The FP7 DNICast project aims to advance current state-of-the-art of concentrating solar technologies by reducing uncertainty of short-term DNI forecasts and thereby contributing to increase the overall plant efficiency. Since October 2013, a multidisciplinary consortium of meteorological scientists, solar engineers and energy analysts, has investigated different methods for DNI nowcasting, with the aim to identify main advantages and drawbacks and suggest possible combinations depending on the user requirements. A full coupling of all methods is out of the scope of the project, but an approach on how to merge different information sources is intended. A portfolio of complementary methods for the nowcasting of the DNI and their combinations in order to cover the complete nowcasting horizon from now to 6 hours are the expected results of DNICast.
Project intermediate results have been extensively discussed with a large number of experts, including the members of the Advisory Board, recipients of the project newsletter and several other experts gathered through end-user workshops. Indeed, one of the main project aims is to exchange with the potential end-users of DNICast results, including industry.
The 3rd workshop had the aim to present and discuss overall results, as well as to collect feedback for streamlining before the end of the project, next October. The focus was to:
- disseminate the results for the benefit of the industry
- present and discuss lessons learnt and areas for further improvement
- move forward to make sure that the results are of use and can be further exploited by the research and industry communities
The 3rd DNICast end user workshop was held at Munich, Germany during June 1st 2017, as a side event of the Intersolar exhibition held at the same area. The speakers were members of the DNICast project consortium, and the workshop attracted several end users from the energy sector as audience.
The speaker’s topics focused on results, information and end products that were derived from the project. Special attention was given to communicate the project results to the CST community. During the talks enough time was given so that the consortium members answer questions of the audience that comprised of CST end-users, stakeholders and plant operators. The aim was not only to inform the audience but also to discuss, interact with potential end-users and receive feedback. The workshop concluded with a discussion on the nowcasting products and the optimum way they should be delivered to the end users.
A synthesis of the End-User workshop (based on deliverable 5.3 – workshop report for the 3rd workshop) is presented below. More information about the project can be found at: www.dnicast-project.net
1. Workshop Introduction
Mr. Abdelghani El Gharras (OME) opened the session by introducing the consortium, including both DNICast general and scientific coordination team. He continued by laying out some facts concerning renewable energy sources with a focus on CSP. The point that CSP is projected to account for 11% of the global electricity production by 2050, instead of 1% today (according to IEA), was stressed.
The introductory talk was continued by Dr. Marion Schroedter-Homscheidt (DLR). She continued by describing what the DNICast project is about and stressed that the project brings together energy sector specialists with meteo sector scientists. A presentation with major objectives from the project was shown that also included a brief explanation of the overall target of the project: CST optimization. Each WP of the project was introduced with special focus on WP4. Concerning the significance of aerosol in DNI nowcasting, it was mentioned that another workshop was dedicated to discussing this point and therefore little would be mentioned in the current workshop. The talk was concluded by summarizing the workshop’s agenda and by asking the audience to actively participate.
2. Session I. Synthesis of nearly final results
The session was initiated by a small speech by the session’s chair, Dr. Chris Gueymard (Solar Consulting Services), who explained that each presentation would last for 15 minutes with additional 5 minutes for discussion. He also asked the audience for good questions and introduced the first presentation’s speaker.
2.1 Requirements and accuracy needs for concentrating solar technologies by Mario Biencinto (CIEMAT)
The speaker initiated his talk stating that he is replacing Dr. Tobias Hirsch (DLR), who was unable to attend. The talk was initiated by an outline of what would be presented. The hypothetical question of “what actions would I take if I knew the DNI diurnal profile in advance” was raised as the talk’s motivation. The talk followed by explaining several definitions, such as the nowcast horizon, the characteristic time and length of a plant, and stressed the difference in definition between meteorology and CST professionals concerning nowcasting. The speaker noted the concept of linking nowcast horizon with the characteristic time via an application specific factor. It was also noted that the same method could also be applied to link the characteristic time with the temporal resolution. Two types of applications, with respect to nowcasting, were distinguished and two examples for each type were shown. In the first application, stabilization of CSP plant, it was stressed that the question to be answered related to the decision when to initiate a start-up procedure in a plant, and was noted that it depends on the nowcast horizon. How the characteristic time is involved in the example was also presented. The second example involved the optimization of battery storage utilization. The following two examples shown included the use of DNI maps, and it was stressed that their use reduces high deviations between observed and forecasted DNI values in a plant. Following the examples an overview of the CSP nowcasting applications were shown and explained. The speaker continued by explaining the notion of nowcasting accuracy again showing examples and stressed that it can be asserted to financial benefits. The talk concluded with a summary of what was shown and mentioned the talk was part of D2.1.
During the discussion, the speaker was asked by a member of the audience, who identified himself as a “newbie”, about how the pixel size can affect performance. The speaker replied that no such results were available for solar tower plants. The speaker added that the effect depends on the specific plant characteristics and it is thus complicated.
2.2 Presentation of the investigated nowcasting methods with all sky cameras and its validation by Andreas Kazantzidis (U. Patras) and Stefan Wilbert (DLR)
Prof. Andreas Kazantzidis initiated his talk by explaining how different partners were involved in the specific task he is presenting, followed by a summary of what to be presented. In order to demonstrate how all sky imagers operate, images with varying aerosol conditions were shown. The good agreement with of AOD calculation, established against AERONET observations was shown concluding that AOD can alternatively be delivered by all sky imagers. The calibration process was shown followed by examples of how polar to Cartesian coordinates can be converted. The all sky imager system operating at PSA was introduced for this purpose. It was highlighted that a novelty was to identify cloud type on a specific area of the image instead of appointing one cloud type per image. The importance of retrieving the cloud height was stressed and was noted that a 2- or 4-camera system is required for this purpose. It was mentioned that the cloud velocity is also calculated based on this method. It was also mentioned that a four-camera system, such as the one operating at PSA, improves the accuracy of the results. Examples of the camera system discussed were shown and evaluated against a ceilometer. The benefits of the 4-camera system were further discussed. It was stressed that the end product is a DNI map of the area the cameras inspect. The speaker invited Dr Stefan Wilbert (DLR) to continue the talk.
The presentation focused on the state-of-the-art with respect to validation methods before the project, which included only ground measurements as control systems. It was stressed that the proposed method included in addition spatial gradients which is an improvement. The suite of sensors used for validating the results at PSA was introduced and it was mentioned that spatial resolved irradiance maps created from radiometers and photos of the ground were utilized for validating spatial averages. The importance of considering temporal and spatial aggregations was stressed throughout the remaining talk and how such considerations reduce nowcasting uncertainties was repeated throughout. A summary of technical knowledge that came out of the validation process was discussed and was mentioned that the cloud height, as derived by a ceilometer is not relevant for DNI nowcasting. The limitation of simple models used for the same purpose was also stressed.
During the discussion, the speakers were asked if there is an optimum distance for placement of the cameras. Dr. Stefan Wilbert (DLR) replied that if the cameras of a two-camera system are too far they cannot be matched. He also mentioned that if 4 cameras are used they can be further apart. He continued by mentioning that the optical distance has been calculated only for specific cases were a constellation with cameras in the edges of a 1.2 km square yielded better results compared to a square with approximately 500 m side length.
A second question was raised related to the applicability of the methods shown to GHI. It was replied that GHI is implemented and that for GHI one can seek for the clear sky index.
A third question regarding the maintenance was posed. Dr. Stefan Wilbert (DLR) replied that at PSA the cameras were cleaned daily despite that not being the typical routine. He also mentioned that weekly cleaning is performed at a second test site to check the performance of the system. He also mentioned that for a new two-camera system method not both cameras need to be cleaned daily thus facilitating the maintenance process.
2.3 Presentation of the investigated nowcasting methods based on satellite images by Marion Schroedter-Homscheidt (DLR).
The speaker initiated her talk by showing the outline of her presentation in which the audience was posed three principal questions with respect to DNI forecasting. How these questions have been addressed by the scientific community before DNICast was discussed. It was noted that satellite images analysis have been applied to solar energy related issues for more than 20 years, yet the inherent limitations of this method, related to the image resolution, were also pointed out. The second generation Meteosat satellite was introduced and the discrepancy of identifying clouds in a non-sharp satellite image was mentioned. The talk continued discussing further the three principal questions raised in the beginning of the talk. Other methods currently in use were also presented and an overall summary table was shown. The talk continued by presenting results based on varying DNI diurnal scenarios using the new proposed method and their evaluation at PSA. The talk concluded by providing a list of relevant publications.
During the discussion, the speaker was asked about the use of WRF on calculating cloud height. Dr. Marion Schroedter-Homscheidt (DLR) replied that different heights were tested but did not make significant difference in the resulting calculations.
2.4 Presentation of the investigated nowcasting methods using NWP models by Thomas Landelius (SMHI).
The speaker initiated his talk by presenting the state of the art before DNICast. The speaker noted that despite the sun being the driver of earth’s weather, its direct result, DNI, has not been the focus in NWP models. It was also noted that cloud simulation in NWP is problematic. The audience was encouraged to read a recent publication that summarizes the state of the art concerning NWP. The talk continued by pointing out the prerequisites for a good quality forecast in general, followed by introducing the HARMONIE NWP model to the audience. The talk continued discussing two data assimilation methods used by the model, the pros and cons of which were laid out. It was pointed out, using examples, that caution should be exercised when comparing NWP solar radiation results with point data. The talk continued by discussing methods used to initiate clouds in HARMONIE and was noted that possible improvements in the calculation of cloud height and type could be of assistance. The talk concluded by showing what need to be done in the future and it was mentioned that ensemble assimilation methods seem to be the way forward.
During the discussion, the speaker was asked whether cloudy pixels were screened out and whether any effort was given to include this sort of information. The speaker replied that this was not an effort with the DNICast project, but it has been sought for outside of the project with promising results. It was noted once again that ensemble forecasts was the way to go.
2.5 Validation and uncertainty assessment of the DNICast nowcasting method by Jean Dubranna (ARMINES) and Yves-Marie Saint-Drenan (ARMINES).
The speaker initiated his talk by discussing the motivation behind it. It was mentioned that a set of ground DNI measurements were used for the validation process with several parameters taken into account. It was noted that from a network of 14 stations that was shown, instead of focusing on a single year, 4 three-month periods were selected. This strategy allowed different meteorological patterns to be assessed. The speaker pointed out that 10 different models were validated each using a different nowcasting strategy, a summary of which was also presented. The difference among the models was discussed and was noted that their heterogeneity posed problems in the validation procedure. It was also noted that PSA was well represented. An overview of the model performances for all stations was presented. It was stressed that some models deteriorated under certain scenarios but not all under the same ones and therefore the use of specific models for specific station was suggested. The concept of performance summary sheet was introduced and it was demonstrated how they can be interpreted followed by two case studies regarding PSA. The speaker continued providing some more brief information on the performance summary sheets, but continued to explain thoroughly the DNI variability classes used in the performance sheets. Dr. Yves-Marie Saint-Drenan (ARMINES) was invited to continue the talk. The new speaker mentioned that information should be synthesized to predict trends which would assist in selecting the appropriate model to nowcast DNI. It was also mentioned that a great cause of discrepancy is that nowcasting is initiated with errors even at time zero. Cloud transport was identified as a major source of this discrepancy without being the only one. Examples regarding this point were presented and a summary of the presentation was given to conclude the talk.
During the discussion, the speakers were asked whether a similar validation process was considered using sky cameras. Dr. Jean Dubranna (ARMINES) replied that only NWP and satellite images were used for this analysis and that this point is addressed in a separate report.
2.6 Combination of DNICast predictions: Examples of combination methodologies and improvements achieved by Lourdes Ramírez (CIEMAT) and Stefan Wilbert (DLR).
The first speaker initiated her talk by showing the outline of her presentation explaining its main focus, followed by a summary of the nowcasting inputs that were utilized. It was mentioned that the main goal was to harmonize them. She explained that two sites, out of the 14 available, were considered for the presented work mainly due to data availability issues and noted that the same periods, as presented by Dr. Jean Dubranna (ARMINES) earlier, were used. It was also noted that effort was given to identify periods of great variability instead of “good” subsets. The speaker presented an overview of the two methods that were employed for the combination of the results. At this point Dr. Stefan Wilbert (DLR) was invited to continue with the talk. He presented the uncertainty weighed combination method thoroughly, including equations and results regarding PSA. The talk was continued by Dr. Lourdes Ramirez (CIEMAT) who explained equally thoroughly and showed results from the second method, the multi-regressive approach. The speaker noted that in the presented method a dynamic fitting was employed, and she stressed that both methods presented performed better than any single method used as input for the combination, noting that the second method she presented is still under development.
During the discussion, Dr. Lourdes Ramirez (CIEMAT) was asked whether another model is considered. She replied that the presented models are the suggestions at the moment. Another member of the audience asked Dr. Stefan Wilbert (DLR) details about the uncertainty estimation. He replied that the uncertainty calculation was left to the data providers, adding that the methods, used by each individual data provider to calculate the uncertainty, is different and available. Dr. Marion Schroedter-Homscheidt (DLR) also noted that each data provider used very different method for calculating the uncertainty. A member of the audience also exerted the problem of calculating uncertainties and suggested to use the plain average as a reference. Dr. Stefan Wilbert (DLR) replied that uncertainty calculations should not burden the end user but only the data provider.
3.1 Preliminary assessment of the impact of spatially-resolved DNI on the electricity production of Concentrating Solar Technologies by Mario Biencinto (CIEMAT)
The speaker initiated his talk by discussing its outline. He then explained how the current method of implementing an average DNI value over an entire plant is problematic, followed by a summary of objectives aimed to cope with this problem. A selection of non-ideal DNI diurnal profile case studies was shown, serving as basis for the presented model implementation. It was noted that 2 scenarios would be investigated that differ in the number of DNI values for the control system. It was discussed that results using the typical method of using one average DNI would also be shown as reference. The focus of the talk was oriented on the case study of June 1st, based on which results concerning CSP from all methods were individually shown and thoroughly explained. The weaknesses of each method were also discussed. The talk continued by exhibiting cumulative results from all methods using all profiles mentioned earlier with respect to maximum temperature, which was stressed to be an important parameter, and electricity production. It was noted that there is no clear rule that the proposed method will always perform better than the currently used method with respect to DNI. The intercomparison of methods was shifted to CPV where 2 plant types were distinguished. For each type examples were shown and explained. Emphasis was given to maximum and minimum forecasted DNI values. It was stressed that in the CPV case the use of spatially resolved DNI resulted in avoiding overestimation and/or underestimation of the forecasted DNI.
During the discussion, the speaker was asked whether another method was used as reference. He replied that he has done so even though he did not show any of these results, adding that it is interesting to investigate these methods using different controls.
3.2 Assessment of the effect of DNI nowcasting on the yield of exemplary CSP plants for a time dependent feed-in tariff by Jürgen Dersch (DLR)
The speaker initially presented his motivation for his work, followed by the methodology implemented for the goals to be achieved. He mentioned the software utilized and stressed that the aim of his study was to produce a measurable outcome. The speaker mentioned that only specific power plant cases were examined with the aim to maximize revenues, stressing that heuristic methods were used. The audience was then familiarized with two plants of different type (parabolic trough and solar tower) whose characteristics were shown. The speaker then showed the characteristics of the forecasts datasets used, including the concept of the ideal forecast dataset. He also mentioned that a combined forecast from the presented suite was also used. Graphical examples of these datasets were shown with respect to PSA and revenue yields for the same plant based forecasted values were shown. It was stressed that the combined forecast performed better than the rest but always worse than the ideal case. It was noted that the combined forecast data were only available for 3 months and not a full year. The speaker continued focusing on a case study day discussing the performance of all forecast datasets. The talk moved to discussing the performance of the datasets with respect to annual revenue for multiple years at two sites (PSA and Ghardaia), including the discrepancies of each dataset and site. It was stressed several times that the combined forecast performed better in all examined cases, apart from one.
During the discussion, the speaker was asked why the benefit is greater in the solar tower than parabolic trough case. He replied that he is not certain. Dr. Marion Schroedter-Homscheidt (DLR) mentioned that it has to do with technicalities such as the sensitivity of each plant to the time resolution of the forecast method.
3.3 A web demonstrator for DNICast results by Carsten Hoyer-Klick (DLR)
The speaker initiated his talk by laying out the objectives of building a web based application. He stressed that a huge amount of data is to be visualized and that he has considered a classical architecture of client server approach that reduces data transfer loads. The speaker continued by explaining the home screen of the web app and its capabilities. He mentioned that different forecasting methods can be assessed with respect to several time periods. Examples of the interactive plots were shown live and mentioned that some features are not yet implemented.
During the discussion, Dr. Stefan Wilbert (DLR) asked the audience to share ideas on how to visualize results. A suggestion was made that the user should manipulate the complexity of the graphs. A checkbox feature was suggested as an example. A second question was raised whether the complexity of the graphs relates to the diversity of the forecasting methods and their varying results. Dr. Marion Schroedter-Homscheidt (DLR) replied that it is still an open issue how to decipher varying results and she stressed that this question relates on how the message of varying forecasts can be delivered to the end user.
3.4 Integration of DNICast methods in CSP plants by Martin Gaston Romeo (CENER)
The speaker presented an overview of the DNICast project and its main goal, followed by a summary of the tools used in the project (satellites, NWP models, measurements, sky cameras). The main research outputs of the project and their impact was also discussed. It was also mentioned that data validation methods were also improved. The speaker continued discussing each scientific topic covered by the project and presented the key scientific questions regarding nowcasting. The use of sky cameras as nowcasting tools was thoroughly discussed focusing on the end-product, the DNI maps. It was noted several times that at least two cameras are required and that the results are produced live at the CSP facility. The use of satellite images and NWP models as forecasting tools was also discussed, stressing their implementation limitations. It was noted that all the tools mentioned before can be integrated into one combined forecasting model, followed by the validation methods that should be sought when forecasting DNI. How consortium partners have implemented the project’s results was mentioned and it was noted that additional effort aiming at further implementation should be sought for. The speaker concluded his talk by laying out the DNICast project’s achievements. Dr. Marion Schroedter-Homscheidt (DLR) commented that DNICast is a project oriented on results and open to the end user.
4. General discussion and conclusions: How to move forward for industrial exploitation.
Dr. Martin Gaston Romeo (CENER), who chaired the session, raised questions on how the DNICast project results can be communicated to the DNI community. It was mentioned that this question specifically referred to the plant operators. Mr. Zeyad Yasser (Flagsol) replied that it is not clear how the results will be implemented. He mentioned that graphs should be clear, not burdened by a surplus of information. He stated that this is essential for rapid decision making. Dr. Martin Gaston Romeo (CENER) wondered if a time-series can hold enough information for such decisions. He also asked about the usefulness of dealing with a single time-series. Mr. Zeyad Yasser (Flagsol) replied that it should suffice. Dr. Martin Gaston Romeo (CENER) mentioned that quality aspects should be considered and provided some examples. Dr. Marion Schroedter-Homscheidt (DLR) stated that she is well aware that single curve plots are sought for when dealing next day forecasts. She wondered how the uncertainty can be taken into account. Mr. Zeyad Yasser (Flagsol) mentioned that the topic of uncertainty in decision making has been discussed among operators but he stressed that it was done having one curve in mind.
Mr. Juan Ignacio Burgaleta (SENER) asked what the average DNI value represents. Dr. Marion Schroedter-Homscheidt (DLR) replied that it is a spatial average covering an individual plant. She also stressed that the related uncertainty is not fixed and can vary from case to case. She also noted that a single value can be easily interpreted but it is not accurate. Based on her expertise an average value can be assorted with RMSE up to 80%. She wondered if it would be helpful to forecast a range of values and stressed that the single value strategy is not helpful and that the community should move further from this aspect. Dr. Stefan Wilbert (DLR) added that the selection of the plant control strategy could be made including the forecast uncertainty and the plant characteristics. He mentioned that a method to determine applicability classes for this purpose would be developed and noted that the experience of CSP users would be very helpful at this task. He suggested collaborating outside of the workshop.
The topic of the talk changed. It was mentioned by a member of the audience that decisions should be related to financial costs. Dr. Stefan Wilbert (DLR) replied that different camera systems were not tested. He also mentioned that it is intended to do so within the DNICast project. He also noted the large difference in price concerning cameras. He mentioned that the camera systems used at PSA cost approximately 700€, while a single camera cost can rise up to 11,000€ or 50,000€. He also mentioned that the quality is in some popular cases not related to the price.
Mr. Juan Ignacio Burgaleta (SENER) asked if the temperature and relative humidity will be included and noted that sandstorms should be considered in the forecast. Dr. Martin Gaston Romeo (CENER) wondered if large variations in these parameters have been observed. Mr. Juan Ignacio Burgaleta (SENER) replied that typically the temperature and humidity are not important but sandstorms can be on specific cases. He stressed the importance of aerosol and of sandstorms. Dr. Marion Schroedter-Homscheidt (DLR) wondered if they change within an hour. Mr. Juan Ignacio Burgaleta (SENER) insisted that sandstorm should be forecasted. Dr. Martin Gaston Romeo (CENER) replied that such forecast is qualitative. Dr. Marion Schroedter-Homscheidt (DLR) added that such forecasts were discussed in WP2. She noted that it is different for the nowcasting process. Mr. Juan Ignacio Burgaleta (SENER) also mentioned that soiling should be a known parameter. Dr. Marion Schroedter-Homscheidt (DLR) replied that this is a big subject. She added that deposition is numerical simulation input but it is not produced as an output mainly because there are not enough data to validate the simulation results. The option to participate in a deposition (soiling) data comparison was brought forward.
Dr. Martin Gaston Romeo (CENER) concluded the meeting by thanking everyone for their participation in the workshop.