Method for generating artificial intelligence model for predicting solar energy-generated power, solar energy-generated power prediction system using artificial intelligence model, and solar energy generation device failure diagnosis system using artificial intelligence model

Abstract

Provided are a method for generating an artificial intelligence model for predicting solar-energy generated power, wherein generated power is further effectively predicted through effective machine learning, a solar energy-generated power prediction system using the artificial intelligence model, and a solar energy generation device failure system using the artificial intelligence model. The method for generating an artificial intelligence model for predicting solar-energy generated power comprises the steps of: collecting, by using a GPS module, a set of measurement times based on GPS time, a set of actual measurement data of a solar power plant, and a set of environmental data of an installation area of the solar power plant; generating a set of prediction data by using a physical model by configuring an equivalent physical model for calculating generated power of the solar power plant; comparing values of all data included in the set of prediction data and the set of actual measurement data for each measurement time, and then generating a reduced training data set only with a pair of prediction data and actual measurement data having the magnitude of a relative error less than or equal to a reference value, and environmental data having a measurement time coinciding with the pair; and generating a trained artificial intelligence model for predicting the amount of generated power of the solar power plant on the basis of environmental variables including the solar radiation quantity and the atmospheric temperature by means of machine learning using the reduced training data set.

Description

A method for generating an artificial intelligence model for solar power generation power prediction, a solar power generation power prediction system using an artificial intelligence model, and a solar power generation device fault diagnosis system using an artificial intelligence model.

[0001] The present invention relates to a technology for predicting the power generation of a solar power plant using an AI model, and more specifically, to a method for generating an artificial intelligence model for predicting solar power generation, a solar power prediction system using such an AI model, and a solar power generation device failure system using such an AI model.

[0002] Solar power generation is a technology that directly generates electricity from light, applying the photovoltaic effect or photoelectric conversion principle. While it might seem theoretically feasible to ensure a steady supply of power with sufficient light, actual power production often deviates from expectations. For example, solar modules or panels are configured as arrays to maximize the power produced by solar cells. However, power generation can vary depending on factors such as array arrangement (tilt angle, azimuth, etc.), installation topography (steep slope, flat land, etc.), and the surrounding environment (temperature, irradiance, altitude, etc.).

[0003] In particular, solar power plants are being installed in diverse locations, and the potential for additional power generation benefits, such as with bifacial solar modules, makes estimating power generation performance increasingly difficult. Furthermore, for economic reasons, installation sites are often built according to terrain without leveling, or even at inclinations on mountainous areas. This increases the number of additional variables that make power generation difficult to predict, further complicating power generation performance estimation. Differences in module tilt and azimuth can lead to power generation losses due to differences in array surface irradiance. Even when irradiance is measured, accurate estimation of power generation can be difficult.

[0004] Furthermore, this leads to inaccurate assessments of solar power generation capacity, making it difficult to determine whether individual arrays or modules are operating normally or whether some or all of them are malfunctioning. This problem poses challenges both in the installation and operation of solar power plants, which are becoming increasingly important due to demands for carbon neutrality and other factors. Therefore, a solution is needed.

[0005] [Prior Art Literature]

[0006] [Patent Document]

[0007] (Patent Document 1) Republic of Korea Patent Publication No. 10-2024-0008653, (January 19, 2024)

[0008] The technical task of the present invention is to solve such problems, and to more accurately predict the power generated by a solar power plant by applying an artificial intelligence model, and in particular, to provide a method for generating an artificial intelligence model for solar power generation prediction that enables more accurate prediction by improving the accuracy of power data used in machine learning, a solar power generation power prediction system using such an artificial intelligence model, and a solar power generation device failure system using such an artificial intelligence model.

[0009] The technical problems of the present invention are not limited to the problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

[0010] The method for generating an artificial intelligence model for predicting solar power generation according to the present invention comprises the steps of: (a) collecting a set of measurement times based on GPS time using a GPS module, a set of actual data including current, voltage, and power of a solar power plant for each measurement time, and a set of environmental data including solar irradiance and atmospheric temperature of an installation area of ​​the solar power plant for each measurement time; (b) constructing an equivalent physical model for calculating the power generation of the solar power plant using the environmental data as an input variable, and generating a set of predicted data including predicted current, predicted voltage, and predicted power of the solar power plant for each measurement time matching the set of environmental data from the physical model; (c) comparing the values ​​of all data included in the set of predicted data and the set of actual data for each measurement time, and generating a reduced learning data set including only pairs of the predicted data and the actual data having a relative error less than or equal to a reference value, and the environmental data having a matching measurement time with the pair; And (d) a step of generating a trained artificial intelligence model that predicts the power generation amount of the solar power plant based on environmental variables including solar irradiance and atmospheric temperature through machine learning using the reduced learning data set.

[0011] The above step (c) can generate a ratio of the predicted data and the actual data for each measurement time, and determine the degree to which the ratio deviates from 1 as the relative error.

[0012] In the above step (c), the size of the learning data set can be adjusted by changing the reference value.

[0013] In the above step (c), the environmental data that falls short of the standard solar radiation and the pair of predicted data and actual data that match the measurement time can be removed from the learning data set.

[0014] The above standard solar radiation may be 300W / ㎡.

[0015] The physical model of step (b) may include an air temperature-surface temperature conversion model that calculates the surface temperature of the solar module applied to the solar power plant from the air temperature and the solar irradiance.

[0016] In the above step (d), the machine learning algorithm includes a supervised learning algorithm, and the supervised learning algorithm may include a regression algorithm.

[0017] A solar power generation power prediction system using an artificial intelligence model according to the present invention comprises: a GPS module; a first data collection unit that collects a set of measurement times based on GPS time using the GPS module, a set of actual data including current, voltage, and power of a solar power plant for each measurement time, and a set of environmental data including irradiance and air temperature of an installation area of ​​the solar power plant for each measurement time; a physical model unit that constructs an equivalent physical model for calculating power generation of the solar power plant using the environmental data as an input variable, and generates a set of predicted data including predicted current, predicted voltage, and predicted power of the solar power plant for each measurement time that matches the set of environmental data from the physical model; a learning data set generation unit that compares the values ​​of all data included in the set of predicted data and the set of actual data for each measurement time, and then generates a reduced learning data set including only pairs of the predicted data and the actual data having a relative error less than or equal to a reference value, and the environmental data having a matching measurement time with the pair; A first artificial intelligence model generation unit that generates a first artificial intelligence model for predicting the power generation of the solar power plant based on environmental variables including solar irradiance and atmospheric temperature through machine learning using the reduced learning data set; and a first model operation unit that applies the first artificial intelligence model to predict the power generation of the solar power plant based on the environmental variables.

[0018] The above first data collection unit can synchronize the local time of the sensor or external measuring station providing the environmental data and the local time of the electrical equipment of the solar power plant providing the actual measurement data with the measurement time based on the GPS time.

[0019] The above learning data set generation unit can generate a ratio of the predicted data and the actual data for each measurement time, and then determine the degree to which the ratio deviates from 1 as the relative error.

[0020] The reduced learning data set may be one in which the environmental data that falls short of the standard solar irradiance and the pairs of the predicted data and the actual measured data that match the measurement time are removed.

[0021] The above physical model may include an air temperature-surface temperature conversion model that calculates the surface temperature of a solar module applied to the solar power plant from the air temperature and the solar irradiance.

[0022] A fault diagnosis system for a solar power generation device using an artificial intelligence model according to the present invention comprises: a solar power generation power prediction system using the artificial intelligence model; a second data collection unit that collects a set of inference data regarding the generated power of a solar power plant, which is an output of the solar power generation power prediction system, a set of actual generated power data provided from the solar power plant at the time of generating the inference data, and a set of operation information including operation information regarding the normal and fault states of the solar power plant at the time of generating the inference data provided from the solar power plant; a second artificial intelligence model generation unit that generates a second artificial intelligence model that determines the normal and fault states of the solar power plant based on a ratio of the inference data and the actual generated power data through machine learning using a learning data set composed of the set of inference data, the set of actual generated power data, and the set of operation information; And it includes a second model operation unit that determines the normal and fault status of the solar power plant by applying the second artificial intelligence model and inputting the inference data and the power generation measurement data.

[0023] The machine learning algorithm of the second artificial intelligence model generation unit includes a supervised learning algorithm, and the supervised learning algorithm may include a regression and classification algorithm.

[0024] The above second artificial intelligence model can determine normal and fault states from the location of the vector space by mapping the three ratios of each of the inferred current, inferred voltage, and inferred power included in the inferred data and each of the actual current, actual voltage, and actual power included in the generated power measurement data as independent components into a three-dimensional vector space.

[0025] According to the present invention, changes in power generation due to changes in environmental conditions (e.g., solar irradiance) of solar power generation facilities, solar power plants, or solar modules can be more accurately calculated through prediction using artificial intelligence. In particular, an AI model for power generation prediction can be trained with more accurate learning data (e.g., extracted power and environmental data that eliminates error-prone data to ensure more accurate inference), thereby improving inference performance. This allows for an AI model to predict changes in solar power generation due to environmental conditions with greater accuracy. Furthermore, by generating and providing customized AI models based on data collected from each facility, more accurate power predictions that reflect regional characteristics are possible. Furthermore, based on the accurate inferences of such AI models, power generation calculations and evaluations can be more accurately performed, enabling faster and more accurate identification of failures and / or malfunctions in solar power generation facilities or equipment. Therefore, utilizing the present invention, more accurate and efficient operation of solar power facilities is possible without errors.

[0026] FIG. 1 is a flowchart illustrating a method for generating an artificial intelligence model for solar power generation prediction according to one embodiment of the present invention.

[0027] Figure 2 is a block diagram of a solar power generation power prediction system using an artificial intelligence model according to one embodiment of the present invention.

[0028] Figure 3 is an example of a physical model that illustrates the physical model of solar power generation as an equivalent circuit.

[0029] Figures 4 to 7 are diagrams illustrating a method of producing predicted data using the physical model of Figure 3.

[0030] Figures 8 to 10 are conceptual diagrams illustrating the principle of generating a learning data set according to the present invention and graphs showing the data distribution of a learning data set generated therefrom.

[0031] Figure 11 is a block diagram of a solar power generation device fault diagnosis system using an artificial intelligence model according to one embodiment of the present invention.

[0032] Figure 12 is a conceptual diagram illustrating the fault diagnosis principle of the fault diagnosis system of Figure 11.

[0033] The advantages and features of the present invention, as well as the methods for achieving them, will become clearer with reference to the embodiments described in detail below together with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. These embodiments are provided only to ensure that the disclosure of the present invention is complete and to fully inform those skilled in the art of the scope of the invention, and the present invention is defined solely by the claims. Like reference numerals designate like elements throughout the specification.

[0034] Hereinafter, with reference to FIGS. 1 to 12, a method for generating an artificial intelligence model for solar power generation prediction, a solar power generation power prediction system using the artificial intelligence model, and a solar power generation device fault diagnosis system using the artificial intelligence model according to the present invention will be described in detail. Since the present invention utilizes an artificial intelligence model, the method for generating an artificial intelligence model will first be described in detail, and then, based on the method, a power prediction system and a fault diagnosis system using the artificial intelligence model will be described in sequence.

[0035] FIG. 1 is a flowchart illustrating a method for generating an artificial intelligence model for solar power generation prediction according to one embodiment of the present invention.

[0036] Referring to FIG. 1, the method for generating an artificial intelligence model for predicting solar power generation according to the present invention generates a valid learning data set using actual data of power generation measured from a solar power plant, environmental data regarding solar irradiance and temperature around the solar power plant at the time of generating the actual data, and predicted data, which is power generation theoretically predicted from the environmental data. If the learning data is invalid when generating an artificial intelligence model, noise is learned, resulting in a decrease in inference performance. Therefore, by generating a learning data set that provides a more accurate correlation (linear and / or nonlinear) through data preprocessing and then performing machine learning, the performance of the artificial intelligence model can be improved without changing the characteristics of the model itself. The present invention removes the time error of data based on GPS time, compares the actual measured value (actual measured data) and the predicted value predicted through theory (predicted data - calculated from a physical model) under the same conditions (or at the same time), and configures a learning data set using only data in which the theoretical value and the actual measured value match within the error range, so that the prediction accuracy of the artificial intelligence model can be drastically improved through machine learning of more effective learning data.

[0037] Furthermore, because the training data includes both observed and theoretical predictions, it simultaneously captures the accuracy of theoretical predictions and hidden real-world variables not accounted for in theory, allowing predictions to reflect both accuracy and flexibility. Therefore, it is possible to implement an AI model that combines the advantages of both theoretical and reality-based predictions: higher adaptability to reality than predictions based solely on theory, and higher theoretical correlation than predictions based solely on observed data.

[0038] The artificial intelligence model generation method for solar power generation power prediction of the present invention (hereinafter, the artificial intelligence model generation method) is configured as follows. The artificial intelligence model generation method is as follows.

[0039] A step [(a) step-S100] of collecting a set of measurement times based on GPS time using a GPS module, a set of actual measurement data including current, voltage, and power of a solar power plant for each measurement time, and a set of environmental data including solar irradiance and air temperature of the solar power plant installation area for each measurement time.

[0040] A step [(b) step-S200] of constructing an equivalent physical model that calculates the power generation of a solar power plant using environmental data as input variables, and generating a set of predicted data including the predicted current, predicted voltage, and predicted power amount of the solar power plant for each measurement time matching the set of environmental data from the physical model.

[0041] A step [(c)-S300] of generating a reduced learning data set by comparing the values ​​of all data included in the set of predicted data and the set of actual data by measurement time, and only the pairs of predicted data and actual data whose relative error size is less than or equal to a reference value, and the environmental data whose measurement time matches the pair, and

[0042] It includes a step [(d)-S400] of generating a trained artificial intelligence model that predicts the power generation of a solar power plant based on environmental variables including solar irradiance and air temperature through machine learning using a reduced learning data set.

[0043] These steps can be performed using the following system: For example, the system may include a GPS module (see 100 in FIG. 2 - time information generation based on GPS time), a first data collection unit (see 200 in FIG. 2 - basic data collection including a set of measurement times, a set of actual data for each measurement time, and a set of environmental data for each measurement time), a physical model unit (see 300 in FIG. 2 - generation of a set of predicted data for each measurement time based on a physical model), a learning data set generation unit (see 400 in FIG. 2 - data reclassification based on relative error after collating predicted data and actual data), a first artificial intelligence model generation unit (see 500 in FIG. 2 - generation of an artificial intelligence model for power generation prediction by machine learning of reduced learning data), and a first model operation unit (see 600 in FIG. 2 - application of the generated artificial intelligence model to power generation prediction at the present time), and all of the remaining components excluding the GPS module may perform one or more of the above steps, calculations and / or data processing and / or It can be formed with a computer device (including a storage device) capable of collecting data.

[0044] For example, the first data collection unit may be formed by including a storage device capable of inputting / outputting data and storing it, and the physical model unit may include a simulation program for producing theoretical predictions, etc. The learning data set generation unit may include a data processing program capable of data collection and comparative analysis, and the first artificial intelligence module generation unit may be implemented in a form capable of distributed and / or parallel computing by configuring one or more computer devices in parallel to improve machine learning performance. The first model operation unit may include a data processing program for inputting / outputting data in conjunction with the learned artificial intelligence model. Since a recording medium storing such a program may also be included in a computer device, each component may be operated on a computer. In this way, the above steps can be performed by a computer system in which one or more computer devices are organically combined. The solar power generation power prediction system described below (see 10 in FIG. 2) is a specific example of such a system, and therefore, a specific description of the system will refer to the solar power generation power prediction system.

[0045] Below, each step of the AI ​​model creation method is described in more detail. The explanation proceeds based on the flowchart in Figure 1, with reference to other drawings.

[0046] Referring to Fig. 1, first, as a basic data collection step, a step is performed to collect a set of measurement times based on GPS time using a GPS module, a set of actual data including current, voltage, and power consumption of a solar power plant for each measurement time, and a set of environmental data including solar irradiance and air temperature of the solar power plant installation area for each measurement time (S100). This step can be performed using a GPS module (see 100 of Fig. 2) and a first data collection unit (see 200 of Fig. 2) that collects basic data as described above.

[0047] The measurement time can be provided from the GPS module (100). For example, the GPS signal provided by the GPS module (100) can be received, and the first data collection unit (200) can generate and collect the measurement time based on the GPS time at regular intervals. The measurement time can be a time standard for matching data (actual data, environmental data, and predicted data described below) to the same point in time based on the GPS time, and can be a standard time for synchronizing the time of the data by synchronizing the different local times of the data.

[0048] The GPS signals received by the GPS module from satellites provide GPS time, which can compensate for time errors based on location for accurate distance measurements. Therefore, data collected from different locations can be synchronized to a single time based on the measurement time based on GPS time. There are various time synchronization methods using GPS signals, so any known method can be utilized.

[0049] For example, even if there is a difference between the local time of electrical equipment (e.g., inverter, etc.) that generates actual measurement data (current, voltage, power) at a solar power plant and the local time of a measuring device (e.g., sensor) that generates environmental data (irradiance, temperature) at or around the solar power plant, the difference can be corrected by converting the time information of each data (local time - which can be provided in the form of a timestamp for each data) to a measurement time based on GPS time. The predicted data described below can also be synchronized based on this measurement time, thereby eliminating errors that may occur due to differences in the time standards between data.

[0050] The present invention compares actual measured data under the same conditions (irradiance, temperature) with predicted data predicted by a physical model (learning data set generation step-S300 described below) to construct a more efficient learning data set, so the smaller the time error of the compared data, the more desirable it is. The data of the present invention are collected and classified based on the measurement time based on GPS time with extremely high accuracy, so more accurate data comparison at the same time is possible. The method for generating a learning data set through data comparison is described in detail below.

[0051] In this way, by synchronizing the time based on the measurement time, all other necessary basic data can be collected. That is, a set of measurement times, a set of actual measured data including the current, voltage, and power consumption of the solar power plant for each measurement time, and a set of environmental data including the amount of solar radiation and air temperature of the solar power plant installation area for each measurement time can be collected by a data collection device such as the first data collection unit (see 200 in Fig. 2).

[0052] At this time, the actual measurement data may be data that reflects the amount of power generation that varies depending on the environmental conditions (irradiance, temperature, etc.) of the solar power plant for the purpose of power prediction, such as the actual values ​​of current, voltage, and power amount measured at the solar power plant. The actual measurement data is continuously measured at the solar power plant and may be provided from electrical equipment (e.g., inverters) that output power. The environmental data is data that reflects environmental conditions that directly affect solar power generation, such as irradiance and air temperature, and may be provided from sensors (light intensity sensors, temperature sensors, etc.) and / or measuring stations installed at and / or around the solar power plant. In this way, a set of measurement times based on GPS time, a set of actual measurement data (current, voltage, power amount of the solar power plant) for each measurement time, and a set of environmental data (irradiance and air temperature, etc.) of the solar power installation area for each measurement time can be collected.

[0053] Afterwards, an equivalent physical model for calculating the power generation of a solar power plant using environmental data (insolation, atmospheric temperature, etc.) as input variables is constructed, and a step is performed to generate a set of predicted data including the predicted current, predicted voltage, and predicted power of the solar power plant for each measurement time that matches the set of environmental data from the physical model (S200). This step can be performed using the aforementioned physical model unit (see 300 in Fig. 2). A more detailed description of the physical model is as follows.

[0054] Fig. 3 is an example of a physical model that illustrates a physical model of solar power generation as an equivalent circuit, and Figs. 4 to 7 are diagrams illustrating a method of producing predicted data using the physical model of Fig. 3.

[0055] A physical model can be a physical and / or mathematical model that theoretically predicts the power generation or power output of a solar power plant. The inputs of the final physical model are environmental data (solar irradiance, atmospheric temperature), and the outputs are the current, voltage, and power values ​​(predicted current, predicted voltage, and predicted power) expected to be produced by the solar power plant under given conditions. Therefore, the physical model can theoretically calculate the power generated by a solar power plant based on environmental conditions (solar irradiance and atmospheric temperature). This predicted data is compared with previously collected measured data to form a learning data set, as described below. Specifically, the physical model can be structured as follows.

[0056] The physical model is based on the fact that a solar power plant is based on solar cells, and that solar cells can be interpreted as a diode model. For example, a solar power plant is formed by an array of multiple solar modules, and each solar module is in turn composed of multiple solar cells, so a solar module can be modeled as a collection of solar cells. Since a solar cell generates photocurrent from a PN junction formed by the joining of a P-type semiconductor and an N-type semiconductor, a single solar cell can be interpreted as an equivalent circuit including a diode, the simplest semiconductor. Figure 3 (a) shows the equivalent circuit of a unit solar cell interpreted as a one-diode model including a single diode and series and parallel resistors. [Here, I ph is the photocurrent, V d is the diode voltage, R sh is the parallel resistance, R s represents series resistance]

[0057] Referring to Fig. 3(b), the solar module can be modeled as a circuit in which a plurality of equivalent circuits (i.e., unit cells) of Fig. 3(a) are connected in series. At this time, it is assumed that all unit cells are equal [i.e., I at the top of Fig. 3(b)]. ph1 =I ph2 =...=I phN =I ph , I d1 =I d2 =...=I dN =I d , I sh1 =I sh2 =...=I shN =I sh And, in the middle V d1 =V d2 =...=V dN =V d , R sh1 =R sh2 =...=R shN =R sh , R s1 =R s2 =...=R sN =R s Therefore, NR at the bottom s(=N×R s ), NV d (=N×V d ), NR sh (=N×R sh ) (where N is the number of solar cells) is derived], the circuit topology can be finally converted into the lower equivalent circuit of Fig. 3 (b). [In Fig. 3 (b), I ph1 , I ph2 , I phN are the photocurrents of the 1st, 2nd, and Nth solar cells, respectively, and I d1 , I d2 , I dN V is the diode current of the 1st, 2nd, and Nth solar cells, respectively. d1 , V d2 , V dN are the diode voltages of the 1st, 2nd, and Nth solar cells, respectively, and R sh1 , R sh2 , R shN I is the parallel resistance of the 1st, 2nd, and Nth solar cells, respectively. sh1 ,I sh2 , I shN R is the parallel resistance current of the 1st, 2nd, and Nth solar cells, respectively. s1 , R s2 , R sN [I and V represent the series resistance of the 1st, 2nd, and Nth solar cells, respectively, and the output current and output voltage, respectively]

[0058] In the final equivalent circuit model [bottom of Fig. 3(b)], V (output voltage) = NV d -I(output current)×NR s Considering this point, the diode voltage (V d ) into the known Shockley diode equation, the output voltage equation of the solar module can be derived as follows.

[0059]

[0060] <Mathematical Formula 1>

[0061]

[0062] Also, in the final equivalent circuit model [bottom of Fig. 3(b)], V(output voltage) = I sh ×NR sh -I(output current)×NR s Considering this point, the diode current (I d ) can be substituted into the Shockley diode equation, and the output current equation for the solar module can be derived as follows.

[0063]

[0064] <Mathematical Formula 2>

[0065]

[0066] [In the above mathematical equations 1 and 2, q is the electron charge, k is the Boltzmann constant, T is the temperature (absolute temperature), n is the ideal factor, I0 is the saturation current, I ph is the photocurrent, V d is the diode voltage, I d is the diode current, R sh is the parallel resistance, R s is the series resistance, and NR s =N×R s , NV d =N×V d , NR sh =N×R sh , (where N is the number of solar cells)]

[0067]

[0068] From the mathematical model of the solar module, a curve showing the relationship between the output voltage (V) and the output current (I) as in Fig. 4 can be obtained. At this time, the photocurrent (I) of the relationship ph) can be approximated by the short-circuit current when the module voltage is 0, and as is known, the short-circuit current varies depending on the irradiance [e.g., it can be calculated as the ratio of the variable irradiance to the reference irradiance, where the reference irradiance can be 1000 W / ㎡, which is the irradiance of the STC (Standard Test Condition)], curves in which the maximum power point changes depending on the irradiance change can be simulated as in Fig. 5 [in Fig. 5, each curve is an IV curve for a different irradiance]. By repeating this simulation, the irradiance and voltage (V) of the solar module as in Figs. 6 and 7 mp: maximum voltage) or irradiance and current (I mp: The correlation curve of the maximum current can be obtained, and by curve fitting it, the relationship between the irradiance and the voltage of the solar module, the relationship between the irradiance and the current of the solar module, and the equation for the power (product of current and voltage) can be derived as follows.

[0069]

[0070] <Mathematical Formula 3>

[0071]

[0072] <Mathematical Formula 4>

[0073]

[0074] <Mathematical Formula 5>

[0075]

[0076] [In the above mathematical equations 3, 4, and 5, V mp_module Silver module maximum voltage, I mp_module Silver module maximum current, P mp_module is the maximum power of the module, a v , b v , c v is the voltage coefficient, I r refers to the amount of solar radiation (solar intensity: W / ㎡)]

[0077]

[0078] At this time, considering that the band gap of the solar cell is affected by temperature, the equations for the maximum voltage and maximum current of the module can be corrected again as follows.

[0079]

[0080] <Mathematical Formula 6>

[0081]

[0082] <Mathematical Formula 7>

[0083]

[0084] [In the above mathematical equations 6 and 7, T m is the module temperature, a tv is the voltage temperature coefficient, a ti means the current temperature coefficient, and the rest is the same as mathematical expressions 3 and 5]

[0085]

[0086] Therefore, knowing the number of modules connected in series and in parallel, we can calculate the voltage and current of the entire array as follows. Of course, power is calculated as the product of voltage and current.

[0087]

[0088] <Mathematical Formula 8>

[0089]

[0090] <Equation 9>

[0091]

[0092] [In the above mathematical equations 8 and 9, V mp_array is the total array voltage, I mp_arry is the total current of the array, Array sr is the number of modules connected in series, Array pr refers to the number of modules connected in parallel, and the rest is the same as mathematical formulas 6 and 7]

[0093]

[0094] However, in this modeling, the temperature of the module (T m) is directly considered, the module temperature can be converted into an equation for the air temperature and solar irradiance by applying the following air temperature-surface temperature conversion model.

[0095]

[0096] <Mathematical Formula 10>

[0097]

[0098] [In the above mathematical expression 10, T mod is the module temperature, T amb is the atmospheric temperature, K Ross is the coefficient, E poA refers to the solar irradiance (measured value). Reference: E.Skoplaki, JA Palyvos, Operating temperature of photovoltaic modules: A survey of pertinent correlations, Renewable Energy, Vol. 34 (2009) 23-29]

[0099]

[0100] That is, the physical model is the surface temperature (T) of the solar module applied to the solar power plant. m ) may also include an air temperature-surface temperature conversion model that calculates the air temperature and solar radiation. Therefore, the power generation of a solar module can be calculated more effectively based on environmental variables such as air temperature. However, the air temperature-surface temperature conversion model need not be limited to an example, so if there is another valid model, other models (or other correlations based on them) can be applied.

[0101] Therefore, using the above-described physical model and the mathematical formula derived from it, the current, voltage, and power output from the array of the target solar power plant can be calculated as a function of irradiance and temperature (atmospheric temperature). The coefficients during modeling can be adjusted according to the characteristics of the solar power system, etc. In this way, by constructing an equivalent physical model that calculates the power produced by the solar power plant (especially current, voltage, and power) using environmental data (irradiance and atmospheric temperature) as input variables, a set of predicted data including the predicted current, predicted voltage, and predicted power of the solar power plant for each measurement time that matches the set of environmental data from the physical model can be generated.

[0102] That is, given the solar irradiance and atmospheric temperature, the predicted voltage of the solar power plant (solar array) (e.g., V in Equation 8) is predicted from the solar irradiance and atmospheric temperature from the physical model (e.g., Equations 8, 9, and 10 above). mp_array ), predicted current (e.g., I in Equation 9 mp_array ) and predicted power (voltage × current) can be calculated, so that the predicted current, predicted voltage, and predicted power for each measurement time can all be calculated for each solar irradiance and atmospheric temperature collected in the basic data collection step [(a) step-S100]. In this way, a more valid learning data set can be generated by comparing the set of predicted data predicted from the physical model with the collected actual data. The set of predicted data thus produced can be stored in the physical model unit or the first data collection unit described above and classified by measurement time.

[0103] It should be understood that physical models are not necessarily limited and can be adjusted, such as by modifying and / or altering at least some of them, if necessary. Furthermore, it is not impossible to construct one or more physical models and selectively and / or complementarily utilize them to generate predictive data. Therefore, the method of generating predictive data using physical models is not limited to these examples.

[0104] Thereafter, after comparing the values ​​of all data included in the set of predicted data and the set of actual data by measurement time, a step is performed to generate a reduced learning data set consisting only of pairs of predicted data and actual data whose relative error is less than or equal to a reference value, and environmental data whose measurement time matches the pair (S300). This step can be performed by the learning data set generation unit described above (see 400 in FIG. 2).

[0105] In this step (learning data set creation step), the data collected in the previous step are compared by measurement time and errors are determined. In particular, environmental data (insolation and air temperature) are collected by measurement time, actual measured data (current, voltage, power consumption) which is the power production amount of the solar power plant by measurement time are also collected, and predicted data (predicted current, predicted voltage, predicted power consumption) which is the predicted power production amount of the solar power plant by measurement time are also collected, so each data can be matched and compared at the same time using the measurement time as a parameter.

[0106] This kind of comparison is essentially the same as comparing the measured power production (current, voltage, and power values ​​at the same measurement time) with the predicted power production (predicted current, predicted voltage, and predicted power value at the same measurement time) under specific environmental conditions (irradiance and air temperature at the specific measurement time) using the measurement time as a medium. In other words, even if the environmental conditions are the same, there may be differences between the theoretical predictions (predictions based on physical models) and the actual measurements (actual measurements). Therefore, by comparing them with each other, data with relatively large errors between theory and practice can be removed from the training data, effectively preventing the degradation of artificial intelligence model performance due to noise. More specifically, this step can generate a ratio of the predicted data to the actual data for each measurement time, determine the degree to which the ratio deviates from 1 as the relative error, and extract only the data whose relative error is within a reference value.

[0107] Figure 8 summarizes the process of creating a learning data set as a block diagram. Referring to the drawing, the collected basic data set [Data Load - Measurement Time (Date), Insolation (I r ) and ambient temperature (Temp) (i.e. environmental data), voltage (V dc ), current (I dc ) and power (P ac )(i.e., actual data)] among environmental data [solar radiation (I r ) and atmospheric temperature (Temp)] are applied to the physics model to produce predicted data [predicted voltage (V est ), predicted current (I est ), predicted power (P est ) - Here, V est is the aforementioned V mp_array Wow, I est is the aforementioned I mp_array Wow P est is used to produce [which can be practically equal to their product], and the produced predicted data (e.g., V est , I est ) is the actual data (e.g., V) by measurement time. dc , I dc ) are matched with each other (e.g., V dc / V est , I dc / I est ) is generated. That is, for each measurement time, the voltage ratio (V) of the actual data and the predicted data dc / V est ) and the current ratio (I) of the measured data and predicted data dc / I est ) to generate all of these ratios (V dc / V est and I dc / I est ) is different from 1 [e.g., 1 and voltage ratio (V dc / V est ) absolute value of the difference, 1 and current ratio (I dc / I est) can be determined by the relative error and only data with a relative error less than or equal to the standard value can be extracted (Data Extraction) to create a reduced learning data set.

[0108] For example, the reference value that gives the limit of relative error can be set within 0.5. That is, when the limit of relative error is set to an absolute value of 0.5, the ratio of predicted data to actual data (e.g., V dc / V est , I dc / I est ) is at least 1±0.5, so that a more valid set of learning data can be created before extraction. As the reference value approaches 0, the degree of agreement between the predicted data and the measured data increases, but the number of learning data may decrease. Therefore, it is highly desirable to appropriately adjust the range of the reference value to extract learning data of an appropriate size. In other words, the size of the extracted learning data set can be adjusted by changing the reference value that sets the relative error limit between the predicted data and the measured data in the learning data set generation step.

[0109] More preferably, the reference value can be 0.1, in which case the ratio of predicted data to actual data (e.g., V dc / V est , I dc / I est ) is in the range of 1±0.1, so the consistency between the predicted data extracted as learning data and the actual data increases further. That is, as in the example of Fig. 8, the voltage ratio (V dc / V est ) and current ratio (I dc / I est ) By extracting the actual data and predicted data with a high degree of mutual agreement between 0.9 and 1.1 (the boundary value may or may not be included as needed), a reduced learning data set can be created using only pairs of such predicted data and actual data and environmental data whose measurement times match the pairs.

[0110] In other words, the reduced training data set includes pairs of predicted and measured data with relative errors below a threshold, along with environmental data corresponding to their causes. Therefore, changes in environmental data, including solar irradiance and air temperature, can be effectively learned from both measured and predicted data, resulting in changes in power generation.

[0111] At this time, additionally, environmental data (irradiance and atmospheric temperature by measurement time) that is less than the standard irradiance and pairs of predicted and actual data whose measurement time matches can be removed from the learning data set. In other words, data at a time when the amount of light is low and the power generation does not reach an appropriate level are likely to not effectively reflect the correlation between irradiance and power generation, so by additionally removing data based on irradiance, more reliable learning data can be generated. At this time, the standard irradiance can be 300 W / ㎡, and therefore, in the final learning data set, pairs of environmental data with irradiance less than 300 W / ㎡ (including irradiance less than 300 W / ㎡ and atmospheric temperature at the same time) and predicted and actual data whose measurement time matches can also be removed.

[0112] The reduced training data set through these steps more effectively represents the correlation between irradiance and solar power generation current, and irradiance and solar power generation voltage for an appropriate irradiance (e.g., a reference irradiance of 300 W / ㎡ or more), as shown in Fig. 10. Fig. 10 is a data plot between irradiance and current, and irradiance and voltage of the training data set reduced through the data extraction process illustrated in Fig. 8. Comparing this with the data plot between irradiance and current, and irradiance and voltage of the raw data before processing in Fig. 9, it can be confirmed that noise is extremely effectively reduced.

[0113] That is, by examining the consistency between the predicted data predicted from the physical model for specific environmental conditions (solar irradiance and atmospheric temperature) and the actual measured data, and reducing the data based on the consistency between the theoretical and actual values ​​(relative errors being below the standard value), a noise-removed learning data set [Fig. 10] can be extracted that significantly reduces the inference error. By machine learning the reduced learning data set in this way, an artificial intelligence model that can very accurately predict solar power generation according to changes in environmental conditions such as solar irradiance can be created.

[0114] Next, a step is performed to create a trained artificial intelligence model that predicts the power generation of a solar power plant based on environmental variables, including solar irradiance and air temperature, through machine learning using the reduced training data set (S400). This step can be performed using the first artificial intelligence model generation unit (see 500 in FIG. 2) described above.

[0115] Since artificial intelligence can learn data in different ways depending on the algorithm, it is desirable to use an algorithm more suitable for predicting solar power generation. In this embodiment, the machine learning algorithm includes a supervised learning algorithm, and the supervised learning algorithm may include a regression algorithm. Of course, deep learning using artificial neural networks can also be applied, so an artificial intelligence model can be created using various algorithms that can learn linear or nonlinear relationships and predict output. The learning data set of the present invention includes environmental data including irradiance and atmospheric temperature, actual solar power generation data (current, voltage, power consumption) corresponding to the data, and corresponding physical model prediction data (predicted current, predicted voltage, predicted power consumption). However, noise is extremely effectively removed through preprocessing that increases the consistency between the actual data and the predicted data, so that a trained artificial intelligence model can be created that infers generated power (current, voltage, power consumption, etc.) from environmental variables (irradiance, atmospheric temperature, etc.) with higher accuracy through regression learning. The artificial intelligence model generated in this way (i.e., the artificial intelligence model for solar power generation power prediction, which is the same as the first artificial intelligence model described below) can more accurately infer solar power generation power that changes according to environmental conditions (solar irradiance, atmospheric temperature, etc.).

[0116] Hereinafter, with reference to FIG. 2, a solar power generation power prediction system using an artificial intelligence model according to the present invention will be described in detail. This system is a system for generating an artificial intelligence model for predicting solar power generation and using it for actual inference, and is substantially identical to the system for generating the aforementioned artificial intelligence model. Therefore, this system will be described in connection with each step of the aforementioned artificial intelligence model generation method, and since the contents of the aforementioned artificial intelligence model generation method are equally applicable to this system, a repeated explanation thereof will be omitted. The following will focus on explaining aspects not described.

[0117] Referring to FIG. 2, the solar power generation power prediction system (10) (hereinafter, power prediction system) using the artificial intelligence model according to the present invention comprises: a GPS module (100), a first data collection unit (200) that collects a set of measurement times (as described above) based on GPS time using the GPS module, a set of actual data including current, voltage, and power of the solar power plant for each measurement time (as described above), and a set of environmental data including solar irradiance and air temperature of the solar power plant installation area for each measurement time (as described above); a physical model unit (300) that constructs an equivalent physical model (as described above) that calculates the power generation of the solar power plant using the environmental data as an input variable, and generates a set of predicted data (as described above) including predicted current, predicted voltage, and predicted power of the solar power plant for each measurement time that matches the set of environmental data from the physical model; and a processor that compares the values ​​of all data included in the set of predicted data and the set of actual data for each measurement time. Then, the learning data set generation unit (400) that generates a set of learning data (same as described above) reduced to only pairs of predicted data and actual data whose relative error is less than or equal to a reference value and environmental data whose measurement time matches the pairs, a first artificial intelligence model generation unit (500) that generates a first artificial intelligence model (same as an artificial intelligence model for solar power generation power prediction) that predicts the power generation of a solar power plant based on environmental variables including irradiance and atmospheric turbidity through machine learning using the reduced learning data set, and a first model operation unit (600) that predicts the power generation of a solar power plant based on environmental variables by applying the first artificial intelligence model.

[0118] Since the system of the present invention is structured with a focus on functional aspects, the physical form of the system or the classification of the installed programs may not necessarily match the configuration. Each configuration described below may be formed by installing one or more computer programs for implementing each function on a computer device, and each computer device may include one or more programs or sets of programs that can perform the functions of each configuration described below. One or more computer devices may be used. Within such limits, the physical form of the system (i.e., the form of the computer device or the installed programs) can be changed in any way, and therefore, it is not necessary to understand it in a limited manner. The configuration of the power prediction system (10) will be described in more detail as follows.

[0119] The GPS module (100) is configured to receive GPS signals and transmit GPS time included in the GPS signals. The GPS module can receive GPS signals from satellites and may be configured to enable satellite communication for this purpose. The GPS module (100) may be installed in a solar power plant (A), but need not be installed in such a manner. It may be placed in various locations where it can be connected to the system via a wired and / or wireless connection to transmit information including GPS time. Of course, if there is a pre-installed GPS module, it is also possible to utilize that module. The GPS module (100) may be capable of signal analysis on its own, and in such a case, it may be integrated with a computer device capable of signal analysis.

[0120] The first data collection unit (200) can collect data from a GPS module (100), a solar power plant (A), a sensor (not shown) measuring environmental data, and / or a measuring station (B). The first data collection unit (200) can include a storage device capable of inputting / outputting data and a program for operating the storage device, and can also have a time synchronization function capable of time conversion. The first data collection unit (200) generates and collects measurement times (e.g., GPS time or universal time synchronized based on GPS time, etc.) at regular intervals based on the GPS time provided by the GPS module (100), and synchronizes the local time of the remaining collected data [actual measurement data (current, voltage, power) provided from the solar power plant and environmental data (irradiance, air temperature) provided from the measuring station, etc.] based on the measurement time to set a time standard. That is, the first data collection unit (200) can synchronize the local time of the sensor or external measuring station providing environmental data and the local time of the electrical equipment (e.g., inverter, etc.) of the solar power plant providing actual data through time synchronization to the measurement time based on GPS time. The first data collection unit (200) collects a set of measurement times based on GPS time, a set of actual measurement data including the current, voltage, and power amount of the solar power plant (A) for each measurement time, and a set of environmental data including the amount of solar radiation and air temperature of the solar power plant installation area for each measurement time.

[0121] At this time, the measuring station (B) provides environmental data (irradiance, atmospheric temperature, etc.) and may be located in the same area as and / or adjacent to the solar power plant (A). The measuring station (B) includes measuring devices that can measure environmental information such as irradiance and temperature, and may be installed together with the solar power plant (A). However, it does not necessarily have to be the same, and if possible, data may be provided from public measuring stations that provide atmospheric information, weather information, etc. in a location adjacent to the power plant, so the measuring station needs to be understood in a broad sense that includes public measuring stations, etc. In addition, it is possible to deploy and use sensors (including measuring devices) that can measure irradiance, atmospheric temperature, etc. in the solar power plant (A) and / or the surrounding area, so the application of sensors need not be ruled out. In other words, environmental data may be provided from sensors (not shown) and / or measuring stations, and the sensors or measuring stations need not be limited within the scope of their ability to provide environmental data.

[0122] The physical model unit (300) may have a function of implementing a physical model (as described above) that calculates the power generation of a solar power plant from environmental data (irradiance, atmospheric temperature, etc.) and producing output (predicted data) therefrom. The physical model unit (300) may include, for example, a computer simulator and / or a simulation program that outputs predicted data (predicted current, predicted voltage, predicted power as described above) based on the physical model. The physical model unit (300) may receive environmental data, which is an input of the physical model, through data communication with the first data collection unit (200), and may also have a data storage to store simulation results. The physical model may include an air temperature-surface temperature conversion model that calculates the surface temperature of a solar module applied to a solar power plant from the air temperature and irradiance, as described above, to improve its performance (as described above). In this way, the physical model unit (300) constructs an equivalent physical model that calculates the power generation of a solar power plant using environmental data as input variables, and can generate a set of predicted data including the predicted current, predicted voltage, and predicted power amount of the solar power plant for each measurement time that matches the set of environmental data from the physical model. The physical model can be modified and changed as needed.

[0123] The learning data set generation unit (400) may have functions such as processing and handling data collected through data comparison and classification. The learning data set generation unit (400) may receive a set of predicted data, a set of measured times, a set of actual data, and a set of environmental data collected through data communication with the physical model unit (300) and the first data collection unit (200), and compare them based on the measurement time. The learning data set generation unit (400) may be formed by a data processing program having such a function and / or a computer device including such a processing program. In particular, since the time information of the data is synchronized based on the measurement time, the learning data set generation unit (400) can compare the data by measurement time, determine the error between the actual value and the predicted value therefrom, and extract learning data only from data with a relatively small error.

[0124] That is, as described above, the learning data set generation unit (400) can determine the degree of agreement by comparing the predicted data with the measured data, and can determine the degree to which the ratio deviates from 1 as a relative error after generating a ratio of the predicted data and the measured data by measurement time (as described above). In addition, as described above, environmental data that falls short of the standard irradiance and pairs of predicted data and measured data whose measurement time matches the environmental data can be removed from the learning data set, and data that fall short of the standard irradiance can be excluded from the reduced learning data set (as described above). In this way, the learning data set generation unit (400) can compare the values ​​of all data included in the predicted data set and the measured data set by measurement time, and can generate a reduced learning data set using only pairs of predicted data and measured data whose relative error size is less than or equal to a reference value, and environmental data whose measurement time matches the pair.

[0125] The first artificial intelligence model generation unit (500) performs machine learning based on the learning data set generated by the learning data set generation unit (400). The first artificial intelligence model generation unit (500) may be formed by a computer device capable of machine learning, and may also be implemented in a form capable of distributed and / or parallel computing, such as one or more computer devices configured in parallel to improve machine learning performance. The first artificial intelligence model generation unit (500) may, for example, generate a trained artificial intelligence model by training an initialized artificial intelligence model with a learning data set. At this time, the machine learning algorithm includes a supervised learning algorithm, and the supervised learning algorithm includes a regression algorithm to generate a trained artificial intelligence model (first artificial intelligence model) that predicts the power generation amount of a solar power plant (inference data described later) by taking environmental data (irradiance, atmospheric temperature) as input (as described above). In this way, the first artificial intelligence model generation unit (500) can generate a first artificial intelligence model (i.e., the artificial intelligence model for solar power generation power prediction described above) that predicts the power generation amount of a solar power plant based on environmental variables including solar irradiance and atmospheric temperature through machine learning using a reduced learning data set.

[0126] The first model operation unit (600) may be linked to the generated first artificial intelligence model (an artificial intelligence model for solar power generation power prediction generated by machine learning of a learning data set) and may have a function of inputting and outputting data. The first model operation unit (600) may include a type of data processing auxiliary program that provides data for inference to the first artificial intelligence model or receives output data (inference data) and outputs it to the outside. The first model operation unit (600) may receive current environmental data (solar irradiance, atmospheric temperature) from the first data collection unit (200) and / or the aforementioned measuring station (B), input the data into the generated first artificial intelligence model, and output inference data (inference current, inference voltage, inference power) to the outside as an output result. If necessary, the first model operation unit (600) may also be integrated with the first artificial intelligence model generation unit (500). In this way, by applying the first artificial intelligence model to the inference using the first model operation unit (600), the power generation (inference data) of a solar power plant can be predicted based on environmental variables (solar radiation, atmospheric temperature, etc.).

[0127] The power prediction system (10) of the present invention is designed to improve the accuracy of an artificial intelligence model by performing machine learning by constructing a reduced learning data set as described above. That is, the first artificial intelligence model (artificial intelligence model for solar power generation prediction) generated by the first artificial intelligence model generation unit (500) is the same as the trained artificial intelligence model generated according to the artificial intelligence model generation method described above (see FIG. 1), and predicts output by learning a linear or nonlinear relationship between environmental data (irradiance, atmospheric temperature) and solar power generation (power, voltage, power amount). The inference data, which is the predicted result from the first artificial intelligence model, includes current prediction (inferred current), voltage prediction (inferred voltage), and power prediction (inferred power amount), and each of these is inferred based on the machine learning described above regarding the power generation of a solar power plant according to irradiance and atmospheric temperature, so the accuracy is greatly improved by the effectiveness of the learning data set. That is, as described above, the learning data set of the present invention includes environmental data including irradiance and atmospheric temperature, actual solar power generation data (current, voltage, power), and predicted solar power generation data (predicted current, predicted voltage, predicted power) of a physical model corresponding thereto, and noise is also very effectively removed through preprocessing that increases the consistency between the actual data and the predicted data, so that more accurate inference data (inferred current, inferred voltage, inferred power) can be output by the first artificial intelligence model trained with data that reflects the advantages of both theoretical and actual values. With such a power prediction system (10), the solar power generation that changes according to irradiance, temperature, etc. can be more accurately predicted by utilizing the artificial intelligence model.

[0128] Hereinafter, with reference to FIGS. 11 and 12, a solar power generation device fault diagnosis system using an artificial intelligence model according to the present invention will be described in detail. The solar power generation device fault diagnosis system using an artificial intelligence model of the present invention includes, as a part thereof, the solar power generation power prediction system (10 in FIG. 2) described above in order to utilize the artificial intelligence model for power prediction (first artificial intelligence model) of the aforementioned system. Therefore, in the following description, repeated explanations of the aforementioned matters will be omitted, and the added parts will be described with emphasis.

[0129] FIG. 11 is a block diagram of a solar power generation device fault diagnosis system using an artificial intelligence model according to one embodiment of the present invention, and FIG. 12 is a conceptual diagram illustrating the fault diagnosis principle of the fault diagnosis system of FIG. 11.

[0130] Referring to FIG. 11, the solar power generation device fault diagnosis system (20) (hereinafter, “fault diagnosis system”) using the artificial intelligence model according to the present invention is configured in a form in which a second data collection unit (700), a second artificial intelligence model generation unit (800), and a second model operation unit (900) are added to the aforementioned power prediction system (10). As illustrated in FIG. 11, the configuration of the power prediction system (10) is substantially the same as the aforementioned configuration, and therefore, the description thereof will be replaced with the aforementioned description.

[0131] The second data collection unit (700) collects a set of inference data (as described above) regarding the power generated by a solar power plant, which is the output of the solar power generation power prediction system (10), a set of measured power generation data provided from the solar power plant (A) at the time of generating the inference data, and a set of operation information including operation information regarding the normal and fault states of the solar power plant at the time of generating the inference data provided from the solar power plant. The second data collection unit (700) may include a storage device capable of inputting / outputting data and a program for operating the storage device, and may also have a time synchronization function capable of time conversion like the first data collection unit (200). That is, the second data collection unit (700) may be understood in accordance with the first data collection unit, except for the difference in the type of collected data.

[0132] The second data collection unit (700) collects the inference data (inferred current, inferred voltage, inferred power) output by the first artificial intelligence model (artificial intelligence model for solar power generation prediction) of the aforementioned power prediction system (10). In addition, it collects a set of actual power generation data (actual current, actual voltage, actual power) measured from the solar power plant (A) at the same time as the inference data generation time and the operation information including the normal and fault status of the solar power plant (A) (e.g., signal or code information that can determine whether the entire power plant is fixed and / or operating normally) together. The actual power generation data may be substantially the same as the actual measurement data described above, but there is a difference in what is collected by the second data collection unit (700). An example of the operation information is information indicating whether it is normal or faulty detected or recorded throughout the entire power plant, and may include status information recorded in the power plant system and designated code information that classifies the status.

[0133] A more specific example of operational information is as follows: Operational information is recorded in the power plant system, for example, Code 0 - Normal, Code 1 - PID (Potential Induced Degradation), Code 2 - Deterioration of solar cells and modules (Degradation), Code 3 - Voltage mismatching, Code 4 - Current mismatching, Code 5 - Partial shading and soiling of solar cell modules (Partial shading & soiling), Code 6 - Partial shading and series resistance reduction of solar cell modules (Partial shading & reduce Rs), Code 7 - Large-area shading of solar cell modules with reduced voltage (Severe shading_Vmp), Code 8 - Large-area shading of solar cell modules with reduced current (Severe shading_Imp), Code 9 - Temperature sensor error (TC sensor error), Code 10 - Irradiance sensor error (IR sensor error), Code 11 - Irradiance sensor and temperature sensor It may be a record of whether the status is normal or faulty (including the type of fault) by dividing the status into details such as abnormality occurrence (IR & TC sensor error), etc. by code.

[0134] The second data collection unit (700) can receive and collect operational information recording normal and fault states from the power plant and actual power generation measurement data (actual current, actual power, actual power amount) transmitted at regular intervals. In addition, the inference data (inferred current, inferred voltage, inferred power amount) output by the first artificial intelligence model generated by the first artificial intelligence model generation unit (500) at regular intervals can also be collected. As described above, these data can be synchronized in time information based on GPS time, and therefore, a set of inference data, a set of actual power generation measurement data, and a set of operation information can also be matched for the same time based on the time based on GPS time (e.g., the measurement time described above). Therefore, a set of inference data, a set of actual power generation measurement data at the time of inference data generation, and a set of operation information at the time of inference data generation can be classified and collected as a single reference time. The data collected in this manner by the second data collection unit (700) is used as learning data for generating a second artificial intelligence for fault determination. The second data collection unit (700) can use the time information provided by the aforementioned GPS module (100) to synchronize the time information included in the data.

[0135] The second artificial intelligence model generation unit (800) generates a second artificial intelligence model (artificial intelligence model for fault judgment) that determines the normal and fault status of a solar power plant based on the ratio of the inference data and the actual power generation data through machine learning using a learning data set composed of a set of inference data collected by the second data collection unit (700), a set of actual power generation data, and a set of operation information. The second artificial intelligence model generation unit (800) may be formed as a computer device capable of machine learning, and may also be implemented in a form capable of distributed and / or parallel computing in which one or more computer devices are configured in parallel to improve machine learning performance. The second artificial intelligence model generation unit (800) may also be formed to generate a trained artificial intelligence model by training an initialized artificial intelligence model with a learning data set. However, the second artificial intelligence model generation unit (800) can generate an artificial intelligence model (second artificial intelligence model) through classification learning by selecting 3D coordinate points, etc.

[0136] The machine learning algorithm of this second artificial intelligence model generation unit (800) includes a supervised learning algorithm, and the supervised learning algorithm may include regression and classification algorithms. It is possible to classify and learn the fault status according to the operating information through classification learning. More specifically, the second artificial intelligence model generation unit (800) performs a preprocessing step of first configuring the inference data and the generated power measurement data in a ratio among the learning data, and then generates a learned artificial intelligence model (second artificial intelligence model) capable of determining the normal and fault status through machine learning that learns the distribution of the ratio of the inference data and the generated power measurement data and the correlation between the normal and fault status information (e.g., the code information described above) included in the operating information.

[0137] For example, the second artificial intelligence model generation unit (800) generates three ratios (e.g., actual current / inferred current, actual voltage / inferred voltage, actual power / inferred power) for each of the inferred current, inferred voltage, and inferred power included in the inferred data, and each of the actual current, actual voltage, and actual power included in the actual power measurement data, and maps points (or vectors) having the three ratios as components to a vector space (e.g., the coordinate space of FIG. 12) in which the ratios are designated as components of three coordinate axes, thereby learning the distribution of the points, and learns the correlation between the points and the corresponding operational information, thereby generating a trained artificial intelligence model (second artificial intelligence model) that infers the corresponding operational information (e.g., the code information described above) according to the ratios of the inferred data and the actual power measurement data (e.g., actual current / inferred current, actual voltage / inferred voltage, actual power / inferred power).

[0138] That is, as shown in the example of Fig. 12, the second artificial intelligence model can map points (or vectors) on a three-dimensional vector space that has the three ratios as independent components when given three ratios (e.g., actual current / inferred current, actual voltage / inferred voltage, actual power / inferred power) for each of the inferred current, inferred voltage, and inferred power included in the inferred data and each of the actual current, actual voltage, and actual power included in the actual power measurement data. From there, as shown, the second artificial intelligence model can determine normal and fault states from the correlation between the positions of the learned points and the operating information (e.g., Normal - code 0, PID (Potential Induced Degradation) - code 1, Voltage or Current Mismatching - codes 3, 4, Partial shading - codes 5, 6, etc.). Moreover, since the failure status of the operational information is classified and judged according to the spatial distribution of points, the type of failure (e.g., code) can be identified, making it very effective in diagnosing the failure of solar power generation equipment (e.g., electrical equipment of solar power plants, etc.).

[0139] That is, the fault diagnosis system (20) of the present invention collects inference data from the first artificial intelligence model (artificial intelligence model for solar power generation power prediction) generated by the first artificial intelligence model generation unit (500), and utilizes both inference data and actual power generation data from the second artificial intelligence model (artificial intelligence model for fault judgment) generated by the second artificial intelligence model generation unit (800) to more accurately diagnose whether a solar power generation device is faulty. As described above, the first artificial intelligence model is capable of accurate inference by a valid learning data set, and precise status changes can be diagnosed based on the ratio of such inference data and actual data (actual power generation data), so that it is possible to more accurately determine whether a solar power generation device is faulty, thereby assisting in the operation and maintenance of the power plant.

[0140] The second model operation unit (900) may be linked with the second artificial intelligence model (artificial intelligence model for fault judgment) and may have a function of inputting and outputting data. The second model operation unit (900) may have a function of providing data for inference (inference data and actual power generation data) to the generated second artificial intelligence model or receiving output data (whether normal or faulty, type of fault in case of fault) and outputting it externally. The second model operation unit (900) can be understood in accordance with the first model operation unit described above and, if necessary, may be integrated with the second artificial intelligence model generation unit (800). Through such a system configuration, the first artificial intelligence model (power inference model) and the second artificial intelligence model (fault judgment model) can be combined to infer the generated power (inference data) of a solar power plant based on environmental variables (irradiance, atmospheric temperature, etc.) or to effectively diagnose whether a solar power generation device is faulty by utilizing the inferred data.

[0141] Although the embodiments of the present invention have been described with reference to the attached drawings, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without altering the technical concept or essential features thereof. Therefore, the embodiments described above should be understood to be illustrative in all respects and not restrictive.

[0142] [Explanation of symbols]

[0143] 10: Solar power generation power prediction system using an artificial intelligence model

[0144] 20: Solar power generation device fault diagnosis system using artificial intelligence model

[0145] 100: GPS module

[0146] 200: First Data Collection Unit

[0147] 300: Physical Model Department

[0148] 400: Learning data set generation section

[0149] 500: First AI Model Generation Unit

[0150] 600: 1st Model Operation Department

[0151] 700: Second Data Collection Unit

[0152] 800: Second AI Model Generation Unit

[0153] 900: 2nd Model Operation Department

[0154] A: Solar power plant

[0155] B: Measuring station

[0156] The present invention utilizes AI-based prediction to accurately predict and calculate changes in power generation based on environmental conditions (e.g., solar irradiance) of solar power generation facilities, solar power plants, or solar modules, thereby providing advantages in renewable energy management and supply. Furthermore, by using inference based on power predictions to determine whether power generation facilities and equipment are malfunctioning, the present invention also provides advantages in the operation and maintenance of energy facilities. Furthermore, by generating, providing, and applying an AI model capable of predicting with higher accuracy based on more accurate learning data, the present invention enables more accurate and effective operation of solar power facilities. Therefore, the present invention is expected to be useful in energy-related and related industries.

Claims

1. (a) A step of collecting a set of measurement times based on GPS time using a GPS module, a set of actual measurement data including current, voltage, and power amount of a solar power plant for each measurement time, and a set of environmental data including solar irradiance and atmospheric temperature of the solar power plant installation area for each measurement time; (b) a step of constructing an equivalent physical model that calculates the power generation of the solar power plant using the environmental data as input variables, and generating a set of predicted data including the predicted current, predicted voltage, and predicted power amount of the solar power plant for each measurement time that matches the set of environmental data from the physical model; (c) a step of comparing the values ​​of all data included in the set of the predicted data and the set of the actual data by the measurement time, and then generating a reduced learning data set using only pairs of the predicted data and the actual data whose relative error is less than or equal to a reference value, and the environmental data whose pairs and the measurement time match; and (d) A method for generating an artificial intelligence model for predicting solar power generation, comprising a step of generating a trained artificial intelligence model that predicts the power generation of the solar power plant based on environmental variables including solar irradiance and atmospheric temperature through machine learning using the reduced learning data set.

2. In paragraph 1, The above step (c) is a method for generating an artificial intelligence model for solar power generation power prediction, which generates a ratio of the predicted data and the actual measured data for each measurement time, and determines the degree to which the ratio deviates from 1 as the relative error.

3. In paragraph 2, A method for generating an artificial intelligence model for solar power generation power prediction, wherein in step (c) above, the size of the learning data set is adjusted by changing the reference value.

4. In paragraph 2, A method for generating an artificial intelligence model for solar power generation prediction, wherein, in the step (c), the environmental data that falls short of the standard solar irradiance and the pairs of the predicted data and the actual measured data that match the measurement time are removed from the learning data set.

5. In paragraph 4, A method for creating an artificial intelligence model for predicting solar power generation, where the above standard solar irradiance is 300 W / ㎡.

6. In paragraph 1, A method for generating an artificial intelligence model for solar power generation power prediction, wherein the physical model of step (b) includes an air temperature-surface temperature conversion model that calculates the surface temperature of a solar module applied to the solar power plant from the air temperature and the solar irradiance.

7. In paragraph 1, A method for generating an artificial intelligence model for solar power generation power prediction, wherein in the above step (d), the machine learning algorithm includes a supervised learning algorithm, and the supervised learning algorithm includes a regression algorithm. 8.GPS module; A first data collection unit that collects a set of measurement times based on GPS time using the above GPS module, a set of actual measurement data including current, voltage, and power amount of the solar power plant for each measurement time, and a set of environmental data including solar irradiance and atmospheric temperature of the solar power plant installation area for each measurement time; A physical model unit that constructs an equivalent physical model for calculating the power generation of the solar power plant using the above environmental data as input variables, and generates a set of predicted data including the predicted current, predicted voltage, and predicted power amount of the solar power plant for each measurement time that matches the set of environmental data from the physical model; A learning data set generation unit that compares the values ​​of all data included in the set of the predicted data and the set of the actual data by the measurement time, and then generates a reduced learning data set using only pairs of the predicted data and the actual data whose relative error is less than or equal to a reference value, and the environmental data whose pairs and the measurement time match; A first artificial intelligence model generation unit that generates a first artificial intelligence model that predicts the power generation of the solar power plant based on environmental variables including solar radiation and atmospheric temperature through machine learning using the reduced learning data set; and A solar power generation power prediction system using an artificial intelligence model, including a first model operation unit that predicts the power generation of the solar power plant based on the environmental variables by applying the first artificial intelligence model.

9. In paragraph 8, The above first data collection unit is a solar power generation power prediction system using an artificial intelligence model that synchronizes the local time of the sensor or external measuring station providing the environmental data and the local time of the electrical equipment of the solar power plant providing the actual measurement data with the measurement time based on the GPS time.

10. In paragraph 8, The above learning data set generation unit is a solar power generation power prediction system using an artificial intelligence model that generates a ratio of the predicted data and the actual data for each measurement time, and then determines the degree to which the ratio deviates from 1 as the relative error.

11. In paragraph 10, A solar power generation power prediction system using an artificial intelligence model, wherein the reduced learning data set is obtained by removing the environmental data that falls short of the standard solar irradiance and the pairs of the predicted data and the actual measured data that match the measurement time.

12. In paragraph 8, The above physical model is a solar power generation power prediction system using an artificial intelligence model, including an air temperature-surface temperature conversion model that calculates the surface temperature of a solar module applied to the solar power plant from the air temperature and the solar irradiance.

13. A solar power generation power prediction system using an artificial intelligence model according to any one of Articles 8 to 12; A second data collection unit that collects a set of inference data on power generation of a solar power plant, which is an output of the solar power generation power prediction system, a set of actual power generation data provided from the solar power plant at the time of generating the inference data, and a set of operation information including operation information on normal and fault states of the solar power plant at the time of generating the inference data provided from the solar power plant; A second artificial intelligence model generation unit that generates a second artificial intelligence model that determines the normal and fault status of the solar power plant based on the ratio of the inference data and the generated power measurement data through machine learning using a learning data set consisting of the set of the inference data, the set of the generated power measurement data, and the set of the operation information; and A solar power generation device fault diagnosis system using an artificial intelligence model, including a second model operation unit that determines the normal and fault status of the solar power plant by applying the second artificial intelligence model and inputting the inference data and the actual power generation measurement data.

14. In paragraph 13, A solar power generation device fault diagnosis system using an artificial intelligence model, wherein the machine learning algorithm of the second artificial intelligence model generation unit includes a supervised learning algorithm, and the supervised learning algorithm includes a regression and classification algorithm.

15. In paragraph 13, The second artificial intelligence model is a solar power generation device fault diagnosis system that uses an artificial intelligence model to map three ratios of each of the inferred current, inferred voltage, and inferred power included in the inferred data and each of the actual current, actual voltage, and actual power included in the generated power measurement data as independent components into a three-dimensional vector space, and determines normal and fault states from the location of the vector space.

Images