KR20200057821A - solar photovoltatic power generation forecasting apparatus and method based on big data analysis - Google Patents

Abstract

The present invention relates to an apparatus for predicting a generation amount of solar photovoltaic power based on big data analysis which may collect various data on solar photovoltaic power and systematically apply various analysis schemes including multiplex artificial neural network analysis based on the collected data to forecast power generation, and confirm factors affecting the prediction of the generation amount of the solar photovoltaic power for every analysis to increase prediction accuracy with respect to various environmental changes, and a method thereof. In consideration of various climate factors associated with outside temperature, quantity of solar radiation, humidity and the like affecting the solar photovoltaic power generation and various factors including a physical change of a machine such as pollution and temperature of a module, by providing volatility prediction of power yield according to a climate state and physical elements of a machine through accurate power generation prediction and influence factor analysis and fundament information for efficient load operation and management of a power generation system, the fundament information may be widely used as fundament data to increase economics of an existing installation area, to review business of a solar power station, to determine an intention.

Description

Solar photovoltatic power generation forecasting apparatus and method based on big data analysis

The present invention relates to an apparatus and method for predicting the amount of photovoltaic power generation through big data analysis, and in particular, collects various data related to photovoltaic power generation and systematically conducts various analysis techniques including multi-composite artificial neural network analysis based on the collected data. It is possible to predict the amount of power generation by applying it, and it relates to an apparatus and method for predicting the amount of photovoltaic power generation through big data analysis to check the factors affecting the photovoltaic power generation prediction every time to increase the prediction accuracy for various environmental changes. .

Photovoltaic power generation, a renewable energy, is expected to play a very important role in achieving the goal of greenhouse gas reduction along with climate change in the fourth industrial revolution. Therefore, the lifespan and efficiency of the solar panel has been improved with the steady support policy and technology development for the photovoltaic power generation business, and the facilities related to the photovoltaic power generation are being upgraded. Is becoming.

In the case of such photovoltaic power generation, it is very important to predict the amount of power generation because it is dependent on external influences. In the case of excessive power generation, there are physical and electrical hazards due to power generation exceeding the design capacity, and when there are multiple photovoltaic power generation facilities in the connected system Problems such as the occurrence of reverse algae may occur, and poor power generation may cause a problem of falling profits.

As described above, solar power generation prediction plays an important role in preventing physical and economic risk factors, and many profits and losses fluctuate in economic terms depending on the accuracy of solar power generation forecasts, thus expanding the introduction of technology for power generation forecasts in various fields. have.

As is known, in the case of solar power generation prediction, various climatic factors related to outdoor temperature, solar radiation, humidity, etc., and physical changes of the machine such as module contamination, module temperature, etc. must be considered together. Because it occurs, it is not easy to analyze vast amounts of information effectively.

Solar power plants currently operating in many places are experiencing many problems due to climate change, foreign matters, and defects in the module despite increasing demand and expansion and supply. Without it, it is difficult to find, and this problem not only affects the lifespan or power generation of not only individual modules but also other connected modules, and in severe cases, often causes fatal problems such as fire due to electrical abnormalities.

As a method to solve this problem, research is being conducted on various solar power generation predictions, but most of the studies related to a prediction technology or a single prediction model that depend on weather forecasts require improvement of the prediction method. Precise and inaccurate prediction technology is accompanied by problems such as deterioration in power production quality and economic loss due to insufficient management of power production.

In particular, the prediction technology dependent on the weather forecast is difficult to cope with the problem due to foreign matters, defects, etc. of the actual power generation module, and even when a single prediction model or a plurality of these single prediction models are used, it affects the actual power generation prediction by various factors. Because it is difficult to cope with the situation in which the influencing factor affecting is variable, the result is not accurate.

Furthermore, there is a study to apply a prediction model using a neural network such as a machine learning method or deep learning, which has recently been in the spotlight, but in predicting solar power generation, various weather information, various sensing information, and various situations and information about equipment, as well as panels Since it is necessary to consider several different types of data, such as the image information of, it is difficult to determine the target of learning, and the neural network analysis in which only some information is learned has a problem in that it is difficult to trust the prediction result compared to its complexity.

Korean Registered Patent No. 10-1635450, [Photovoltaic power generation prediction system for urban energy management system using weather information] Korean Registered Patent No. 10-1856320, [Solar power generation prediction device and method]

The purpose of the embodiments of the present invention is to predict more accurate power generation by considering various factors including physical changes in the machine such as module temperature, module pollution, and various climate factors related to outdoor temperature, solar radiation, and humidity, which affect solar power generation. It is to provide an apparatus and method for predicting the amount of photovoltaic power generation through the analysis of big data so that it can provide.

Another object of the present invention is to collect environmental data and facility data together with real-time weather information, and hierarchically manage each analysis technique in order to systematically apply various analysis techniques, while predicting power generation through various analysis prediction techniques. Prediction of solar power generation through big data analysis that determines the importance of various input variables applied to the analysis at each analysis time, adjusts the weights and increases the reliability of the weights, thereby increasing the accuracy of predictions according to various environmental changes. It is to provide an apparatus and method.

Another object of the present invention is to apply a neural network analysis optimized for image analysis to image information for checking whether the photovoltaic module has a serious influence on power generation, as well as quantified information such as weather information or power generation information, and In addition, it is possible to perform a highly reliable analysis by applying a multiplexed artificial neural network technology that analyzes the pollution information obtained through neural network analysis and a plurality of quantified development related information through different neural network learning. In parallel, whenever the analysis of power generation is performed, it is possible to reliably determine the weight by analyzing the influence factors that affect the forecast of power generation. Through big data analysis, it is possible to intelligently respond to various environmental changes and increase the hit rate of power generation prediction. It is to provide an apparatus and method for predicting solar power generation.

In order to achieve the above object, the apparatus for predicting solar power generation through big data analysis according to an embodiment of the present invention includes weather data, measurement data for the environment, power generation status data, and observation image data of the solar power module. A data collection unit that collects solar power-related data, and a data storage and preprocessing unit that periodically stores the data collected by the data collection unit and processes the collected data to be used as an influencer for predicting and analyzing power generation. In order to predict the amount of photovoltaic power generation, multiple analysis methods including learning methods and statistical methods, parameters to be used in each analysis method are defined, and the defined parameters are changed according to detailed analysis settings. Manage the parameters, and perform the power generation prediction process with multiple analysis methods that reflect the defined and changed parameters.Whenever the analysis is performed, the influence factor used in the analysis is analyzed to determine its impact and the weight is reset to follow up. It includes a predictive analysis unit to be reflected in the analysis, and an information providing unit that provides estimated power generation amount information and impact factor analysis information according to each analysis method of the prediction analysis unit in a set manner.

As an example, solar power-related data includes external data including past, present, and future weather data provided by an external weather forecasting agency, and solar irradiation, outdoor temperature, module temperature, and humidity data measured in a solar power module installation area, and It may include internal environmental data including image data for detecting contamination and abnormality of the panel, and internal facility data including module power generation data and inverter power generation data.

As an example, the data storage and preprocessing unit may include a storage server that collects data from the data collection unit, a remote shared storage unit that periodically receives and stores the data collected by the storage server, and data stored in the storage server or remote shared storage unit. It may include a data pre-processing unit that performs at least one of noise removal, data normalization, and format change so that it can be obtained and used as an influence factor that is input information for predicting solar power generation.

As an example, the predictive analysis unit defines a learning type including a learning type and a statistical analysis type, defines an analysis type including regression analysis, classification analysis, and cluster analysis, and defines parameters to be used for the defined analysis, thereby defining multiple learning types. The core technique management unit that defines the solar energy predictive analysis method, the general technique management unit that sets the detailed analysis method by changing the parameters of the analysis technique defined in the core technique management unit, and the detailed analysis method according to the general technique according to the set analysis goal It may include a module management unit for setting and managing a group and setting and managing an analysis schedule.

As an example, the prediction analysis unit includes a plurality of analysis processing units for predicting the amount of photovoltaic power generation with defined and set parameters, but as one of them, a deep learning artificial neural network for image analysis that outputs pollution state information from image information of the photovoltaic power generation module. And, multi-composite comprising a predicted power generation neural network that outputs the expected power generation by inputting quantified power generation-related information including at least one of solar radiation, module temperature, and solar altitude and pollution state information output from the deep learning artificial neural network It may include an analysis processing unit using an artificial neural network.

Meanwhile, the deep learning artificial neural network that analyzes module contamination in the analysis processing unit using the multiplexed artificial neural network is a CNN (Convolutional neural network) -based neural network optimized for image analysis, and the power generation prediction neural network includes a plurality of quantified contamination status information and It may be an artificial neural network that outputs a predicted amount of power generation using information related to power generation as an input for learning.

As an example, the predictive analysis unit includes an influencer collection unit that collects influence factors, a predictive analysis unit that analyzes solar power generation prediction results, an influencer accuracy determination unit that determines the accuracy of the predictive analysis unit, and a machine learning method analysis. A cross-validation unit that cross-verifies the training data set used for analysis whenever it is made, a weight determination unit that determines the contribution weight of each factor's power generation through mutual information, and a random forest method. An influencer analysis unit including an influence determination unit for determining an influence degree by obtaining an inference probability for each influencer may be included.

According to another embodiment of the present invention, a method for predicting a photovoltaic power generation through big data analysis collects data related to photovoltaic power generation, including weather data, measurement data on the environment, power generation status data, and observation image data of the photovoltaic module. Data collection step, data storage and pre-processing step to periodically store the data collected in the data collection step, and process the collected data to be used as an influencer for predicting power generation, and learning to predict solar power generation Method management steps to define multiple analysis methods including methods and statistical methods, parameters to be used in each analysis method, to change defined parameters according to detailed analysis settings, and to manage analysis methods and parameters to be used according to analysis goals , The power generation prediction step of performing the power generation prediction process by a plurality of analysis methods reflecting the parameters defined and changed in the technique management step, and the impact factor analyzed by analyzing the influence factor used in the analysis whenever the analysis is performed in the power generation prediction step And an information providing step of providing an estimated generation amount information and an effect factor analysis result obtained in the generation factor prediction step and the influence factor analysis step in a set manner so as to determine and reset the weight to be reflected in the subsequent analysis.

As an example, the power generation prediction step analyzes the image information of the photovoltaic power generation module through a deep learning artificial neural network optimized for image analysis, obtains pollution state information, and quantifies at least one of solar radiation, module temperature, and solar altitude. It may include a multi-assembly artificial neural network prediction step of predicting the amount of power generation through the power generation prediction neural network as input to the power generation-related information and the pollution state information analyzed in the deep learning artificial neural network.

As an example, the influencer analysis step performs machine learning until a value equal to or higher than the accuracy set by the K-fold cross-validation separation method is performed on the machine learning data set every time the machine learning method PV analysis is performed. The step of verifying the weight, the step of determining the contribution weight of each influencer's power generation through Mutual Infotmation, and the step of determining the degree of influence by obtaining the reasoning probability for each influencer in a random forest method It may include at least one step.

The apparatus and method for predicting the amount of photovoltaic power generation through big data analysis according to the embodiments of the present invention include various climate factors related to the photovoltaic power, solar radiation, humidity, etc. By considering various factors including physical changes, more accurate power generation prediction and impact factor analysis provide basic information for efficient load operation and management of the power generation system, as well as forecasting volatility of power generation according to weather conditions and physical elements of machinery. By providing it, there is an effect that can be widely used as basic data for improving economic feasibility of existing installation areas and reviewing and making business decisions of new solar power plants.

The apparatus and method for predicting solar power generation through big data analysis according to embodiments of the present invention hierarchize each analysis technique to collect environmental data, facility data together with periodic real-time weather information, and systematically apply various analysis techniques In addition, it predicts the amount of power generation through various multi-composite analysis forecasting techniques while managing it as an enemy, and determines the importance of various input variables applied to the analysis at each analysis time, adjusts the weights, and increases the reliability of the weights, thereby changing various environments. It is possible to increase the accuracy of prediction according to the effect of responding to the changing state of the device or the changing weather.

Furthermore, the image information of the photovoltaic module, which has a significant influence on power generation, can be applied to the neural network analysis optimized for image analysis to classify the pollution status, so it accurately considers not only the quantified information but also the status of the module that is difficult to distinguish without an expert. , By applying the power generation prediction neural network that considers these pollution information and quantified information collectively, more accurate power generation prediction is possible, thereby increasing the hit rate and reliability of power generation prediction.

1 shows a conceptual configuration diagram of a photovoltaic power generation prediction device through big data analysis according to an embodiment of the present invention.
2 shows a configuration of a data storage and preprocessing unit according to an embodiment of the present invention.
3 conceptually shows the processing contents of the technique management components included in the predictive analysis unit according to an embodiment of the present invention.
4 conceptually shows the configuration of the predicted power generation analysis unit included in the prediction analysis unit according to an embodiment of the present invention.
5 is a conceptual diagram for explaining a configuration and an operation method of a multiplexed artificial neural network which is one of an analysis processing unit of an expected generation amount analysis unit according to an embodiment of the present invention.
6 to 8 is an exemplary embodiment showing the configuration of a multiplexed artificial neural network.
9 is a conceptual diagram showing the configuration of an influencer analysis unit according to an embodiment of the present invention.
10 is a flowchart illustrating a method of predicting solar power generation through big data analysis according to an embodiment of the present invention.

The present invention as described above will be described in detail through the accompanying drawings and embodiments.

It should be noted that the technical terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the present invention. In addition, technical terms used in the present invention should be interpreted as meanings generally understood by a person having ordinary knowledge in the technical field to which the present invention belongs, unless otherwise defined in the present invention. It should not be interpreted as a meaning or an excessively reduced meaning. In addition, when the technical term used in the present invention is a wrong technical term that does not accurately represent the spirit of the present invention, it should be understood as being replaced by a technical term that can be correctly understood by those skilled in the art. In addition, the general terms used in the present invention should be interpreted as defined in the dictionary or in context before and after, and should not be interpreted as an excessively reduced meaning.

In addition, the singular expression used in the present invention includes a plural expression unless the context clearly indicates otherwise. In the present invention, terms such as “consisting of” or “comprising” should not be construed to include all of the various components, or various stages, described in the invention, including some of the components or some stages. It may or may not be construed as further comprising additional components or steps.

Further, terms including ordinal numbers such as first and second used in the present invention may be used to describe elements, but the elements should not be limited by terms. The terms are used only to distinguish one component from other components. For example, the first component may be referred to as a second component without departing from the scope of the present invention, and similarly, the second component may be referred to as a first component.

Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the accompanying drawings, but the same or similar elements will be given the same reference numbers regardless of the reference numerals, and repeated descriptions thereof will be omitted.

In addition, in the description of the present invention, when it is determined that detailed descriptions of related known technologies may obscure the subject matter of the present invention, detailed descriptions thereof will be omitted. In addition, it should be noted that the accompanying drawings are only for easy understanding of the spirit of the present invention and should not be interpreted as limiting the spirit of the present invention by the accompanying drawings.

Meanwhile, the apparatus referred to in the present invention may be implemented as a hardware component, a software component, and / or a combination of a hardware component and a software component. The systems, devices and components described in embodiments include, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors (micro signal processors), microcomputers, field programmable arrays (FPAs), PLUs ( It may be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit, microprocessor, or any other device capable of executing and responding to instructions. In addition, such a server or device may run an operating system (OS) and one or more software applications running on the operating system. In addition, the device may access, store, manipulate, process, and generate data in response to the execution of the software. For convenience of understanding, a server or a device may be described as one being used, but a person having ordinary skill in the art, the processing device may include a plurality of processing elements and / or a plurality of types of processing. It can be seen that it can contain elements. For example, a server or device may include a plurality of processors or a processor and a controller. In addition, other processing configurations, such as parallel processors, are possible. In addition, the device can control various sensors directly through the interface or through a separate hardware / software board or module for controlling a separate driver or sensor, and can cover all various logical, electrical, and optical modifications for this purpose. have.

Furthermore, in the embodiments of the present invention, devices for collecting data and components thereof may be devices including hardware or software for collecting information of physical and electrical sensors, and receiving and processing electrical signals. Interface, drive driver, power supply, microcontroller for control, storage, operating system or firmware, and sensors as needed.

Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1 to 10.

1 shows a conceptual configuration diagram of a photovoltaic power generation prediction device through big data analysis according to an embodiment of the present invention.

As shown, it is largely divided into a data collection unit 100, a data storage and pre-processing unit 200, a prediction analysis unit 300, and an information providing unit 400, each of which may be composed of a plurality of independent or interlocked devices. Since it can be configured as a system in practice, it is also referred to as a device as a comprehensive concept because it is possible to unify the configuration to process them according to the scale. Meanwhile, these components may also be configured as a single device or a plurality of devices.

Looking at the illustrated configuration, the data collection unit 100 is an external data collection unit 110 for collecting weather data including information about the past, present and future related to the weather from an external agency, and the area where the photovoltaic power generation is made An internal environmental data collection unit 120 for collecting various environmental measurement data obtained from the inside and observation image data of a solar power module, and an internal facility data collection unit for collecting data according to the power generation state in internal facilities for solar power generation ( 130).

Data collected by each illustrated collection unit is not limited to the illustrated types, and more data not illustrated may be collected, and some of them may not be used.

The data storage and pre-processing unit 200 periodically stores data collected by the data collection unit 100 and processes the collected data so that it can be used as an influencer for predicting power generation. In conjunction with each collection unit (110, 120, 130) of the collection unit 100 stores the data provided by the storage server 210 and the storage server 210, which receives and stores or transmits the data as needed, If necessary, the remote shared storage unit 220 that can be shared with other photovoltaic power generation prediction devices and data stored in the remote shared storage unit 220 or data provided by the storage server 210 is retrieved or searched for the sun. And a data pre-processing unit 230 that performs pre-processing to be used as an input for photoelectric power generation prediction analysis, that is, to be used as an influence factor.

The predictive analysis unit 300 sets a plurality of analytical methods including a learning method and a statistical method for predicting the amount of photovoltaic power generation, and a key technique management unit 310 that defines parameters to be used in each analysis method, and a detailed analysis setting of the defined parameters The general technique management unit 320 that varies according to the analysis, the module management unit 330 that manages the analysis method and parameters to be used according to the analysis target, and the expected generation amount analysis that performs the analysis process in a plurality of analysis methods with defined and changed parameters The unit 340 includes an impact factor analysis unit 350 that analyzes an influence factor used in the analysis every time the analysis is performed, determines its impact level, and resets the weight to be reflected in the subsequent analysis. Each of these components will be described in detail later with reference to FIGS. 3 to 9.

How the information providing unit 400 uses the predicted generating amount information providing unit 420 and the predicted generating amount information or the impact factor analysis information to provide the predicted generation amount information and the impact factor analysis information according to each analysis method of the predictive analysis unit 300 It includes a user UX unit 410 to set whether or not to provide to the user according to the setting. Of course, it may include various visual and audible interface components for providing to the user.

2 shows a configuration of a data storage and preprocessing unit according to an embodiment of the present invention.

As shown, the storage server 210 collects data related to photovoltaic power generation from the data collection unit 100, and this data collection unit 100 is a meteorological agency API module, solar radiation sensor, and outdoor air for conceptually obtaining external weather data. It includes a temperature sensor, a module temperature sensor, a humidity sensor, a camera that provides image data for detecting contamination and abnormality of a panel, a module power generation sensor, and an inverter that provides inverter power generation information. Here, it should be considered that each sensor, camera, and inverter includes a collection unit that collects information from them, rather than that it is included as a direct component. Meanwhile, the instantaneous voltage and the instantaneous power value of the photovoltaic module can be collected from the module power generation sensor, and the cumulative production, DC total power, DC average power, and DC average voltage can be collected from the inverter.

The illustrated storage server 210 may periodically (in the illustrated embodiment, one minute unit) store the collected data in the remote shared storage unit 220 or transmit it to the data preprocessing unit 230 through the communication unit. The storage server 210 collects data, analyzes its type, designates a storage location, and pre-processes some information in that various environmental data or facility data generated in one solar power plant is continuously generated. In addition to the functions of the general database server provided to the unit 230, it may include a function of a distributed processing server that supports a distributed processing architecture (for example, Hadoop or Spark, etc.) for processing vast collection of data whose data volume increases exponentially according to accumulation. .

The shared storage unit 220 may store environment data or facility data provided by the storage server 210, and may take the form of a distributed storage unit or a cloud storage unit in that the information is vast, and may be in the form of one if necessary. It is also possible to integrate and store data in conjunction with storage servers of photovoltaic power generation prediction devices for a solar power plant other than the solar power plant storage server 210 of.

The data pre-processing unit 230 obtains data stored in the storage server 210 or the remote shared storage unit 220, and removes noise and normalizes the data so that the data can be used as input information for predicting solar power generation, that is, as an influence factor , At least one of the format changes may be performed.

In an embodiment of the present invention, the prediction analysis unit 300 is largely composed of technique management components 310, 320, 330 for systematically managing analysis techniques, and a plurality of analysis processing units for substantially analyzing expected power generation. It includes a power generation analysis unit 340 and an influence factor analysis unit 350 for analyzing an influence factor used in analyzing the expected generation amount and changing a weight.

First, the technique management configurations of the prediction analysis unit 300 will be described with reference to FIG. 3.

FIG. 3 conceptually shows the processing contents of the technique management components included in the prediction analysis unit according to an embodiment of the present invention, and as shown, a core technique management unit defining a plurality of analysis methods and parameters to be used in each analysis method ( 310) and the general technique management unit 320 that changes the parameters defined in the core technique management unit according to the detailed analysis setting and the module management unit 330 that manages the general technique as a group according to the analysis goal. Through this hierarchical configuration, it is possible to effectively manage a plurality of solar power generation prediction methods, and to effectively respond to changes in influence factors (such as classifying influence factors with high influence according to environmental changes and reflecting changed weights), Depending on the situation, appropriate analysis methods or analysis methods can be selected or detailed analysis conditions can be set, and analysis groups according to situations can be managed as modules (modules such as analysis methods and settings set in a group).

The core technique management unit 310 defines basic analysis techniques for predicting the amount of photovoltaic power generation, learning to define machine learning types such as supervised learning / unsupervised learning / semi-supervised learning / strong learning / ensemble and general statistical analysis types. It provides a type definition function, an analysis type definition function that defines regression / classification / cluster analysis, and a parameter definition function that can utilize collected data as parameters in the defined learning type and analysis type.

The general technique management unit 320 provides a function to classify the parameters of the prediction technique defined in the core technique into various techniques in detail, and for this purpose, interlocks the core technique and the general technique (details using the core technique selected from the core techniques) It provides the function to define the analysis method as a general technique) and to define general technique parameters.

The module management unit 330 is a main function that sets and manages various general techniques as a group based on a target value to be predicted or analyzed, and as an auxiliary function, analysis such as a prediction period / model update period / usability by general technique It provides a function to set the schedule of work and to set tags for input and output.

In other words, the module management that manages the basic analysis methods that are the core of the analysis as the core technique, manages the general technique that materializes the core technique as specific analysis conditions, and sets and manages these general techniques as a group hierarchically. Because it is done, it is consistent, free of modification, and can provide a flexible structure to add necessary core techniques or general techniques.

The predictive analysis unit 300 includes a predicted generation amount analysis unit 340 including a plurality of analysis processing units for actually predicting solar power generation by actually performing specific general techniques managed by such technique management configurations. As illustrated in FIG. 4, the predicted power generation analysis unit 340 has various analysis processing units that perform actual analysis using analysis methods and parameters substantially defined in technique management configurations. There may be several analysis processing units including learning methods.

In an embodiment of the present invention, at least a core prediction analysis technology applied to photovoltaic power generation may include an analysis processing unit using regression analysis and an analysis processing unit utilizing a multiplexed artificial neural network.

First, the analysis processing unit through regression analysis performs machine learning through regression analysis, and an exemplary machine learning function is as follows.

 Here, x includes external collected data and internal collected data (environmental data, facility data) collected as input variables (independent variables), and y is predicted power generation as output variables (dependent variables).

 For example, the independent variable x used in the regression analysis can utilize solar radiation, module temperature, solar altitude, etc. In addition, it is possible to predict solar power generation by combining various variables.

On the other hand, another of the core predictive analysis techniques used in the embodiments of the present invention is an analysis processing unit utilizing a multiplexed artificial neural network, which will be described with reference to FIGS. 5 to 8.

FIG. 5 is a conceptual diagram for explaining the configuration and operation method of the multiplexed artificial neural network, which is one of the analysis processing units of the predicted power generation analysis unit according to an embodiment of the present invention, unlike other types of analysis methods based on quantified collection data. In the case of the illustrated multi-complex artificial neural network, the camera image for checking the pollution or abnormality of the photovoltaic power generation module is analyzed to determine the pollution degree of the module and reflect it to predict the amount of photovoltaic power generation.

As illustrated, the multiplexed artificial neural network analysis processing unit 340a included in the predicted power generation analysis unit 340 is collected from the camera of the data collection unit 100 and stored and pre-processed in the data storage and pre-processing unit 200 The neural network pollution level measuring unit 341a which outputs the pollution state information from the panel image information of the power generation module, and the amount of solar radiation collected and pre-processed in the data storage and preprocessing unit 200 collected from sensors of the data collection unit 100, and Estimated amount of power generated by inputting module temperature, quantified power generation related information including solar altitude information obtained from system reference information (latitude, longitude, time) and pollution state information (contamination degree) output from the neural network pollution level measuring unit 341a. It includes a neural network power generation prediction unit (342a) for outputting.

That is, the multiplexed artificial neural network analysis processing unit 340a is composed of a first neural network that analyzes a module image and outputs a pollution level, and a second neural network that outputs an expected power generation amount using information related to the pollution level and quantified solar power generation. .

Here, the first neural network for analyzing an image uses a deep learning artificial neural network (for example, a convolutional neural network (CNN) -based neural network) optimized for image analysis, and the second neural network for outputting an expected amount of generation using quantified information is a numerical value. Neural networks optimized for analysis can be used.

6 shows a CNN configuration as a deep learning artificial neural network optimized for image analysis constituting the neural network contamination measurement unit 341a. Of course, CNN is known as a deep learning method that is optimized for image analysis, but since there are improvements or variations based on it, this new deep learning method can be adopted, which is the core technique management unit 310 and general techniques described above. It can be used after registration and specification through the management unit 320.

As shown, select the target area from the pre-processed image of the photovoltaic module (especially the panel), apply the convolution operation through the filter (C1), apply the activation function, go through the pooling (S2) process, and filter again After applying the convolution operation through (C3) and applying the activation function, the feature map constructed through the process of pooling (S4) can be learned through the neural network learning unit (pollution measurement model) corresponding to the classifier to output the pollution level.

Here, the convolution and pooling process is grouped and viewed as a layer. The illustrated example is only applied to two layers. If necessary, it can be stacked into multiple layers, and the higher the number of layers, the higher the performance. can do. Accordingly, the number of image pixels in the illustrated example, the size of the filter, and the size of the pooling are merely exemplary and may be variously changed.

FIG. 7 shows the configuration of the neural network learning unit (contamination measurement model) shown in FIG. 6 in more detail, and as shown, FCV [1,1] to FCV [of the n × m matrix output through the S4 layer as shown in FIG. n, m] is feature information extracted from the module image through CNN, and when this input is provided to a hidden layer composed of a fully connected network, as a result, it provides pollution degree as output y.

At the output layer, RC stands for the module's real current, TIP stands for the module's theoretical instant power, and RV stands for the module's real voltage. Through the number of neurons in the hidden layer, the learning rate, the batch size, and the number of learning can be defined.

As described above, the pollution level of the module may be provided as a value less than 1 through a photographed image of the photovoltaic power generation module, and a higher value may be displayed when the pollution degree is high.

FIG. 8 shows the configuration of the neural network power generation predictor 342a illustrated in FIG. 5, and the illustrated configuration is the pollution degree C obtained through the pollution measurement model (CMM) applied to the neural network pollution measurement unit 341a. ) And solar radiation obtained through solar radiation sensor (Real Radiation (W / ㎡)) (RR), module temperature obtained through module temperature sensor (Real Module Temperature (℃)) (RMT), obtained using system reference information ( Modeled as a hidden layer consisting of a fully connected network using Sun Azimuth Angle (°) (SAA) as input (X, learning data) Instantaneous power prediction model Outputs the estimated amount of solar power generated through the neural network. The RIP provided as the output (Y) means the average instantaneous power of the predicted module. Through the number of neurons in the hidden layer, the learning rate, the batch size, and the number of learning can be defined.

In the case of predicting the amount of photovoltaic power generation, it is not important to continuously output in real time, but it is important to accurately predict the amount of photovoltaic power generation in response to a relatively slowly changing environment. By applying a complex neural network such as 340a), it can be judged only by a skilled practitioner and focusing on it. Depending on the situation, the decision maker can check the contamination level of the module through a large image based on big data. The pollution level can be judged more accurately, and the generation rate is predicted by the neural network that reflects the more accurate panel pollution level.

Now, it is configured in the prediction analysis unit 300 of the embodiment of the present invention, and the influencer analysis unit 350 for analyzing the influence factor used in the expected generation amount analysis and changing the weight will be described.

9 is a conceptual diagram showing the configuration of an influencer analysis unit according to an embodiment of the present invention.

As described above, it is important to accurately predict the amount of photovoltaic power generation in response to an environmental change in which various changes are applied variably in the case of predicting photovoltaic power generation, but this is simply because several types of analysis methods are used. It is not going to be high.

That is, whether the factors affecting the amount of power generated by changes in the environment are solar radiation, solar altitude, ambient temperature, humidity, module contamination, module temperature, expected weather change, or module abnormality or power generation Depending on the environment, whether the inverter efficiency decreases or which of these is the main cause is changed depending on the environment. Therefore, it is necessary to constantly check and re-adjust whether parameters or weights applied to various analysis methods are suitable.

Furthermore, since the data collection unit according to an embodiment of the present invention collects real-time unstructured data, as data increases, problems such as error or loss of data may occur due to many calculations for learning.

Therefore, the influencer analysis unit 350 according to an embodiment of the present invention analyzes the correlation and weight of the predicted amount and the influencer, as shown in the figure, the influencer collection unit for collecting the influencer, and the photovoltaic A prediction analysis unit for analyzing the power generation prediction result, an influence factor accuracy determination unit for determining the accuracy of the prediction analysis unit, a cross-validation unit for cross-validating a set of learning data used for analysis whenever a machine learning method analysis is performed, It may include an influencer analysis unit including a weight determination unit for determining the weighted contribution weight of each influencer, and an influence determination unit for determining an influence degree by obtaining an inference probability for each influencer.

Here, the predictive analysis unit, the accuracy determination unit, and the cross-validation unit, for example, whenever a machine learning-type photovoltaic power generation predictive analysis is performed, the machine learning data set has a value higher than the accuracy set by the K-fold cross-validation separation method. Until then, machine learning can be performed to verify the weight of the influencer.

On the other hand, the weight determining unit can determine the weight of each influencer's power generation contribution through mutual information (Mutual Infotmation), and the influence determination unit can determine the degree of influence by obtaining the inference probability for each influencer in a random forest method. have.

It is desirable to change the weights of the influence factors obtained through these processes in subsequent analysis (parameter setting of the core technique management department, parameter setting of the general technique management department), and the contribution or influence of the generation amount to these influencers is variable. Therefore, it is possible to appropriately respond to various environmental changes, and by providing the information on the analyzed impact factor to the information providing unit 400 together with the expected generation amount, it can be used as basic information for the manager's proper judgment.

10 is a flowchart showing a method of predicting solar power generation through big data analysis according to an embodiment of the present invention, as shown, weather data, measurement data for the environment, power generation status data, and solar power module observation image data And collecting, storing, and pre-processing the photovoltaic power generation-related data.

Thereafter, a plurality of analysis methods including a learning method and a statistical method for predicting the amount of photovoltaic power generation, and parameters to be used in each analysis method are defined, and the defined parameters are changed according to detailed analysis settings, and an analysis method to be used according to the analysis goal And managing parameters. If the weight of the influencer is changed and the influencer with high impact is determined through the analysis of the previous generation forecast, an appropriate analysis method and parameter change can be performed.

Thereafter, a power generation prediction step of performing an analysis process by a plurality of analysis methods with defined and changed parameters is performed. In the embodiment of the present invention, the multi-generation artificial neural network power generation prediction calculates module contamination through a neural network and includes contamination. By using power generation-related data, it is possible to predict the amount of power generation based on neural networks, so that the pollution level, which is an important factor in power generation prediction, can be calculated from module images, which are unstructured information, and can be actively utilized.

In addition, it includes a step of performing predictive analysis processing using other analysis methods such as regression analysis power generation prediction, statistical model power generation prediction, deep artificial neural network analysis, etc., and can be performed in parallel with the multiplexed artificial neural network power generation prediction. .

And providing the predicted power generation information obtained as described above to a manager or a user through various information or auditory information through the information providing unit.

In addition, each time the analysis is performed, performance evaluation of the influencer, weight determination, impact degree determination, etc. are performed, and the reset weight or influencer influence information is reflected in subsequent analysis. Analysis information on the affected factor may also be provided to an administrator or user through the information providing unit.

Furthermore, as all these processes are continuously repeated, the analysis accuracy using big data is gradually improved and it is possible to intelligently respond to various environmental changes.

In the above, preferred embodiments according to the present invention have been illustrated and described. However, the present invention is not limited to the above-described embodiments, and various modifications can be made to any person skilled in the art to which the present invention pertains without departing from the gist of the present invention appended in the scope of the claims. .

100: data collection unit 110: external data collection unit
120: internal environment data collection unit 130: internal facility data collection unit
200: data storage and pre-processing unit 210: storage server
220: remote shared storage unit 230: data pre-processing unit
300: predictive analysis unit 310: core technique management unit
320: general technique management unit 330: module management unit
340: expected power generation analysis unit 350: impact factor analysis unit
400: information providing unit 410: user UX unit
420: Estimated power generation information provider

Claims (10)

A data collection unit for collecting data related to photovoltaic power generation, including weather data, measurement data for the environment, power generation status data, and observation image data of the photovoltaic power generation module;
A data storage and pre-processing unit that periodically stores data collected by the data collection unit and processes the collected data to be used as an influence factor for predicting and analyzing power generation;
In order to predict the amount of photovoltaic power generation, a plurality of analysis methods including a learning method and a statistical method and parameters to be used in each analysis method are defined, and the defined parameters are varied according to detailed analysis settings, and an analysis method to be used according to an analysis target and Manage the parameters, and perform the power generation prediction process with a plurality of analysis methods that reflect the defined and changed parameters.Whenever the analysis is performed, the impact factor used in the analysis is analyzed to determine its impact and the weight is reset to follow up. A prediction analysis unit to be reflected in the analysis;
An apparatus for predicting photovoltaic power generation through big data analysis, including an information providing unit that provides estimated power generation amount information and impact factor analysis information according to each analysis method of the prediction analysis unit in a set manner.
The method according to claim 1, The data related to the photovoltaic power generation, external data including past, present and future weather data provided by an external weather forecasting agency, and solar radiation measured in the solar power module installation area, outside temperature, module temperature, The amount of photovoltaic power generation through big data analysis, which includes internal environment data including humidity data and image data for detecting panel contamination and abnormalities, and internal facility data including module power generation data and inverter power generation data. Prediction device.
The method according to claim 1, Data storage and pre-processing unit
A storage server collecting data from the data collection unit;
A remote shared storage unit for periodically receiving and storing the data collected by the storage server;
And a data pre-processing unit that acquires data stored in the storage server or the remote shared storage unit and performs at least one of noise removal, data normalization, and format change so that it can be used as an influence factor that is input information for predicting solar power generation. A device for predicting the amount of photovoltaic power generation through big data analysis, characterized in that.
The method according to claim 1, The predictive analysis unit
Multiple photovoltaic power generation prediction analysis methods to be used by defining a learning type including a learning type and a statistical analysis type, defining an analysis type including regression analysis, classification analysis, and cluster analysis, and defining parameters to be used for the defined analysis. A core technique management unit defining a;
A general technique manager configured to set a detailed analysis method by changing parameters of an analysis technique defined in the core technique manager;
A device for predicting solar power generation through big data analysis, comprising a module management unit that sets and manages a detailed analysis method according to a general technique as a group according to a set analysis goal and sets and manages an analysis schedule.
The method according to claim 1, The prediction analysis unit includes a plurality of analysis processing unit for predicting the amount of photovoltaic power generation by the defined and set parameters, as one of them,
Deep learning artificial neural network for image analysis that outputs contaminant status information from image information of a solar power module, and quantitative power generation related information including at least one of solar radiation, module temperature, and solar altitude, and output from the deep learning artificial neural network An apparatus for predicting photovoltaic power generation through big data analysis, comprising an analysis processing unit using a multiple artificial neural network including a power generation prediction neural network that outputs expected power generation by inputting the polluted state information as input.
The deep learning artificial neural network that analyzes module contamination in the analysis processing unit using the multiplexed artificial neural network is a CNN (Convolutional neural network) based neural network optimized for image analysis, and the power generation prediction neural network is polluted state information and quantification. An apparatus for predicting solar power generation through big data analysis, characterized in that it is an artificial neural network that outputs predicted power generation by using a plurality of information related to power generation as input for learning.
The method according to claim 1, The predictive analysis unit is an influence factor collection unit for collecting an influence factor, a prediction analysis unit for analyzing the solar power generation prediction result, an influence factor accuracy determination unit for determining the accuracy of the prediction analysis unit, and a machine learning method. A cross-validation unit that cross-validates the training data set used for analysis whenever the analysis is performed, a weight determination unit that determines the weight contribution weight of each influencer through mutual information, and a random forest. ) A method for predicting solar power generation through big data analysis, comprising an influence factor analysis unit that includes an influence determination unit to determine the degree of inference by determining the inference probability for each influence factor in a manner.
A data collection step of collecting data related to photovoltaic power generation, including weather data, environmental measurement data, power generation status data, and solar power module observation image data;
A data storage and pre-processing step of periodically storing the data collected in the data collection step and processing the collected data to be used as an influence factor for predicting power generation;
In order to predict the amount of photovoltaic power generation, multiple analysis methods including learning methods and statistical methods, parameters to be used in each analysis method are defined, and the defined parameters are changed according to detailed analysis settings, and analysis methods and parameters to be used according to the analysis goal Techniques for managing the management step;
A power generation prediction step of performing a power generation prediction process by a plurality of analysis methods reflecting the parameters defined and changed in the technique management step;
An influence factor analysis step of analyzing an influence factor used for analysis every time an analysis is performed in the predicting amount of power generation to determine its influence and resetting the weight so that it is reflected in subsequent analysis;
A method of predicting solar power generation through big data analysis, including an information providing step of providing predicted power generation information and an influencer analysis result obtained in the power generation prediction step and the influencer analysis step in a set manner.
The method according to claim 8, The step of predicting the amount of power generation
Analyze image information of the photovoltaic power generation module through deep learning artificial neural networks optimized for image analysis to obtain pollution status information, and quantify power generation related information including at least one of solar radiation, module temperature, and solar altitude and the deep learning A method for predicting solar power generation through big data analysis, comprising a multi-assembly artificial neural network prediction step of predicting power generation through a neural network.
The method according to claim 8, The step of analyzing the influencer
Verifying the weight of the influence factor by performing machine learning until a value equal to or greater than the accuracy set by the K-fold cross-validation separation method of the machine learning data set is performed whenever the machine learning method predictive analysis of photovoltaic power generation is performed;
Determining a weighted contribution amount of power generation of each influencer through mutual information (Mutual Infotmation);
A method of predicting solar power generation through big data analysis, comprising at least one of the steps of determining the degree of impact by obtaining an inference probability for each influencer by a random forest method.

Images