TRUST-PV webinar Operational Stability in the era of large RES penetration
SolarPower Europe is a consortium member of this 4-year research Horizon 2020 funded project, which intends to improve the performance and reliability of solar power plants. The research project will be completed in June, and has some exciting findings to share.
During the webinar, the project will present key findings from cutting-edge research on advanced power plant controllers to support operational stability and advanced ancillary services and will dive into forecasting for advanced operational stability.
hello everyone um first of all I’d like to welcome everyone to our trustview webinar about operational stability in the era of large rest penetration I am Juliano and I work at stcraft one of the consortial partners of trust PV where we have been working quite a lot in the demonstration activities of the project being some of them related to the topics that you’re going to hear about today I’m going to be the moderator of this webinar and before we start I’d like to thank solar power Europe for organizing the webinar and also CH Marion lafuma from unat and vot katsiki from in access power factors that will be presenting their very interesting topics our webinar today centers around the core topic of operational stability in the context of significant renewable energy source integration as many of you may know trust PV is a 4-year EU Horizon 2020 project focus on solar photovolatic performance and reliability as we approach the pro the Project’s conclusions later this year I believe it’s essential to highlight our achievements trpv has generated valuable insight and reports accessible via our website these reports cover diverse topics including energy yield optimization PV module reuse recycle F detection decision support systems and the challenges posed by large scale res integration into the grid among many other relevant topics so if you’re interested to know a little bit more about it we invite you to visit our website and have a look at it um on the topic of rest penetration and GD friendliness throughout these almost four years the consortion partners have researched and produced impactful result in areas such short and Midterm forecasting methods aiming to enhance Market participation services and operational stability scenarios for both virtual and utility scale power PL PL and also that further on the topics of fully flexible and interoperable PV plants on today’s webinar we will focus on two topics the first one forecasting for advanced operational stability we will explore how forecasting models can be improved by the use of different methods improving operational stability and the second one Advanced power plant controllers understanding how advanced controllers can support operational stability and also ancillary services so the first presentation of today and our first Speaker Marion lafuma uh Marion lafuma holds a BBA obtained at EAC business school followed by an MBA in operational mement and sustainable performance she joined runat in 2011 when she launched the company’s macian subsidiary she then joined the provision office back in 2013 where she’s where she’s in now in charge of the firm’s Business Development with a passion for sustainable and responsible energy Marion and her colleague mat to have been in charge of a leading of leading the forecasting aspects of trust PV project this led to the publication of a report entitled forecasting for advanced operational stability available also on the Project’s website uh it is the results of these reports that you will be presenting in today’s webinar hello everyone thanks a lot Juliano for this introduction thank you also to solar power Europe for this opportunity to talk about forecasting within trust BV so Juliano made a great job at introducing what trust BV is and uh now I’ll try to do the same regarding forecasting um so forecasting for advanced operational stability is uh the title of my presentation and it is uh because we were in charge at reuni of a task dedicated to this particular topic it also involved a few other partners from the Consortium namely urak iMac T and data providing from Sol Mony each of um reuni W urak iMac and Tuda were in charge of a specific subtask which I’ll go into details today and uh this specific task lasted for about a year and a half and as Juliano said it led to the publication of a report entitled forecasting for advanced operational stability which you can download on the trust PV website so uh for you know um each subtask we focused on a specific time Horizon so I I will talk to you about why this is important uh here is the overview of each subtask iMac which is a Belgian Institute Research Institute focused on short-term forecasting modeling based on lowcost Sky cameras a reuni that I’m representing today focused on using satellite data for intra hour forecasting T worked on Advanced forecasting in digital twins digital twins based on artificial intelligence and as said they used solar monkey data and finally urak focused on day head forecasting using numerical weather prediction models now maybe all of you are not particularly familiar with forecasting and that’s why I want to start with this graph which shows why we have several methods used depending on the time horizon horizon and the spatial resolution so um on the x-axis here you can see the temporal resolution and on on the y- axis you can see the space faal resolution the idea is that forecasting needs to beat persistence at all times what is persistence well it’s saying that um today well the current weather now at this time T will be the same in one minute one hour one day so um our focus is to beat this um persistence model and to do so there are several Technologies which exist and which are more impactful depending on a specific scale or on a specific location here in uh green you can see TSI that stands for total Sky imager and this is actually the best technology for forecasting anything between 1 minute to 30 minutes in advance on a very local site because you need to install sky cameras locally um satellite data are the most interesting technology to be used from 10 15 minutes up to 6 hours ahead but I’ll show that within trust PV we’re actually focusing on five minute ahead forecasting using satellite data and uh that’s quite an Innovative aspect and that’s what I’ll talk about later on um satellite data they have a larger field of view because you’ll see later on as well uh we focus on larger visions compared to Sky Images statistics are also a good approach for intra hour forecasting or in intraday forecasting and the these methods are based on neural networks or artificial intelligence models machine learning models and finally nwp stands for numerical weather prediction and these are the best Technologies to be used to forecast up to several days ahead and starting from um a few hours ahead so what we call day ahead forecasting and within this task we looked at all four of these uh different time Horizons and time scales to begin with uh imch has said they focused on using lowcost cameras I’m insisting on the fact that we’re talking about lowcost cameras because other cameras exist but for this particular aspect it was um low cost ones and what was used um so was was were Sky cameras positioned on site looking at the sky above an installation and the idea here was to look at algorithms to determine power production forecasting up to 15 minutes ahead with a 1 second resolution the first step to do so is to calibrate your instruments and to look at uh the the mask of the instrument so looking at the surroundings removing all that and classifying clouds depending on the type of cloud we can have metrics showing the accuracy of the forecast here you can see eight images in reality there are seven Cloud classes using this technique because we put uh two of them in the same class um the results of this first subtask led by IMC was that um they installed a camera on a trust PV site in Italy but unfortunately they didn’t have enough data to be able to provide interesting results so they also used other data sets which had the great advantage of being high quality camera setups one coming from Germany and one coming from China the idea was to use these really high quality data sets to look at the technique using Sky cameras lowcost Sky cameras and in the end it showed that um when you had clear sky uh images so without clouds that what presented the lowest error in terms of relative mean squared error and when you had classes of clouds which were very variable um so very cloudy conditions with a lot of variability that led to the highest errors but anyway it showed also that uh that was really the best technique to get short-term forecasts in the next minutes now runi wat was in charge of a second subtask using satellite data maybe some of you are not familiar with how we use satellite data dat on a daily basis to provide intraday forecasts so I’ll just Linger on this topic for a few minutes um what is done usually is that we look at a full disk image so on the top right hand corner here you have a full disk image providing from the European meteorological satellite Agency youat for a satellite a geostationary satellite entitled metat second generation this satellite takes about 15 minutes to cover the full disk image that you get here um and every 15 minutes you get a new Full disc raw image now we wanted to focus on a specific service provided by MSG which is called rapid scanning service which is the one that you can see the animation that you can see on the bottom right corner and the idea here is that it focuses on a smaller part of the full disk image so it can uh scan it in 5 minutes only and our idea within trust BV was to look at this rapid scanning service to see if it brought any additional added value in terms of forecasting accuracy first we wanted to focus on three trust PV plants one in Italy and two in Spain and it appeared that uh the three sites had quite similar weather conditions so we also added 25 German stations I’ll talk about this uh on the next slides just to give you a a a grasp of how we do the forecasts using satellite data we get a full well um raw image either a full disk image or rapid scanning service image and then we focus on the location of Interest a raw image comprises clouds but also topography relief uh snow mountains everything the first step is to get rid of anything that’s not a cloud to only keep a cloud index map and then we look at the Cloud’s movements what we called what we call Cloud motion Vector fields and by using consecutive images we’re able to look at how the clouds will move in the coming hours from that we’re able to determine ghi Global horizontal erance and from ghi we are able to apply a nance to power conversion model when we have access to all the plants details so the idea within this task was to produce power production forecasts so I said the two well the three sites um within trust PV in Italy and in Spain didn’t present a lot of variability in terms of cloud uh so we decided to also look at uh 25 German stations located all across the country of Germany because obviously uh they are more variable in terms of cloud presence and the conclusions were that the rapid scanning service will will ought to be used because in any case it doesn’t bring a lot of added value when the cloud conditions are very stable like in Spain or in Italy but when you have a lot of variability it can decrease the mean average error by nearly 5% so you might as well use it when it’s available now going to the third subtask uh led by T um the idea here was to look at digital Twins and the first challenge was that well sometimes it’s very difficult to have a lot of details regarding power plants solar power plants due to uh privacy issues difficult access insufficient data quality Etc so luckily within trust BV we had Sol monkey which um produced well had access to a lot of anonymous data which they handed to T elf and through those data they were able to conclude things um the the idea here was to look at four hour Ahad forecasts and once again uh Power production if I’m not mistaken however um before dealing with any artificial intelligence algorithms it was important to look at the literature and to do a review of about a 100 articles that had been published between 2015 and 2020 to look at the most um useful algorithms and so uh this led to a first review paper published in energy reports comparing all these hundred well 100 scientific articles and the main conclusion was that if if you put in place a very complex algorithm it might uh bring some interesting results but it’s too comple Lex to replicate and therefore putting in place a very complex algorithm is often unnecessary because it’s uh the value added is too small compared to the complexity of re producing it um also afterwards um Alba which was the person from telf put in place some physical models and compared that to machine learning models and her conclusion was that um the research led to beat the performance of the physical model previously used thanks to a machine learning model and the best machine learning model um according to her is XG boost apparently it brought the best results so those are some conclusions from this third subtask and now um finally the fourth one um which was led by urak the idea here was to use numerical weather prediction models which reuni watch provided to urak um so numerical weather prediction models for those of you who don’t know they are using um lots of models coming from weather agencies across the world so weather agencies have supercomputers on which they run their models supercomputers cost billions of Euros uh so reu what unfortunately doesn’t own some but uh we purchase streams from uh these different weather agencies and then we blend them and we are able to have some um really thorough um you know um forecasts by putting some weights on the models which perform better depending on the type of climate we’re interested in anyway uh that is to say that urak used nwp models to look at the development of deterministic and probabilistic forecasting methods so they had um a lot of but they put in a lot of work to to look at these different forecasting models um for the deterministic forecasts it showed that the neuron Network models were the outperforming um ones with a skill score of nearly 45% in Italy and nearly 40% in Spain and um the skill score was understood to be the best metric possible for the these types of um comparisions because they’re looking at reliability uh persistance well um um um yeah uh I’m going to to go through this um in terms of resolution reliability and sharpness sorry and then um Power probabilistic forecast it showed that well the the work done by Ur by Mark showed that the analog emble method was the most performing one okay I don’t have a lot of time left so I’m going to conclude and if you need to take uh to to grasp a takeaway for each subtask uh it would be what you see on the slide here uh for iMac uh so looking at Sky cameras they offer the best accuracy for a minute ahead forecasting for the rapid scanning service which we looked at at reuni it always brings added value especially when you have variable Cloud conditions for the T subtask looking at artificial intelligence um the accuracy of the models highly depends on certain conditions like climate location time Horizon and uh regarding urak and numerical weather prediction models analog emble methods provide best probabilistic forecasts that’s it um happy to to answer your questions later on during the Q&A now I’m going to hand over to Juliano again and to vasel all right thank you very much Marion for this great presentation and very interesting results about different forecasting methods um now yeah now we’re going to go move to vaso’s presentations on Advanced power plant controller support and operation stability and Advanced ancillary Services uh Vaso katsiki holds a degree in electrical and computer engineering from ntua with specialization in power system engineering and a master’s degree in Business Administration since 2010 she has worked as project Engineering in rest projects and as a technical product manager in industrial par electronics and batteries her work experience includes B hybrid offgrid and Rural electrification projects she she has participated in several R&D pilot projects on demand response smart energy grids and sector coupling she’s currently working as a product R&D manager in the controls and grid integration Department of in access power factors with special focus on B applications she’s also involved in R&D projects on R and Hybrid Power and Hybrid Power Plant control and asset performance optimization um and now we move to vaso’s presentation uh hello uh thank you so uh let me start with this part of the presentation that um uh mostly focuses on a advanced power plant uh control features further let’s say to production predictability which was the main topic of the first part of the presentation so uh just to start some introductory notes in the context of uh trust PV er er trust PV initiatives on Advanced power control for operational stability so trust PV has attempted to serve as an end to platform for um analysis simulation development and pilot testings of a power system in order to elaborate on the concept of uh a new generation ideally flexible PV plant uh this concept includes several pillars so production predictability that has already been described by um Marion improved operational stability at the point of interconnection enhanced interoperability and integration with uh the complex landscape of energy markets production dispatchability uh following forecast letain employing um several energy resources such as batteries apart from variable Renewables and virtual power plant um operation virtual power plants uh at commercial utility industrial and uh utility scale so in that frame in Access through uh their experience in uh Advanced power control has Le this uh the relevant activities uh and through Trust has attempted to further involve certain aspects of uh Advanced par plan control to support operational stability we will refer in the frame of this webinar in certain Concepts and the um evolved features per se so we will referring three main Concepts that serve both operational H stability and interoperability and Market integration which is the development of um an even faster uh response time system operational uh time uh the ability of service talking in order to respond to today’s complex Market landscape and a better Market integration uh so to begin with a few words on the inherent role of the part plant controller so the part plant controller is actually the heart of a PV plant or of a an ideally flexible PV or hybrid plant of today that performs a realtime control in all energy assets renewable assets the best assets Etc it facilitates seamless um integration of uh Renewables batteries and hydro plants it ensures great Code Compliance and in case of Ops it ensures offgrid stability it enables a dispatchability and frequency it makes the plant H frequency responsive so first through uh capacity fairing techniques and second to allowing the participation and delivery of um a successful delivery of ancillary Services last the power plant controller is in charge to H Implement uh control dispatch schedules and support participation capacity and uh wholesale markets uh a few words on in access advanced par control architecture um we’re talking about a sophisticated framework uh based on building blocks uh which is H extendable adding new uh Logics and uh new inputs and outputs when necessary H it is an open architecture so it is vendor dependent it is modular and it is uh robust uh it ensure interoperability and it offers native um compatibility with a wide range of communication protocols but also with proprietary uh ones so focusing on Logics and entering the first uh let’s say key topic of this part of the webinar which is the uh service tacking uh possibilities we are focusing on uh we will navigate through that uh topic through an example of uh uh Dynamic balancing services and specifically the case of UK uh so today’s complex um energy Market landscape has been has been built in order to um combine ER the requirement of market-based approach and to to achieve demand Supply equilibrium and the increased um request for operational stability uh given the large scale The Continuous the continuously increasing and already large scale renewable integration into the conventional grids so we have a multitude of energy markets and balancing Market the example of UK for example includes five auction based markets uh that differ in terms of um uh delivery requirements glentis gate closure Etc a continuous intraday Market with different payment schem and on top of that um three main Dynamic balancing Services the dynamic containment Dynamic liation and dynamic regulation that are that are employed by the system to maintain the frequency within an acceptable uh range we see for example the delivery cures here for the three balancing um uh Services uh a typical um let’s say uh uh plant that provides ancillary Services usually this the case of a battery uh plant H has um and this is an example of a dynamic regulation High a high frequency uh event it provides a response based on its contracted time and the delivery curve that we uh presented here that is Prett Define from the system um in cases where we where the frequency exceeds a predefined uh range so this is something that is totally um uh let’s say followed however there is a necessity given um let’s say given the um the existence of multiple concurrent Andel Services there is um a necessity and a requirement to be able um to perform form service ping so leverage a single battery asset for example to provide multiple ancillary Services simultaneously this can be done not sorry not only ancill service but multiple services so it could be multiple ancillary service concurrently but participation at wholesale market and ancillary Services also um this is um required ER this is is needed because it can serve it can benefit let’s say the owner with uh maximizing it their revenues through participation in multiple concurrent uh Market offering and it’s also it also helps in terms of risk um management given that the closer we are at the response time H the better forecast um er we have uh uh so it is there is a need let’s say to be able to to stack this service and this has been achieved for example we show here um a case uh with concurrent participation in two different balancing service Dynamic operation Dynamic um contain we can see here for example um in the sorry uh we can see in the graph here for example that in cases we have over frequency ER the uh active uh Power set point for a dynamic containment is employed according to uh specific contracted power so according to the um er engagement from the service and when we have an under frequency event the set point that is um effective is a set point of dynamic moderation service according to to the respective H contracted power so we can see here concurrent operation where service stking is Achieve you can see another example here of service stacking that shows the possibility to concurrently participate um in wholesale market and a dynamic balancing service so the active power Baseline here that we can see we have zoom let’s say in the representative uh case we have a this active power Baseline that is actually uh the scheduled Market dispatch H based on the energy Market participation this is considered as Baseline for the dynamic regulation participating we have a relevant set point from the dynamic regulation and these two set points are stack in order to have the final let’s say response of the plan satisfying both uh requirements so successfully accomplishing both uh Services simultaneously so the first part was regarding this service tacking capability that is essential for interconnected flexible plant so interconnected um H plants this is required the market based approach to ensure operational stability and it’s interconnected uh let’s say um plant uh owner ER can participate to maximize revenues um and better ER create value from their portfolio uh moving along on operational stability um Concepts ER that are even faster let’s say that require Fast Response times we will focus a little bit as an introduction to the response times perent so operational Cycles control operational Cycles which is part of the overall response time of the plant um are required to be more more quick so nowadays operational cycles of 20 to 50 milliseconds are wi widely required and applied to operation battery assets the cases that we have already seen they already operate on such um cycle times uh so these times are essential to support as we as we mentioned before Dynamic balancing um service such as for example UK Services which which was uh the first uh service to be considered fast but nowadays we have even even faster service like for example frequency fast Reserve Services related in the airort market Etc besides the dynamic balancing Services we also have requirement of a even lower cycle time even quicker cycle times down to 10 milliseconds to support not frequency changes through the participation of um in service but to support the rate of change of frequency and provide inertia related services or let’s say F frequency response so uh this Services the necessity of these Services has um ured due to the the gradual Reliance on renewable generators very the conventional um generator and the the uh reduction of the overall rotating rotating mass of the grid uh last but not least Fast Response times are also required for a real time Telemetry that is required in terms of system operat so there is a need to log in on 2050 to L 2050 Herz performance data in order to evaluate the participation services so focusing on the second Point H cycle times down to 10 milliseconds have been achieved in order to be able to support this Advanced response overall response requirements because the control is part of the overall response of service such as sythetic inertia further to this cycle reduction ER some test some tests have been conducted within CHP project to address the uh the power oscillation to uh to address the power illation chapter um that is a a modern requirement that up to now ER refers to the synchronous uh generators and um uh they are which are obliged to address uh let’s say the um uh issues to address these issues of power res relations that derive from lack of synchronization let’s say between the different uh parts of the GDs so up to nowadays synchronous generators are obliged have the duty to address this um Power ulations however due to the increase of better based Generations H we foresee that uh this will be a future obligation also for a PV plans uh so the uh issue is that for example due to certain instabilities and lack of synchronization in different parts of the power relations may arise and relevant volage stability issues in frequencies for example that may vary from 0.1 0 point2 up to one and a half or two Herz and the uh P plants the renewable plants will need to um respond in a way in order not to aggravate these power oscillations but to uh improve them and make them and even eliminate them so we have conducted given the difficulty in conducting a test um involving let’s say high high voltage abrupt instabilities we have cooperated within trust PV with Force um er H we have a cooperated with for within the frame of trust TV in order to address this in order to create to simulate a high voltage grid trip a rotating generator creat power oscillations and then employ um and then let’s say observe the behavior with the power plant controller interconnected in the grid I’m concluding in a few minutes my presentation in order to respect the time and move to uh Q&A uh so we see here an Abrupt trip uh that was achieved through the H real time simulation within the hardware in the loop and we see here the power oscillations that have been created in the blue line without the presence of the power plant controller and then in the Gray Line we see the power res relations that have been registered with the presence of a b plant with a power plant um controller so this test which is a real time simulation so it’s a real time response to a power illation three has been achieved due to Fast Response times however in um real life let’s say real life implementations this in order to counterface such oscillation requires a real fast response time also from the part of the inverter so what we have seen is for The illation Dumping what we are concluding is uh through the studies that for The illation Dumping to be effective we need to have let’s say responses in one tenth of the illation for example if we have a 1 second illation we need to have a 100 millisecond overall response time including the overall inverter communication loss Etc this is overall very very challenging topic that requires also continuous uh let’s say Improvement of all components of uh ER the PV plant uh since uh my the time of my presentation is um uh leading to its end I would like just to to mention let’s say the last part ER that is the work that have been that has been performed I’m I’m skiing some slides here uh sorry for that the work that has been performed the frame of virtual power plant on residential scale energy communities so what we have identified is that for example as we have the requirement of capacity fing of FM response in some cases for utility scale plans the same let’s say requirement for a firm response um will be also essential in the future in the residential scale behind the meter ER PL so in the frame of renewable energy Community optimizing uh the a community energy profile so performing Collective uh sell balancing would be the a or has been a topic of studies we have studied centralized uh base control within um these activities sorry are LED from um our partner in uh urak we’re supporting that so we have studied centralized battery control in the frame of energy Community decentralized but with the centralized management and distributed um this through small distributed batteries with decentralized control uh we have uh also created let’s say a specific um software architecture for a virtual pl we have a pilot installation in the Netherlands and um that comprises 15 households at the moment and we have created a unified U virtual power plant platform through unity in Access Unity platform that enables centralized monitoring and management of all these distributed energy resources uh the the previous um the previous study that I mentioned here is a work in progress this is just a concluding comment and and we have SE that centralized control with one objective or capacity fairing so aligning PV and um battery production to be equal to the load or equal to steady output is the more effective one and the most beneficial for the GD so thank you all um thank you for your time let’s move to uh let’s go back to julan and move to the Q&A section thank you very much vasel and thank you very much for your very interesting presentation as well and we can now move to the Q&A session from the audience um I think let me see I can see while questions are being uh written there I will ask a question myself to to Marion so on the forecasting mod mod that you presented uh I was actually wondering you work quite a lot with satellites and so on and how can large amounts of data or like big data help increase the performance of forecast at all time Horizons okay thanks for this question well it’s actually indispensable if you want to put in place machine learning algorithms um a lot of data are required and I also mentioned this during the presentation um when we looked at at T’s part for instance the use of solar monkey data was necessary to be able to provide um digital twins otherwise you can’t even go into the concept so large amounts of data are a necessity especially when looking at uh machine learning and deep learning and artificial intelligence so the more the data the better all right um thank you Marion um there’s one from the audience I think it’s more related to Vaso is there a paper about the centralized best pilot uh I believe she’s mentioned the one you mentioned your presentation about trust PV uh yeah so there are a few papers uh we can send them so some of them are a um already open already uh published some are in publication uh procedure maybe we can uh uh keep but we can send them to uh we can send them later I guess we can send the links but yes there are three or four papers here all right yeah we can coordinate with solar power Europe to make it uh available somehow um the links um another question um chaso on my side uh from what you showed in the improvements of the power plant controllers um although PV power plants are the most most of them are being built recently but there are some quite old ones and implementing this power controls in Old PV power plants is it a possibility or not really viable you would have to change everything or or not really how how does it work if you’re going to apply these PowerPoint controllers in an old Power Plant no it is uh valuable if also supported by the end devices by The Edge devices also in terms of control weon per it is um viable through overthe a updates and not uh Hardware change of course there are limits that the hard Hardware Set uh but uh yeah all in all it’s let’s say 9% software issue okay interesting uh thank you Vaso um another question from the audience to Marion uh regarding short-term forecasts uh do you know what was the processing time of the rapid rapid scan images yes so a new input data so a New Rapid scanning um scanned image is available every five minutes and then if you mean the processing time to be able to extract a forecast from this input data um it’s maybe a few seconds to a minute maximum okay interesting quite quite fast um another question from the audience to Vaso um Rec and local energ communit uh you use centralized controller from huawe are there other or independent Solutions on the market uh there are already implemented okay uh so in the frame of this uh project what we have done is that we have um a installed project partner is Huawei so we have uh not installed a centralized controller on on field but we are ER monitoring through the um API of uh Huawei and we are calculating centrally uh the frame of the project charge and discharge tasks and we are implementing them through this interface of uh Fusion API uh of course uh there should be other independent Solutions in the market but the solution that we worked on uh was a joint let’s say f support from in Access Huawei [Music] and uh Ur all right um thank you very much vasel um one question uh from my side to Marion you showed quite I mean different forecasting models from different time Horizons theye Etc and does it make sense to put in place a specific type of forecast depending on the plant size more specifically to PV power plants I think what’s interesting mostly to look at is uh other grid requirements and sometimes the grid requirements depend on a plant size sometimes they don’t um and then what we want to focus on is um it’s more costly to put in place instruments on site obviously so perhaps very small sites um do not necessarily necessitate instruments um that being said it depends again on the application so if we’re you know talking about a small rooftop PV plant uh perhaps that’s not necessary to put in place a forecast based on instrument but if you look at um a reduced plant for offgrid um applications so not necessarily what we’ve looked at within trust PV but still um off grid sites are challenging because they’re not you know connected to the rest of the grid and in that case having in place um a forecast based on Sky cameras can be very interesting so a plant size does matter but I think what matters most is what is the application and what do you want to do with the forecast do you want to trade on the energy Market do you want to reduce your spinning Reserve do you want to um lessen your imbalance penalties Etc you know looking at the entire picture is really important all right uh thank you very much Marion um another question from Vaso from the chaso from the audience um for the balancing and frequency Services should these Services come batteries alone or should we place batteries at the PV sites for an effective participations on these markets uh that is um that is a very interesting question in general so from our experience we have seen that for example in balancing inary markets usually stand alone B assets can let’s say follow um in a Dr curve for example this a curve that uh correlates active power with frequency so they follow it effectively they can be auton autonomously integrated um in the stand alone let’s say assets connect NR and provide these responses we have also seen hybrid uh plants ER that are participating so both PV and battery and concurrently let’s say provide a set of services for example in markets such as for example H automatic uh generation control or any other dispatching and regulation set so both are viable choices depending and I think this depends initially when plant is sized on the on an initial let’s say business plan that is stipulate on on initial cost benefit analysis so the different uh sources of uh ER profit what can be sometimes batteries so this is this is um let’s say a conclusion a side conclusion H also for from studies that we have done within the frame of the research pro project that maybe in the future existing um PV sites given that uh let’s say um even the introduction of variability in the grid as they proliferate H they may need to do Advanced grid codes to add to retrofit battery in the PV and for example uh we may have a a requirement to connect the PV in the grid a requirement that the PV is totally dispatchable so it provides a certain response at certain times of the day so in the future maybe there will be certain requirements in order to connect the PV so maybe a battery will not only be an an option but it may be a an obligation a requirement in order to connect let’s say say so besides having a PV plant we will have a hybrid dispatchable H plant so these are this um um also this additional comments so here for example there are in this case there are certain also different architecture in terms of on how we connect the PV we have aced versus DC architecture so the battery may be coupled with the PV panels behind on the DC side of an inverter or we will have a we can have a PV plant and then a battery connected on the AC side the second provides more flexibility each one has certain benefits and suitable for the different business cases but all are viable and certain things that are options now may become obligations in the future given that the grid reaches certain grids depending on their um size and other characteristic reach their limits with hosting the variable energy produced by uh renewable such as PV and wind that are totally let’s say intermittent non dispatchable by Nature okay uh thank you very much uh it does make quite a lot of sense and since we are talking about batteries you showed about the operational cycles for B that they reduce it quite a lot and uh with higher higher penetration of rest do you foresee that these cycle time they will reduce reduce even further or we have actually reached a technical limit on that no for um it seems it seems that the the overall response times will be required to be more and more lower and lower for example what they mentioned power isolations if the the power isolations can be of 0.1 to to two h and a response from the uh from a PV PL should be maybe to effectively eliminate them should be the 1/10th of um h their cycle so we are talking about very um very strict overall response times that require reduction in the operation times in the control operation times but mostly working um Improvement create uh Improvement of the response times of the um inverters ER so yeah it seems that the direction is that uh uh faster and faster overall responses will be required all right um thank you very much Vaso unfortunately we are reaching the the end of our webinar um and I’ve seen there are some other questions from the audience but uh we’ll try to uh reply from email or so on um and I’d like to thank you everyone for participating in the webinar uh again thank solar power Europe for organizing the webinar and both Marion and Vaso for presenting uh thank you very much everyone I hope you have enjoyed the presentations and learned a little bit more as I did thank you thank you so thank you very much