Solar panel system and self-cleaning method

Description

2 environmental conditions. This leads to unnecessarily high water and energy consumption, even at times when cleaning is not strictly necessary. These automatic systems also do not take into account natural cleaning agents, such as rain, causing resources to be wasted and costs to increase.5 These shortcomings underscore the need for a more efficient and environmentally friendly system that can optimize the cleaning of solar panels and at the same time minimize the waste of resources and energy. 10 SUMMARY The goal of the present invention is to offer a solar panel system in which the cleaning of the solar panels is managed in an efficient and optimized manner,where unnecessary cleaning cycles and resource waste are prevented. The system15 aims for maximum energy yield by dynamically predicting the right moment for cleaning based on environmental and performance data. The present disclosure is defined in the attached independent conclusions. The independent conclusions define advantageous20 implementations. According to a first aspect of the disclosure, a computer-implemented method is offered to determine when a self-cleaning mechanism of a self-cleaning solar panel should be activated.25 The method comprises performing the following steps at control times: - obtaining values ​​of at least one performance parameter associated with an energy production efficiency of the solar panel,and where the values ​​of the performance parameter were obtained during the first time period preceding the control time; BE2024 / 5961 3 - obtaining environmental data measured during the said first time period preceding the control time and where the environmental data include, but are not limited to, one or more of the following environmental parameters: temperature, humidity, air pressure, wind speed, wind direction, precipitation amount, dust concentration; 5 - determining a future weather pattern based on the said environmental data and / or based on a weather forecast received from an external weather forecasting service, and where the future weather pattern includes a forecast of values ​​of one or more of the said environmental parameters for a second time period, following the first time period; - predicting a trend of the values ​​of the said performance parameter during the second time period,and where the prediction makes use of a machine learning model that is trained to establish relationships between the values ​​of the performance parameter and the values ​​of one or more environmental parameters; - based on the predicted values ​​of the performance parameter in the second time period, deciding whether cleaning is necessary within the second time period, and if cleaning is necessary, then determining a time tA within the second time period for activation of the mechanism for self-cleaning. By simultaneously using environmental data and a performance parameter of the solar panel associated with energy production efficiency, account can be taken advantageously, when determining a time for cleaning, of the energy production decrease due to pollution on the one hand and weather conditions, such as dust levels, wind precipitation, which can influence pollution negatively or positively on the other hand. The method accurately predicts the optimal cleaning time advantageously,whereby the solar panels can consistently function at their highest efficiency BE2024 / 5961 4 The advantage of the method according to the invention is that the decision not to clean immediately is not a decision based on a single actual measurement time, but on a future prediction of the expected performance of the solar panel over a longer period in the future, taking into account a future weather pattern. 5 For example, if the performance parameter has dropped at a certain moment,Cleaning does not necessarily need to be activated immediately if the weather pattern forecast indicates that the performance parameter will rise again as a result of, for example, a rain shower. In a cost-effective way, the number of cleaning cycles can remain limited while the performance of solar panels is maximized. In general, the decision as to whether cleaning is necessary or not and / or the determination of the time for activation of the self-cleaning mechanism can be based on a comparison between the values ​​of at least one of the predicted performance parameters and predetermined performance criteria. The method for determining when the self-cleaning mechanism of the self-cleaning solar panel must be activated is performed at each control time. In implementation forms, the control times are determined by a fixed control frequency, preferably 20, where the control frequency is equal to or greater than 1 / (TP2), with TP2 being the duration of the second time period. In other implementation forms,upcoming inspection times correspond to a time at which the value of at least one performance parameter of the solar panel or the value of another performance parameter of the solar panel has fallen below a predetermined threshold value. In implementation forms, the method further comprises: -predicting the effectiveness of cleaning in the second time period and where the effectiveness of cleaning is an expression of a relative difference between the performance of the solar panel before and after cleaning at a specific time, preferably where the effectiveness of cleaning BE2024 / 5961 5 is an expression of a relative difference between at least one performance parameter, or another performance parameter of the solar panel, after and before the cleaning of the solar panel. Preferably, for these forms of execution where cleaning effectiveness is also predicted, the method comprises: 5 - based on the predicted values ​​of the performance parameter in the second time period and based on the predicted effectiveness of cleaning in the second time period,deciding whether cleaning is necessary within the second time period, and if cleaning is necessary then determining a time point A within the second time period for activation of the 10 mechanism for self-cleaning. To predict the effectiveness of cleaning, use can be made of the said machine learning model that predicts the performance parameter and that is further trained to establish relationships between the values ​​of the effectiveness of cleaning and the values15 of the one or more environmental parameters. Alternatively, predicting the effectiveness of cleaning can make use of a second machine learning model that is trained to establish relationships between the values ​​of the effectiveness of cleaning and the values ​​of the one or more environmental parameters.20 In implementation forms,includes deciding whether or not to carry out cleaning during the second time period: -calculating an expected energy production during the second time period when no cleaning would be carried out and calculating an expected energy production during the second time period when cleaning would take place at a specific time within the second time period, and -if a gain in energy production with cleaning compared to the energy production without cleaning is less than a predetermined energy limit value, then deciding not to carry out cleaning during the second time period. BE2024 / 5961 6 According to a second aspect of the disclosure, a solar panel system is offered comprising: at least one solar panel; a self-cleaning mechanism for at least one solar panel, one or more panel sensors configured to measure at least one performance parameter associated with an energy production efficiency5 of the solar panel; environmental sensors configured to measure environmental data,and where the environmental data include, but are not limited to, one or more of the following environmental parameters: temperature, humidity, air pressure, wind speed, wind direction, precipitation amount, dust concentration; a data acquisition module coupled with the panel sensors and the environmental sensors and configured to measure the values ​​of at least one performance parameter and the values ​​of the environmental data as a function of time, and where the data acquisition module contains a storage medium to store the measured values; and a control system to control the self-cleaning mechanism and where the control system includes a processor configured to determine, using a machine learning model and based on at least one performance parameter and the environmental data, a time tA when the self-cleaning mechanism should be activated, and where the controller is configured to activate the self-cleaning mechanism at the specified time tA. The system can adapt to changing environmental and weather conditions in a cost-effective manner,such as dust storms or seasonal pollution, making it suitable for a wide range of locations and situations. Indeed, the ML model is continuously trained25 with the self-recorded local data, so that the ML model automatically adapts to the specific environmental conditions of the geographical area where solar panels are installed. The method and the system can be advantageously integrated into both large-scale commercial solar parks and residential30 installations, thereby serving a broad target group. BE2024 / 5961 7 Preferably, the self-cleaning mechanism is a vibration mechanism. This has the advantage that no water or chemical agents need to be used and the system can also be used in dry areas. In this way, the solar panel system is also fully autonomous, both regarding the decision regarding the timing of the cleaning5 and regarding the execution of the cleaning itself. In implementation forms,can the solar panel be provided with a nanocoating to repel dust and dirt. In implementation forms, the control system of the solar panel system can be coupled with an external weather forecasting service or an external weather forecasting station to obtain a future weather pattern, and where the future weather pattern includes a prediction of values ​​of one or more of the environmental parameters. In other implementation forms, the solar panel system can include a weather forecasting module configured to determine a future weather pattern, and where the future weather pattern includes a prediction of values ​​of one or more of the environmental parameters, preferably where the future weather pattern is determined based on the measured environmental parameters. In implementation forms, the machine learning model is one of, or a combination of, one of the following types: random forest,recurrent neural networks of gradient boosting. .BRIEF DESCRIPTION OF THE DRAWINGS The invention will be explained in more detail below with reference to drawings25 on the basis of examples of implementation forms. Fig. 1 is a schematic representation of an example of an implementation form of a solar panel system according to the present invention that comprises at least one solar panel with a self-cleaning mechanism and a control system to activate the self-cleaning mechanism at a time determined by a processor; BE2024 / 5961 8 Fig. 2 is a flowchart illustrating process steps of an implementation form of a method to determine when a self-cleaning mechanism of a self-cleaning solar panel must be activated according to the present invention; Fig. 3 is a schematic representation of an example of a timeline in which a decision is made repetitively at control points tc1, tc2, tc3, tc4, tc5 whether or not to perform a cleaning in a subsequent time period following the control point at a predicted time tA1, tA2,tA3; Fig. 4 is a flowchart that provides an overview of 10 process steps of a further implementation form of a method according to the present invention. The drawings are not drawn to scale or proportioned. In general, identical parts in the figures are indicated by the same reference numbers.15 DETAILED DESCRIPTION The present disclosure is described in terms of specific implementation forms that are illustrative of the disclosure and should not be construed as restrictive. It will be appreciated by persons skilled in the technology that the present disclosure is not limited to what has been specifically shown and / or described and that alternative or modified implementation forms can be developed in light of the general doctrine of this disclosure. The described drawings are merely schematic and25 non-restrictive. For the sake of clarity and a concise description, elements are described herein as part of the same or separate embodiments; however, it will be clear that the scope of the invention,execution forms can include combinations of all or some of the described elements. It should be understood that the BE2024 / 5961 9 shown execution forms have the same or similar elements, apart from where they are described as being different. The use of the verb "include" as well as the respective conjugations does not exclude the presence of elements other than those mentioned. The use of the article "a" or "the" preceding an element does not exclude the presence of a multitude of such similar elements. Moreover, the terms first, second and similar end descriptions in the conclusions are used to distinguish between similar elements and not necessarily to describe an order, whether in time, space, rank or any other way. It is understandable that the terms used in this way,be interchangeable under suitable conditions and that the forms of implementation of the disclosure described herein are capable of operation in series other than those described or illustrated herein.15 When reference is made in this specification to "one implementation" or "one form of implementation", this means that a certain function, feature, structure or characteristic described in connection with the implementations is contained in one or more implementations of this disclosure. Thus, occurrences of the phrases "in one form of implementation" or "in one form of implementation" at various places in this specification do not necessarily all refer to the same form of implementation,but could do that. Solar panel system25 With reference to Fig. 1, a schematic representation of an example of a design form of a solar panel system10 according to the present invention is presented. The solar panel system10 comprises one or more solar panels1 and each solar panel1 has a self-cleaning mechanism2 of the30 solar panel. BE2024 / 5961 10 In design forms, the self-cleaning mechanism2 may be a vibration mechanism, preferably an ultrasonic vibration mechanism. When the vibration mechanism is activated, contaminants are loosened from the surface of the solar panel without the need for water or mechanical brushes.5 In design forms, the solar panel1 in combination with, for example, a vibration mechanism2, may also include a nanocoating3 that is applied to the surface of the solar panel to repel dust and dirt. Denanocoating3 is, for example, a hydrophobic and oleophobic10 nanocoating. Preferably, this coating is UV-resistant, ensuring it remains effective throughout the lifespan of the solar panels,even with prolonged exposure to sunlight. The coating is also transparent and designed to have minimal influence on the light absorption efficiency of the underlying solar cells.15 For each solar panel, the solar panel system comprises one or more panel sensors4 configured to measure at least one performance parameter P1(t), P2(t),…Pi(t) associated with an energy production efficiency of the solar panel. These performance parameters are measured as a function of time; a time t20 is associated with each parameter value. In implementation forms, the one or more panel sensors4 can be integrated into the solar panel. Examples of performance parameters P1(t), P2(t),…Pi(t) are: a light transmission through the panel, a pollution level of the panel,25 an energy conversion efficiency of the panel and an energy production. An example of panel sensors are optical sensors that measure the intensity of the light passing through the surface of the solar panels and reaching the solar cells. With these sensors, therefore, the light transmission through the panel can be measured.and these light transmission is a measure of the pollution of the panels. BE2024 / 5961 11 Another example of panel sensors are sensors that perform a direct measurement of dust accumulation. These sensors can therefore directly measure the degree of pollution of the panels. The degree of pollution of the solar panels can also be derived from the energy conversion efficiency, which is a ratio between the electrical energy produced by the solar panel and the amount of solar energy that falls on the panel. The measurement of energy production can also be used as a performance parameter, but it is dependent on weather factors. The solar panel system further includes one or more environmental sensors configured for measuring environmental data O1(t), O2(t),…Oi(t). These environmental data are measured as a function of the times a time t is associated with each measured parameter value. Examples of ambient data O1(t), O2(t),… Oi(t) are one or more of the following ambient parameters: temperature, humidity, 15 air pressure, wind speed, wind direction,precipitation amounts and dust concentration. The dust concentration is, for example, the amount of PM2.5 particulate matter in the air or another parameter that indicates a dust concentration. The environmental parameters can depend on the weather conditions such as storm, rain, sun, snow, hail, sandstorm and any other possible weather condition that can occur in a specific geographical area. The environmental parameters can also depend on day or night, the time of year or the seasons. In implementation forms, humidity sensors can, for example, use capacitive or resistant elements to detect moisture content, and temperature sensors can, for example, use thermocouple soft thermostats to measure changes in the ambient temperature.or can air quality sensors make use of gas detectors to measure polluting particles in the air.30 BE2024 / 5961 12 The solar panel system further includes a data acquisition module6 which is coupled with the panel sensors4 and the ambient sensors5. The data acquisition module is configured to measure the values ​​of one or more performance parameters and the values ​​of the ambient data as a function of time. This measurement is preferably performed in real-time. The measured values ​​are then stored in a storage medium7. In implementation form, the storage medium is computer memory, such as RAM memory. The solar panel system10 further includes a control system8 to control the self-cleaning mechanism2. In particular, the10 control system8 can activate the self-cleaning mechanism2 to start a self-cleaning cycle. For this purpose, the control system8 includes a processor9, also called a central processing unit,that is configured to execute a software program that calculates the time tA when the self-cleaning mechanism 2 must be activated.15 The software algorithm uses a machine learning model, ML model, to make a prediction of a time tA when the self-cleaning mechanism must be activated during a second time period TP2 following the first time period TP1. The ML model uses as input data at least one 20 performance parameter, and environmental data measured during the first time period and / or predicted environmental data for the second time period. In implementation forms, the control system can obtain weather information from an external weather report. For example, the control system of the solar panel system can be connected, preferably wirelessly 25, to an external weather forecasting service or an external weather forecasting station, in order to obtain a future weather pattern in this way,and where the future weather pattern includes a prediction of values ​​of one or more of the mentioned environmental parameters. Examples of weather forecasting services are OpenWeatherMap, AccuWeather, or a regional weather service. This external service provides data such as expected rainfall, wind speed, and humidity for specific locations. This weather information is then passed to the machine learning model as input data, as discussed further below. In execution forms, the processor can execute a weather forecasting algorithm that makes an extrapolation based on historical weather datasets in order to determine a future weather pattern. For example, the system can analyze trends in the measured rainfall of the past few days and extrapolate these into the future. In doing so, the time of year can be taken into account, for example, rainy seasons,and earlier patterns. The advantage of this algorithm10 based on extrapolation is that one is not dependent on external services. The disadvantage is that it can be less accurate during sudden weather changes. In implementation forms, the solar panel system can have its own weather forecasting module comprising weather forecasts based15 on locally measured environmental parameters, such as temperature, air pressure, wind speed, wind direction and humidity. In this way, a future weather pattern is obtained,and where the future weather pattern includes a prediction of values ​​of one or more of the mentioned environmental parameters.20 The weather forecasting module can, for example, make use of the same or an additional ML model to predict weather changes. A drop in air pressure can, for example, indicate an approaching rain area. Wind speeds and directions can be combined with seasonal trends to predict rainfall. Furthermore, the time of year can also be taken into account to model seasonal influences. The advantage of using a proprietary weather forecasting module is that the system is completely independent, locally optimized, and can be very specific to the location of the solar panels.30 BE2024 / 5961 14 In implementation forms,the software algorithm uses the ML model or a second ML model to predict cleaning effectiveness, and where this prediction of cleaning effectiveness co-determines the time tA for activation of the self-cleaning mechanism. Cleaning effectiveness is an expression of a relative difference between the performance of the solar panel before and after cleaning, preferably where cleaning effectiveness is an expression of a relative difference between the performance parameter after and before cleaning the solar panel. The control system then activates the self-cleaning mechanism with an activation signal at time tA determined by the algorithm, as schematically shown in Fig. 1. In implementation forms, the control system includes an internal clock, such as a real-time clock,to keep track of time. The computer-implemented method executed by the software algorithm is discussed in more detail below. Computer-implemented method With reference to Fig. 2, a flowchart is shown with a number of process steps of an example of an execution form of a method to determine when a self-cleaning mechanism of a self-cleaning solar panel must be activated. The method with the various process steps is executed repetitively at certain control times, etc. In execution forms, the control times are determined by a fixed control frequency and the control is thus performed each time after the lapse of a certain predetermined period. The frequency of control is usually a configuration value. Depending on the execution form, the control can, for example, daily,be carried out weekly or every fixed period.30 BE2024 / 5961 15 In other forms of execution, the inspection time c may correspond to a time at which a performance parameter of the solar panel falls below a predetermined threshold value. In this implementation, the inspection is still carried out repetitively at time t c but the time duration between two consecutive inspections is variable. Fig. 3 shows a schematic representation of an example of a timeline in which a check is performed repetitively at control points tc1, tc2, tc3, tc4, tc5 and a decision is made whether or not to perform a cleaning in a subsequent time period following the control point at a well-defined predicted time tA1, tA2, tA3. For example, at time point tc1 a check is performed based on data obtained in the time period TP1 before time point tc1 and it is decided whether in a subsequent period TP2, after time point tc1,whether or not a cleaning is necessary. In this example, it was decided to perform a cleaning at time tA1 in the period following the control time tc1. At time tc2, a second check is performed based on data obtained from the time period TP1 prior to time tc2, which in this example was the period TP2 prior to control moment tc1. After performing a check at time tc2, it was decided not to perform a cleaning in the period following time tc2. For control times tc3 and c4, in this example, it was decided to perform a cleaning at the predicted times tA2 and A3 respectively, which follow time tc3 and c4 respectively. In implementation forms, the repetitive execution of the cleaning control can take place at a certain frequency equal to or greater than 1 / (TP2), with TP2 being the duration of the second time period. A first process step is shown schematically in Fig. 2.relates to the data entry a second process step200 relates to the processing of the input data and / or the determination of derived characteristics. Depending on the form of execution this can be considered as one or as two steps. BE2024 / 5961 16 In a process step100 the data from the panel sensors P1(t), P2(t),…Pi(t) and the data O1(t), O2(t),…Oi(t) from the environmental sensors are read in. These values ​​from the panel sensors and environmental sensors are therefore values ​​measured before the time tc at which the check is performed. The values ​​were, for example, obtained in a first time period5 TP1 before the time tc of the check. The method according to the present invention comprises obtaining values ​​of at least one performance parameter P1(t) associated with an energy production efficiency of the solar panel, and where the values ​​of the performance parameter are obtained during the first 10 time period TP1 prior to the check time tc. As discussed above,Performance parameters that can directly or indirectly detect the fouling of the solar panels are, for example: light transmission through the panel, a degree of fouling of the panel, an energy conversion efficiency of the panel, and energy production. The method according to the present invention further comprises obtaining environmental data measured during the said first time period TP1 prior to the control time c. These environmental data include, but are not limited to, one or more of the following environmental parameters: temperature, humidity, air pressure, wind speed, precipitation amount, dust concentration. In implementation forms, the data from the panel sensors and the environmental sensors may possibly be raw data. Depending on the implementation forms,can your data first be converted into a 25 adapted form that is usable for reading into the ML model. This processing of the input data can optionally be performed in step 200. Optionally, derived characteristics can also be determined in step 200. For example, optionally, based on a read performance parameter P1(t) of the solar panel, such as the light transmission through the panels, a solar panel characteristic dP1 / dt BE2024 / 5961 17 can be determined, such as the characteristic of the rate of dirt build-up on the panels. This dirt rate dP1 / dt can be derived from the change in light transmission. Such derived characteristics can also be considered as additional performance parameters. In execution forms, time-related characteristics can also be determined based on an internal clock of the control system or the processor, such as: season,day or night. These time-related characteristics can then be associated with the input data. The method further comprises determining a future weather pattern based on the mentioned environmental data and / or based on a weather forecast received from an external weather forecasting service or external weather station. The weather pattern comprises a forecast of values ​​VO1(t'), VO2(t'), …VOi(t') of one or more of the mentioned environmental parameters for a second time period TP2,following the first time period TP1. The 'time' refers to a time in the second time period TP2. In implementation forms, the environmental data are measured during a first time period TP1 of, for example, a week and a forecast is made of the weather pattern for the second time period TP2 of, for example, a second week following the first week. In other implementation forms, the first and second periods may be shorter or longer. Process tap 300 relates to the use of the ML model to predict the performance of the solar panel.25 As input data, the ML model uses at least one performance parameter associated with an energy production efficiency of the solar panel as discussed above and the predicted weather pattern for time period TP2 following a control moment. The ML model can also use additional parameters as input,30 such as for example a solar panel characteristic, such as for example the rate BE2024 / 5961 18 of pollution build-up,derived based on at least one performance parameter or based on a second or more performance parameters obtained from panel sensors. As mentioned above, the ML model can also use time-related parameters as input data such as: season, day-night.5 The ML model further uses as input data the predicted weather pattern for the second period TP2 after the time tc of the control recording. As discussed above, the predicted weather pattern comprises a prediction of values ​​VO1(t'), VO2(t'),…VOi(t') of one or more of the environmental parameters. As a result of these environmental parameters,will possibly change the 10 performance parameter of the solar panel in the second time period TP2. As discussed above, the prediction of the environmental parameters for the future period TP2 can either be obtained from an external weather forecasting service or the future environmental parameters can be determined with an internal 15 weather forecasting module. The method according to the present invention involves predicting a further course of the values ​​of the performance parameter during the second time period TP2, after the time at which the check is performed. For this prediction, use is made of the ML model 20 which is trained to establish relationships between the values ​​of the performance parameter and the values ​​of one or more environmental parameters. The method further comprises deciding, based on the predicted values ​​of the performance parameter P1(t') in the second time period TP2, whether a cleaning is necessary within the second time period TP2. The time t' refers to a time in the future, after the time tc at which the check is performed. Subsequently, if a cleaning is necessary then,In process step 400, the method involves determining a time point tA for performing a cleaning during the second time period.30 BE2024 / 5961 19 In other words, the decision to perform a cleaning in the future, after the time of control tc, is not directly based on the actual experimental performance data taken in the period before the control time tc, but is based on predicted future performance data. This allows the decision to be based on a future 5 evolution of the performance of the solar panels over a longer period and where the evolution depends on the future weather pattern. For deciding whether or not to perform a cleaning based on the predicted values ​​P1(t') of the performance parameter in the second period after the time point tc of control, there are several possibilities,10 depending on the chosen implementation forms. These different possibilities for deciding whether cleaning is necessary or not and / or determining the time for activation of the self-cleaning mechanism generally occur on the basis of a comparison between one of the at least one predicted performance parameters15 and predetermined performance criteria. In an implementation form, the decision to perform cleaning or not can be made by comparing the predicted values ​​of the performance parameter with predetermined performance criteria. If the performance criteria are met, it is decided not to activate the self-cleaning mechanism during the second time period. If the performance criteria are not met, a time G is determined that lies within the second time period at which the performance criteria are no longer met. The determination of the time A for activation of the mechanism25 F2?I6=7B6:?:8:?8<2?52?2=CF@=8DE:D865BE <DG@B56?)D+*D-"KD#G22B3:; KD66?3:;<@> 6?56A6B:@56:C>6DKDJ&%,63:;<@>6?56A6B:@56KD<2? be considered as a delay period to provide some extra time between determining the decay of the performance parameter below certain limits and the time of the actual execution of the cleaning.30 In execution forms is A=tG. BE2024 / 5961 20 .?E:DF@6B:?8CF@B>6?<2?56E:DCD6=A6B:@56KD66?F@@B2736A22=56 4@?7:8EB2D:6G22B56I:;?6?:?2?56B6E:DF@6B:?8CF@B>6?<2?KD36A22=5 be based on the weather pattern. For example, if the weather pattern indicates that large amounts of dust are present in the second time period, and the energy production of the solar panels is consequently zero or very low, it may still be decided to temporarily postpone the cleaning, taking into account the expected weather pattern, despite the fact that the solar panels are dirty and urgently need cleaning. ,6E:DCD6=A6B:@56KD<2?@@<863BE:<DG@B56?@> 56B6:?:8:?8E:DD6 stellentotnavoorbeeldzonsondergang,or can be waited until the 10 period arrives where expected energy production is low. In implementation forms, the performance criteria to determine whether cleaning is necessary may include one or more lower limits for the performance parameter, preferably where the lower limits are configuration data.15 In implementation forms, the determination of the time tA to carry out the cleaning can take place in two steps. In the first step, it is checked at what moment G the predicted values ​​of the performance parameter no longer meet the predetermined performance criteria. For example, if the light transmission through the solar panel falls below a first predetermined limit, it can then be decided that an activation of the cleaning mechanism must take place when 66?D:;5CD:AD+JD-#G22B3:;D-96DD:;5CD:A:CG22B@A56F@@B2736A22=56 first limit is reached. In a second step, the exact time for D+36A22=5G@B56?G22B3:;D+*D-"KD%25 ,636A2=:?8F2?KD<2?@AF6BC49:==6?56>2?:6B6?8636EB6?# depending on the form of execution. In one form of execution, the method comprises predicting the effectiveness of cleaning the solar panel in the second time period 0 / '6?<2?KD36A22=5G@B56?@A32C:CF2?66?F@@BCA6=5667764D:F:D6:DF2?30 cleaning. BE2024 / 5961 21 The effectiveness of cleaning, as discussed above, indicates the extent to which the cleaning improves the performance of the solar panel. The effectiveness of cleaning is therefore an expression of a relative difference between the performance of the solar panel before and after cleaning. In implementation forms, the effectiveness of cleaning can be an expression of a relative difference between performance parameters before and after the cleaning of the solar panel. For example, the performance parameter P1 may have fallen back from an initial value of 100% to 80% over time. When cleaning is then performed, the performance parameter P1 may have risen back to, for example, 95%. In this case, the effectiveness of the cleaning is EF=(95%-80%) / (100%-80%)=0.75. In another example,If after cleaning the performance parameter were to return to 100%, then the effectiveness EF=1. In general terms, the effectiveness of cleaning can be expressed as the difference between the value of the performance parameter after and before cleaning divided by the difference between a maximum achievable value of the performance parameter and the value of the performance parameter before cleaning. In implementation forms, the effectiveness of cleaning has a value between 0 and 1, where a value of 1 indicates that the value of the performance parameter of the solar panel after cleaning will be maximum due to cleaning, and where a value of 0 indicates that after cleaning the value of the performance parameter will not have changed compared to the value before cleaning. The effectiveness of cleaning has a direct influence on the energy output of the solar panels. With an effectiveness of 1, the maximum energy output of the solar panel can be achieved again. By, for example, measuring the energy output of the solar panel before and after cleaning,can the effectiveness of the cleaning be measured. The effectiveness of cleaning can depend on various factors, such as weather conditions. For example, the effectiveness of a BE2024 / 5961 22 cleaning carried out in the middle of a sandstorm will be lower than the effectiveness obtained when the cleaning is carried out on a sunny day. ,636A2=:?8F2?KD<2?E:D86F@6B5G@B56?@A32C:CF2?56D6 expect effectiveness of cleaning at the exact time G the evolution of the effectiveness of cleaning after the exact time G. If the effectiveness of cleaning at a time after the time G is greater than the effectiveness of cleaning at the exact time G, then the cleaning can be postponed by a period of time KDI@52D@A96D@86?3=:<52D56B6:?:8:?8G@B5DE:D86F@6B5#5667764D:F:D6:D of cleaning is greater than at the moment G. The expected effectiveness of cleaning can be predicted with 10 the same or a second, additional,ML model. This ML model is trained to establish relationships between the effectiveness of cleaning and the future environmental parameters determined by a weather pattern. The environmental parameters are the parameters as discussed above. In execution forms, the algorithm can be configured to only carry out a cleaning at a time when the effectiveness of cleaning is above a predetermined minimum effectiveness threshold. After all, if the effectiveness is very low, it makes little sense to perform a cleaning. In execution forms, the algorithm can compare the predicted performance parameter with a second threshold value and, as long as the performance parameter ?:6D@?56B56DG66568B6?CF2=D<2?56B6:?:8:?8>6D66?D:;5KDG@B56? deferred if the effectiveness of cleaning is higher at the later time. If the predicted performance parameter rightly falls below the second limit,it may be decided to carry out the cleaning regardless of the expected effectiveness of cleaning. This is to avoid the energy production with the solar panel becoming extremely low. In other forms of execution, another method D@686A2CDG@B56?@>KDD636A2=6?#5@@B)$96D36B6<6?6?F2?66??:6EG6 value for the performance parameter after a fictitious cleaning would have been performed at time tG, - predicting values ​​of the BE2024 / 5961 23 performance parameter for a third time period after the fictitious cleaning, - calculating a rate of change of the performance parameter D:;56?C5656B56D:;5CA6B:@56#6?$96D36A2=6?F2?56E:DCD6=A6B:@56KD@A basis of the rate of change of the performance parameter in the third time period. In forms of execution, predicting values5 of the performance parameter for the third time period after the first fictitious cleaning use the ML model. 16= <6> 6D9@56@@<863BE:<DG@B5D@> KDD636A2=6?#2=C96D D:;5CD:AD+*D-"KD%36B6:<D:C#I2=96D4@?DB@=6CHCD66> 66?24D:F2D:6C:8?22= "#generateattimetA,determined by the algorithm, to activate the self-cleaning mechanism, as schematically shown in Fig. 1. In implementation forms, the decision whether or not to perform a cleanup can be based on calculating an expected energy production during the second time period when no cleanup would be performed and calculating an expected energy production during the second time period when a cleanup would take place at a specific time within the second time period. If a gain in energy production with cleanup compared to the energy production without cleanup is smaller than a predetermined energy limit value,it may be decided not to carry out cleaning during the second time period. If the gain in energy production resulting from cleaning is higher than or equal to the energy limit value, cleaning is therefore carried out in the second time period, and the time for cleaning can, for example, be determined as the time at which maximum energy production is achieved over the second time period. The calculation of the expected energy production in the second time period depends on the predicted trend of the values ​​of the performance parameter during the second time period and / or on the predicted effectiveness of cleaning during the second time period. BE2024 / 5961 24 In forms of execution, if during the repetitive execution of the cleaning control the frequency of control is such that after the execution of a first control at time tc1, a new control is carried out at, for example, time tc2 and this time tc2 is prior to the time for cleaning tA1 as determined during the earlier control tc1,and thus at moment tc2 the predicted cleaning at time tA1 has not yet been performed, then the old time tA1 for cleaning is overwritten and replaced by a new predicted time tA2 for cleaning as determined at moment tc2. This situation can occur if the time period TP2 is, for example, a time period of one week and the frequency of checking and determining the time tA for cleaning occurs, for example, daily. With reference to Fig. 4, a schematic flowchart is shown which provides an overview of process steps of a further implementation of a method according to the present invention. Process step 500 corresponds to the activation of the mechanism for self-cleaning at a time as determined by the algorithm. The invention uses an ML model to predict the performance parameter and / or to predict the effectiveness of a cleaning and this ML model can be selected from different types,depending on the specific implementation and the conditions in which the solar panel system is applied. Some examples of suitable ML models that are known and can be selected by the professional are: RandomForest (RF), Recurrent Neural Networks (RNN), or Gradient Boosting (GB). RandomForest is an ensemble learning method that combines multiple decision trees to provide a robust prediction. This model is particularly suitable for datasets with complex, non-linear relationships between input data, such as pollution levels, weather conditions, and solar panel performance. RNNs are primarily designed to process time-dependent data and recognize trends across consecutive data points. This model BE2024 / 5961 25 is particularly suitable for predicting changes in performance parameters over time, such as seasonal pollution or daily fluctuations in weather conditions. This type is more suitable for the analysis of long-term trends,such as, for example, the build-up of pollution during dry seasons.5 Gradient Boosting is an ensemble method that iteratively improves weak models to create a strong model. It is particularly effective in addressing small errors in earlier predictions, resulting in high accuracy. For example, optimizing the prediction of cleaning effectiveness by correcting errors in earlier predictions. In implementation forms, the ML model can use a combination of different techniques, also known as ensemble learning. This can increase the accuracy and robustness of the model and algorithm. For example, a combination of RF, RNN, and GB can be applied. A combined approach utilizes the strengths of multiple models to make a better prediction. There are two known ways to combine models, namely “stacking,” where the models work in parallel and their predictions are combined by a meta-model,and “hybrid modeling”, where different models are used in a series and where the output of one model is the input for the next. In a combined approach, RF25 can be used in a form of implementation to determine which parameters contribute most to, for example, pollution, and an initial estimate can be made. Subsequently, RNN can be applied to predict the future development of pollution levels based on historical trends and weather forecasts. Finally, GB can also be used to further refine the 30 BE2024 / 5961 26 forecast and correct small errors in the earlier models. An example scenario of a combined approach could be the following. For example, suppose the input is the following: a light transmission of 85%, a temperature of 30°, a rain forecast of 10 mm within 24 hours,and that historical pollution trends indicate an average decrease of 2% light transmission per day. In a first phase, RF and RNN can then be applied, and RF can, for example, estimate that the pollution will reach a critical threshold within 2 days, and RNN can predict that light transmission will improve by 5%, but not fully recover. Subsequently, in a second phase, GB can be applied and, based on the RF and RNN outputs, predict that, for example, cleaning is necessary in 36 hours. The ML model can be continuously trained based on historical data, namely based on the real performance parameters and real corresponding environmental factors that are continuously measured. The ML model for self-cleaning solar panels can therefore be set up as a so-called supervised learning model, whereby the system learns based on historical data regarding pollution, weather conditions,solar panel performance before and after cleaning cycles.20 In implementation forms, a new training dataset for the ML model can be defined where input and output data for the new training dataset are based on the values ​​of the performance parameter and environmental data measured during the first time period TP1 before the control time and the second time period TP2 after the control time, respectively. In this way, the ML model is continuously trained based on the actual measured performance of the solar panel and the actual measured environmental factors. This feedback to the ML model, which results in continuous training, is schematically shown in Fig. 4 in step 700.30 BE2024 / 5961 27 In implementation forms, as discussed above, the determination of the timing of the cleaning execution can depend on a predicted cleaning effectiveness. In these implementation forms, as schematically shown,