV-Live Station Wendlingen Irradiance [W/m²] © Fraunhofer ISE Detection of Shadow Suspicious behaviour under clear sky condition © Fraunhofer ISE Detection of Shadow Measuring data affected by shadow 1 Year irradiance measurement of east-facing Si Sensor, PV-Live Station Wendlingen Obviously, there are shadow obejcts How can we filter the data? © Fraunhofer ISE Detection of Shadow Method 1 – Shade Tool (horizon recognition) SunEye (source: www.solmetric.com) © Fraunhofer ISE Detection of Shadow Method 1 – Shade Tool (horizon recognition) Annual sunpath generated from SunEye, PV-Live Station Wendlingen © Fraunhofer ISE Detection of Shadow Method 1 – Shade Tool (horizon recognition) Pros: Horizon data can be characterized before project or without measuring data Cons: Inspection on-site with proper weather condition Large plant may need multiple horizon datasets Orientation must be corrected afterwards © Fraunhofer ISE Detection of Shadow Method 2 – Machine Learning Select clear sky situations on the basis of satellite data (not influenced by shading) Plot measuserd clear sky index in dependency of azimuth and sun Elevation k*= G meas / G clear k = G sat / G clear , 0.75 © Fraunhofer ISE Detection of Shadow Method 2 – Machine Learning Data preparation Step 1: input © Fraunhofer ISE Detection of Shadow Method 2 – Machine Learning Data preparation Data processing Step 2: smooth © Fraunhofer ISE Detection of Shadow Method 2 – Machine Learning Data preparation Data processing Applying machine learning – Data clustering algorithm: Find effective threshold distinguishing dark / light Grouping points as dark/light Step 3: mask © Fraunhofer ISE Detection of Shadow Method 2 – Machine Learning Data preparation Data processing Applying machine learning – Data clustering algorithm: Find effective threshold distinguishing dark / light Grouping points as dark/light The horizon line is the upper limit of the dark part Step 4: result © Fraunhofer ISE Detection of Shadow Method 2 – Data Clustering Pros: You only need to sit in front of computer Highly compatible with real measuring data Easily applicable of all measuring spots (to be tested) Also applicable on power data without irradiance measurement Cons: Require a long period of data, at least 6 months © Fraunhofer ISE Detection of Shadow Method comparison SunEye Clustering SRTM © Fraunhofer ISE Detection of Shadow Quality flag of data © Fraunhofer ISE Detection of Shadow Quality flag of data © Fraunhofer ISE Thank you for your attention! Fraunhofer Institute for Solar Energy Systems ISE Jefferson Bor www.ise.fraunhofer.de Jefferson.bor@ise.fraunhofer.de
solarbe文库
