KR102211157B1 - optimaized water cooling apparatus and method for solar photovoltatic power generation module using rain water - Google Patents

Abstract

The present invention collects various data related to photovoltaic power generation, calculates necessary prediction information through various types of neural network prediction algorithms, generates a net present value, and uses this as compensation information to perform panel cooling control utilizing excellence. It relates to a solar power module optimal cooling device and method using rainwater that enables optimized cooling control in consideration of actual power plant operating cost, power generation profit, and cooling effect through a learning method. Net present value calculated by taking into account the increased power generation profit and the cost for water spray control with state information (state) and water spray control to improve solar power generation efficiency as action, and solar power generation efficiency improvement By applying the A3C (Asynchronous Advantage Actor-Critic)-based deep reinforcement learning method as a reward to the optimal cooling of the solar power module, it comprehensively considers the amount of power generation, power generation revenue, and cost required for spraying according to the accumulation of learning. There is an effect that can greatly increase the economics of power plant operation through optimal water spray control.

Description

The optimal cooling apparatus and method for solar photovoltaic power generation module using rain water

The present invention relates to an optimal cooling apparatus and method for a solar power module using rain, and in particular, by collecting various data related to solar power generation and calculating necessary prediction information through various types of neural network prediction algorithms, a net present value is generated. Solar power module optimal cooling device using rainwater that enables optimized cooling control in consideration of actual power plant operation cost, power generation profit and cooling effect through a reinforcement learning method that uses this as compensation information to perform panel cooling control using rainwater And a method.

Solar power generation, a new renewable energy, is expected to play a very important role in achieving the goal of reducing greenhouse gas emissions along with climate change in the Fourth Industrial Revolution. Therefore, the lifespan and efficiency of solar panels have been improved according to the continuous support policy and technology development for the solar power generation business, and the facilities related to solar power generation are also advanced, and not only large solar power plants but also small-scale solar power generation are activated. Has become.

However, in order to further activate such solar power generation, solar power generation operators must be able to obtain sufficient profits through solar power generation. The yield of solar power generation does not simply depend on the power generation capacity of the solar power panel or the weather environment, but effectively sells the solar power produced, reduces the cost of operating the power plant, and the efficiency of the currently built solar power generation panel. Several factors that maximize the value should be considered in an integrated manner.

In the case of such solar power generation, since it is dependent on external influences, it is essential to predict the amount of power generated in consideration of various environmental factors, control panel temperature or control panel pollution to increase power generation efficiency, and to obtain economical efficiency, the system marginal price (System Marginal Price: Various flexible conditions such as SMP), mandatory quota for renewable energy generation, and Renewable Energy Certificate (REC) prices must be considered.

In particular, the efficiency of the solar power module panel decreases by about 0.5% when the temperature rises by 1℃ from the standard temperature of 25℃. Therefore, in the summer season, the panel temperature of the photovoltaic module may rise up to 70°C or higher depending on the region. Therefore, since the decrease in the amount of power generation due to the increase in the panel temperature is significant, various studies to lower the temperature of the panel are being conducted.

The most common method for lowering the panel temperature of such a photovoltaic module is to reduce the temperature by spraying water (tap water) or groundwater to the panel in a water-cooled cooling method using cooling water, and further remove dust or contamination to increase efficiency. . However, in the case of a conventional water-cooled panel cooling method, it is possible to increase some power generation efficiency because water or groundwater is simply sprayed according to the temperature or schedule, but it is possible to increase the various costs required for such water-cooled cooling and whether the power generation efficiency obtained through it will lead to profits. The analysis is not being considered at all.

Therefore, only the technology to lower the temperature of the panel technically exists, but it is not optimally performed for the operation of a solar power plant.

Korean Patent Registration No. 10-1892050, [Water-cooled solar power generation device] Korean Patent Registration No. 10-1811426, [Integrated power management system using solar power generation]

The object of the embodiments of the present invention is to use various data on solar power generation and various cost data required for cooling in order to optimally control a panel of a photovoltaic module in consideration of economics and efficiency. After calculating the net present value based on this, it is used as compensation information, and rainwater spray control for efficiency improvement through in-depth reinforcement learning based on A3C (Asynchronous Advantage Actor-Critic) using current data related to solar power generation as status information It is to provide an optimal cooling device and method for solar power modules using rainwater that enables information to be calculated.

In addition, another object of the embodiments of the present invention is to collect various power generation data related to solar power generation, power plant operation data, and external data to predict generation amount, module temperature prediction, and power generation revenue through a neural network. After calculating the present value of power generation revenue through predictive information, the net present value is generated considering the excellent control operation cost, and reinforcement learning is applied to enable optimal control in the direction of increasing the net present value. It is to provide an optimal cooling device and method for solar power modules using rainwater that can be maximized.

Another object of the embodiments of the present invention is to apply the A3C deep reinforcement learning method, which is known as a learning method for performing digital games, for solar cooling, which is a real environment, but individually applying the A3C deep reinforcement learning method so that the learning is not biased. Provides an optimal cooling device and method for solar power modules using an excellence that can improve learning performance by reflecting learning information for individual solar power plants with different environments into the global neural network by using the reinforcement learning performance unit of the solar power plant as a sub-agent. will be.

In order to achieve the above object, the solar power module optimal cooling device using rain according to an embodiment of the present invention includes external data on weather and power prices and power generation costs, and a plurality of power generation data related to solar power generation. A data collection unit that collects internal measurement data related to panel cooling and power plant operation data for panel cooling control, a data storage unit that stores information collected from the data collection unit as a distributed database or relational database, and data storage An analysis unit that calculates the present value of power generation revenue using collected data stored in the wealth, calculates the operating cost required to provide water for cooling, and calculates the net present value from them, and a control signal for spraying solar power panels A3C-based in-depth reinforcement learning is performed using the control unit that provides a control unit and the state information related to photovoltaic power generation collected by the data collection unit to control the control unit in the direction of increasing the compensation and the net present value according to the analysis by the analysis unit Perform reinforcement learning performing department.

As an example, in order to predict the panel temperature of the solar power module due to cooling, the analysis unit uses the actual panel temperature, solar radiation angle, outdoor temperature, and solar radiation data collected as internal measurement data at least through the data storage unit as learning data to predict the next panel temperature. A panel temperature prediction unit to which a neural network model providing as a result is applied may be included.

Furthermore, the analysis unit predicts at least through the panel temperature prediction unit in order to predict the trend of the panel temperature change of the solar power module in the cooling and cooling stopped state, and then uses the panel temperature, solar radiation, outdoor temperature, and solar radiation data as learning data. It may include a long-term panel temperature prediction unit to which a neural network model providing the next panel temperature as a result is applied.

As an example, the data collection unit is the total amount of REC transaction, REC average price, REC starting price, REC closing price, REC reference price, and REC maximum price, which are information related to Renewable Energy Certificate (REC) price, which are collected as external data on electricity price. , Collects at least a plurality of data among the REC lowest data, and the analysis unit uses at least a plurality of data among the total REC transaction volume, REC average price, REC starting price, REC closing price, REC reference price, REC highest price, and REC lowest data at the current time as learning data. Thus, it may include a REC price prediction unit to which a neural network model for predicting the next REC price is applied.

As an example, the analysis unit includes a sprinkling control operating cost calculator that checks the flow rate and power used by the data collection unit as internal measurement data based on the water bill and electricity bill, and calculates the operating cost required to provide sprinkling for cooling. I can.

As an example, the analysis unit includes the estimated REC price predicted through information related to the renewable energy supply certificate (REC) price, which the data collection unit collects as external data on the electric power price, and the system marginal price of generated power by time. : SMP) and the actual generation amount according to the instantaneous power collected as internal measurement data in the data collection unit, and the estimated generation amount predicted using at least some of the outside temperature, the module temperature of the solar panel, insolation, and humidity. It may include a power generation revenue present value calculation unit that calculates power generation revenue.

As an example, the reinforcement learning execution unit includes data set as an influence factor that reflects seasonal and time dependence among the data collected by the data collection unit, an influence factor on power generation profit, an influence factor of efficiency decrease, and an influence factor on the module cooling effect. The executor (actor) to which the optimal control policy neural network model is applied and the state of the execution unit to which the optimal control policy neural network model is applied and outputs the suitability of each control method as a probability value to provide a control signal for panel spraying by learning the state information A value that outputs the expected compensation value as an actual value based on the net present value calculated according to the generation profit and operating cost when the control unit controls it in the environment according to the state information as input and controls it in a preset control method. It may include a critic to which the computational neural network model is applied.

As an example, further comprising a global reinforcement learning performing unit that collects and reflects the learning results learned by the reinforcement learning performing unit and synchronizes the neural network of the reinforcement learning performing unit with its own neural network, and the global reinforcement learning performing unit includes a plurality of different reinforcement learning units. A network with the learning execution units may be configured to asynchronously collect a learning result from each of the plurality of reinforcement learning execution units and synchronize with the reinforcement learning execution unit providing the corresponding learning result.

In the solar power generation module optimal cooling method using rain according to another embodiment of the present invention, the data collection unit includes external data on weather and power prices and power generation costs, a plurality of power generation data related to solar power generation, and internal panel cooling. Measurement data and power plant operation data for panel cooling control are collected, data collection and storage steps in which the data storage unit stores the collected data, and the analysis unit calculates the present value of power generation profits using the collected data stored in the data storage unit. In addition, the analysis step of calculating the operating cost required to provide water for cooling and calculating the net present value from them, and the reinforcement learning performing unit use the net present value obtained in the analysis step as compensation information, and the solar energy in the direction of increasing such compensation. And performing a reinforcement learning step of performing A3C-based deep reinforcement learning using state information related to photovoltaic power generation collected in the data collection unit to generate a control signal for sprinkling the power generation panel.

As an example, the step of performing reinforcement learning is an influence factor that reflects seasonal and time dependence among the data collected by the data collection unit by an actor to which the optimal control policy neural network model is applied, an influence factor on power generation revenue, and an influence factor of efficiency decline. , Collecting data set as the influencing factor for the module cooling effect as state information, learning the state information, and outputting the suitability of each control method as a probability value for providing a control signal for panel sprinkling in the control unit; and When the critic to which the computational neural network model is applied receives the same state information as the execution unit as input, and the control unit controls it in a preset control method in the environment according to the state information, the expected compensation value is applied to the generation revenue and operating cost. It may include the step of outputting the actual value based on the calculated net present value.

In the solar power generation module optimal cooling apparatus and method using rainwater according to the embodiments of the present invention, the actual measurement data obtained from the solar power generation environment is used as state information, and water spray control for improving solar power generation efficiency is executed ( action), and A3C (Asynchronous Advantage Actor-Critic)-based in-depth reinforcement learning method that compensates the net present value calculated by taking into account the increased power generation profit and the cost for spray control according to the improvement of solar power generation efficiency. By applying to the solar power module optimal cooling, it is possible to greatly increase the economic efficiency of power plant operation through optimal watering control that comprehensively considers the amount of power generation, power generation revenue, and watering costs according to the accumulation of learning.

In order to accurately calculate the compensation, which is a learning criterion for the deep reinforcement learning method, the solar power generation module optimal cooling device and method using excellence according to the embodiments of the present invention is a short-term module that is an important influence factor of solar power generation efficiency. Predicts through a temperature prediction neural network, and accurately predicts changes in module temperature by applying a long-term module temperature prediction neural network that uses this short-term prediction module temperature as one of its inputs, and calculates the REC price, an important factor in power generation profits, with various price calculation information. After accurately predicting through a neural network that takes as an input, by accurately predicting the power generation profit by reflecting such sophisticated prediction information, the performance of deep reinforcement learning can be improved, thereby increasing the economic feasibility of the power plant operation.

Furthermore, the solar power module optimal cooling device and method using the excellence according to the embodiments of the present invention is mainly applied to digital games, and the A3C deep reinforcement learning method that can arbitrarily diversify the learning environment cannot diversify the learning environment. In order to solve the problem that the learning result may be biased and performance may be degraded by the limited learning information with regional characteristics as it is applied to the actual solar power plant, a network with the reinforcement learning performing unit for each solar power plant as an agent was constructed. By configuring a separate main reinforcement learning execution unit, the main reinforcement learning execution unit collects and synchronizes the learning information of the reinforcement learning units of several solar power plants in an asynchronous manner, thereby preventing the learning from being biased and improving the learning performance of the deep reinforcement learning method. It has the effect of making it possible.

1 shows a conceptual configuration diagram of a solar power module optimal cooling device using rain according to an embodiment of the present invention.
2 is a conceptual diagram illustrating a solar power module optimal cooling device for explaining an operation method of an analysis unit according to an embodiment of the present invention.
3 is a conceptual diagram showing a neural network model of a panel temperature prediction unit applied to an analysis unit according to an embodiment of the present invention.
4 is a conceptual diagram showing a neural network model of a long-term panel temperature prediction unit applied to an analysis unit according to an embodiment of the present invention.
5 is a conceptual diagram showing a neural network model of a REC price prediction unit applied to an analysis unit according to an embodiment of the present invention.
6 is a conceptual diagram illustrating a specific operation method of an analysis unit according to an embodiment of the present invention.
7 is a conceptual diagram illustrating a method of calculating a power generation profit due to cooling of a solar power module according to an embodiment of the present invention.
8 is a conceptual diagram showing an expanded configuration of a solar power module optimal cooling device according to an embodiment of the present invention.
9 is a conceptual diagram showing a neural network model of an actor configured in a reinforcement learning performing unit of an analysis unit according to an embodiment of the present invention.
10 is a conceptual diagram showing a critic neural network model configured in a reinforcement learning execution unit of an analysis unit according to an embodiment of the present invention.
11 is a flowchart illustrating an operation method of an optimal cooling device for a photovoltaic module 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, the technical terms used in the present invention should be interpreted as generally understood by those of ordinary skill in the technical field to which the present invention belongs, unless otherwise defined in the present invention, and is excessively comprehensive. It should not be construed as a human meaning or an excessively reduced meaning. In addition, when a technical term used in the present invention is an incorrect technical term that does not accurately express the spirit of the present invention, it should be replaced with a technical term that can be correctly understood by those skilled in the art. In addition, general terms used in the present invention should be interpreted as defined in the dictionary or according to the 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 plurality of expressions unless the context clearly indicates otherwise. In the present invention, terms such as “consisting of” or “comprising” should not be construed as necessarily including all of the various elements or various steps described in the invention, and some of the elements or some steps are included. It should be construed that it may not be, or may further include additional components or steps.

In addition, terms including ordinal numbers such as first and second used in the present invention may be used to describe the constituent elements, but the constituent elements should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may be referred to as a first component.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numerals regardless of the reference numerals, and redundant descriptions thereof will be omitted.

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

Meanwhile, the apparatus mentioned 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. Systems, devices, and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), and a PLU ( It may be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit), a microprocessor, or any other device capable of executing and responding to instructions. In addition, such a server or device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one server or device is used, those of ordinary skill in the art, the processing device is a plurality of processing elements and/or multiple types of processing. You can see that it can contain elements. For example, the server or device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor. In addition, the device can control various sensors directly through an interface or through a separate driver or hardware/software board or module separately provided for sensor control, and can cover all of various logical, electrical, and optical modifications for this purpose. have.

Further, 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. For example, an interface, a driving driver, a communication device, a power supply, a microcontroller for control, a storage unit, an operating system or firmware may be included, and sensors may be included if necessary.

On the other hand, the solar power plant includes all types of photovoltaic power generation companies' power generation devices that sell power generated by using photovoltaic power generation modules, and the photovoltaic power generation module includes a photovoltaic power generation panel and directly It means a power generation module including connected electrical devices or an installation frame.

Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1 to 11.

1 shows a conceptual configuration diagram of a solar power module optimal cooling device using rain according to an embodiment of the present invention. This solar power module optimal cooling device can be configured for a specific solar power plant. Basically, after collecting rainwater (rainwater), it is rectified or reprocessed and stored in a storage tank. It is to improve the solar power generation efficiency by removing pollution while lowering the panel temperature of the solar power plant, but to perform an optimized spray control to increase the economic efficiency of the solar power plant.

Basically, the collected rainwater is recycled for the purpose of sprinkling for panel cooling, but if rainwater is insufficient, other water resources such as water and groundwater can be used, and electricity is consumed to sprinkle it, and in some cases, water usage costs may be incurred. have.

Therefore, the solar power module optimal cooling device according to an embodiment of the present invention increases solar power generation efficiency by cooling in order to maintain an optimal power generation effect (maximization of economic efficiency), and thus considers the expected power generation profit. For improvement, it is to generate control information for spray control, such as cooling time, injection amount, injection pressure, applied panel, nozzle used, etc., which are the most economical in consideration of the cost for cooling the photovoltaic module.

For this purpose, the solar power module optimal cooling device according to an embodiment of the present invention includes external data on weather, power prices and power generation costs, a plurality of power generation data related to solar power generation, and internal measurement data related to panel cooling, as shown. And a data collection unit 100 for collecting power plant operation data for panel cooling control, a data storage unit 200 for storing information collected by the data collection unit 100 as a distributed database or a relational database, and data An optimization unit 300 that generates optimal watering control information using collected data stored in the storage unit 200, and an information providing unit that collects and provides various derived information or optimal watering control information of the optimization process to the manager ( 400) may be included.

Here, the optimization unit 300 calculates the present value of power generation revenue, calculates the operating cost required to provide water for cooling, and calculates the net present value from these, and the analysis unit 310 for spraying the solar power panel A controller 330 that provides a control signal and a solar power generation-related collected in the data collection unit to control the controller in a direction of increasing the compensation by using the net present value according to the analysis of the analysis unit 310 as compensation information It includes a reinforcement learning execution unit 320 that performs deep reinforcement learning based on A3C (Asynchronous Advantage Actor-Critic) using the state information.

In practice, the embodiment of the present invention utilizes an A3C-based deep reinforcement learning method as a learning method for generating the most optimal water spray control information for cooling a solar power generation module, but is optimal by reflecting a realistic solar power generation environment and profit structure. It was designed. The A3C-based deep reinforcement learning method, which is usually used to perform digital game play with artificial intelligence, is for a digital game that can configure the learning environment in various ways through simple copying and setting and can predict the next state relatively accurately. In the embodiment of the present invention, this in-depth reinforcement learning method is applied to an actual photovoltaic power generation environment that uses various data that are not normalized, forecast information, and unclear data such as fluctuating weather. It has its characteristics.

In fact, it was difficult to apply the learning method in that the application target was a complex real-world solar power plant, but among these irregular and irregular environmental data, the solar power module was used to determine the influencing factors that influence the optimization of cooling. By performing a process to classify and increase prediction accuracy for them, the actual reinforcement learning execution unit 320 input data is refined, and the Actor-Critic structure, which is the core component of A3C-based deep reinforcement learning, is optimized for actual solar power generation profit improvement. By designing it so that the A3C-based deep reinforcement learning method is suitable for solar power module cooling optimization.

Furthermore, in order to solve the problem that the learning data may be biased and the reliability of the learning result may be lowered, in that unlike digital games that can create various learning environments arbitrarily, the actual environment is the target of learning. The optimal cooling devices for the photovoltaic module are configured as a wide area network, and the learning contents are synchronized in an asynchronous manner to solve the problem of learning bias. This will be described in more detail later with reference to FIG. 8.

A more detailed look at the configuration of FIG. 1 according to an embodiment of the present invention that can provide the technical features of the present invention.

First, the data collection unit 100 largely includes an external data collection unit 110 and an internal measurement data collection unit 120 that collects measurement information of a photovoltaic power generation environment and facilities as internal measurement data, and a control for panel cooling. It includes an internal operation data collection unit 130 for collecting internal operation data, which is information.

The external data collection unit 110 collects meteorological data (providing external meteorological information such as the Meteorological Agency and collecting through open APIs provided by forecasting agencies) used to reflect regional characteristics for solar power generation forecasting, etc. System Marginal Price (SMP), renewable energy supply certificate from external organizations (e.g., Power Exchange (SMP) or renewable one-stop business information integration portal (REC)) to predict or analyze revenue. Renewable Energy Certificate: Various information about the price (REC total transaction volume, REC average price, REC starting price, REC closing price, REC reference price, REC highest price, REC lowest data, etc.), and other external organizations (e.g. KEPCO, Korea National Oil Corporation) ) From the power reserve rate data and oil price information data. In addition, information based on water rates and electricity rates can be collected from external organizations. That is, the external data collection unit 110 collects external data on weather, electricity prices, and generation costs.

The internal measurement data collection unit 120 collects various power generation data necessary for solar power generation, and collects various actual sensor data including temperature and humidity, solar radiation, and panel temperature as environmental sensing information, and increases solar power generation efficiency. As information on the fields, the flow rate used for sprinkling and the flow rate for the amount of water used, the water pressure for sprinkling, the amount of power used for sprinkling, the water level for the stormwater treatment facility and storage water tank, and the driving status of the pump for sprinkling It collects information on pump information about and pressure, water temperature of storage water tank, cooling system control signal and data transmission/reception. Furthermore, as information on the amount of solar power generation, the total power generation through the inverter, the power generation voltage and current values, the voltage and power values of each panel through the individual photovoltaic module panel power sensor, and photovoltaic module image information for measuring the pollution level of the panel, etc. It may include.

The internal operation data collection unit 130 is information related to actual water spray control with a cooling control signal provided through a solar power module optimization cooling device, and information on pump operation that can be used as basic data for further in-depth reinforcement learning ( Pump operation, non-operation, control pressure information), information on the opening and closing information of valves used for spraying, spraying time and spraying pressure for spraying, and information on the nozzles used (information for selecting the optimal nozzle), etc. I can.

Since the collected data of the data collection unit 100 is used as influence factors used in the analysis of the optimization unit 300, a distributed database (eg Hadoop) server 210 and a relational database ( RDBMS) server 220 is divided and stored and managed.

The information providing unit 400 provides analysis results, learning results and control results of the analysis unit 310, the reinforcement learning execution unit 320 and the control unit 330 of the optimization unit 300, and information on each of these processes. It may include an optimization information providing unit 410, and a user UX unit 420 that provides various types of information provided by the optimization information providing unit 410 as visual and auditory information desired by the administrator.

The optimization unit 300 according to an embodiment of the present invention may be divided into an analysis unit 310, a reinforcement learning execution unit 320, and a control unit 330, and substantially the reinforcement learning execution unit 320 is an A3C deep reinforcement Since it is configured based on a learning method, state information (state) is collected from the data storage unit 200 for an actor-critic function, and a reward provided through the analysis unit 310 is collected. In the direction of maximizing this, optimal control information for the controller 330, which is a result of an action, is generated. The controller 330 performs various sprinkling-related controls (pump operation, valve opening and closing, spraying time, spraying pressure, nozzle selection, etc.), as described above, and the reinforcement learning execution unit 320 is provided by the analysis unit 310 Optimal control information is generated in the direction in which the compensation is maximized and provided to the control unit 330, and information collected by the data collection unit 100 through the data storage unit 200 is checked as status information.

To this end, it is necessary to accurately present compensation information that is a learning target of the reinforcement learning performing unit 320.

In an embodiment of the present invention, the net present value obtained by considering both the present value of the generation income and the excellent control operation cost is designed to be used as compensation information of the reinforcement learning execution unit 320.

2 shows a conceptual configuration diagram of a solar power module optimal cooling device for explaining the operation method of the analysis unit 310 according to an embodiment of the present invention.

As shown, in order to optimize the cooling of the photovoltaic module, an influence factor that reflects seasonal and time dependence among various pieces of information collected by the data collection unit 100, an influence factor on power generation profit, an influence factor of efficiency decrease, All factors influencing the cooling effect of the module must be considered.

Among these, excellent control operating costs can be calculated relatively accurately through accurate measured measurement values and determined cost information, but in the case of the present value of power generation revenue, prediction information to predict how much power generation will be obtained based on the current state is required. To That is, the degree to which the solar power module is cooled through sprinkling must be predicted, and the amount of power generation must be predicted through this, so it is based on relatively inaccurate information. Furthermore, the analysis result of the analysis unit is important because it is necessary to calculate through prediction (consideration of relatively accurate SMP price and flexible REC price prediction) how much economic value the expected generation amount obtained through such power generation prediction has.

Accordingly, as shown in the configuration of the analysis unit 310, the REC price prediction unit that predicts at least the REC price and the module temperature are short-term (predicting the next module temperature) and long-term (the module temperature change trend by prediction of the next module temperature). It consists of a variety of prediction units including a module temperature prediction unit that predicts) and a generation amount prediction unit that predicts the amount of power generation (prediction taking into account the module temperature and the current power generation state of the module (contamination degree through image analysis, solar altitude, etc.)). , The result of the generation income present value calculation unit that calculates the present value of the generation income by considering the prediction results of these prediction units, the excellent control operation cost calculation unit used for measurement information and cost information, the generation income present value calculation unit and the excellent control operation cost calculation unit. It includes a net present value generator that generates net present value that can be considered as revenue in practice.

The net present value generated by the net present value generator may be provided as compensation information to the reinforcement learning performing unit 320, and the reinforcement learning performing unit 320 is a control unit based on the sophisticated compensation information of the analysis unit 310. Generates optimal control information for (320).

3 to 5 illustrate the concept of a neural network model applied to each prediction unit configured in the analysis unit 310.

First, one of the important influencing factors in the embodiments of the present invention is the temperature of the panel that is cooled according to sprinkling, and the cooling information according to sprinkling, the state in which the cooling effect is maintained after sprinkling is terminated, is based on this. Generation profit becomes accurate.

Therefore, as shown in FIGS. 3 and 4, the panel temperature prediction of the solar power module is the neural network model shown in FIG. 3 and the neural network shown in FIG. 3 for predicting the next (t) panel temperature based on information according to the current power generation situation. Using the output of the model as one of the inputs, the neural network model shown in FIG. 4 is used to predict the next (t+1) panel temperature.

As shown in FIG. 3, the short-term panel temperature prediction unit 311 is the measured previous panel temperature (currently measured final temperature) (RMT _{t - 1} ), solar radiation (SAA), outdoor temperature (AT), and solar radiation (RR). ) As an input (X, learning data) and outputs the next panel temperature (RMT _t ) through a panel temperature prediction model neural network modeled as a hidden layer composed of a fully connected network. Provided as (Y). The learning rate, batch size, and number of learning can be defined through the number of neurons in the hidden layer.

Here, the solar angle (SAA) is the solar altitude obtained according to the location, date and time of the panel, and the outside temperature (AT) and the amount of solar radiation (RR) are measured using actual data measured from an actual sensor.

The long-term panel temperature prediction unit 312 as shown in FIG. 4 is a panel temperature (RMT _t ), an insolation angle (SAA), and an outside temperature (AT) predicted by the short-term panel temperature prediction unit 311 of FIG. ) And influencing factors including insolation (RR) as inputs (X, learning data), and the panel temperature prediction model modeled as a hidden layer composed of a fully connected network. RMT _{t + 1} ) is provided as an output (Y). The learning rate, batch size, and number of learning can be defined through the number of neurons in the hidden layer.

Through this first and second multi-panel temperature analysis, you can also use information on the degree of temperature change of the panel to derive more precise expected power generation or check the effect on the actual power generation, and compare it with the actual panel temperature. You can check whether the effect ends.

5 is a conceptual diagram showing a neural network model of the REC price prediction unit 313 applied to the analysis unit according to an embodiment of the present invention, as shown in the new renewable energy collected as external data on the power price by the data collection unit Supply Certificate (REC) price-related information: Total REC transaction volume (TQ _t ), average REC price (ARP _t ), REC starting price (SRP _t ), REC closing price (EP _t ), REC reference price (STDP _t ), REC REC price prediction model modeled as a hidden layer composed of a fully connected network with the highest price (MXRP _t ) and REC lowest price (MNRP _t ) data as input (X, learning data). Provides the price (ARP _t+1 ) as output (Y). The learning rate, batch size, and number of learning can be defined through the number of neurons in the hidden layer.

On the other hand, although not specifically shown, the generation amount prediction unit uses a convolutional neural network (CNN) based on a panel image (one of the internal measurement data) of the solar power module collected from the data collection unit 100 and stored in the data storage unit 200. The pollution degree is calculated from the input panel image by learning through a neural network, and the degree of insolation (RR), the temperature of the module (RMT), and solar altitude (SAA) obtained through the data collection unit 100 or data storage unit 200 are calculated. An instantaneous power prediction model modeled as an input (X, learning data) as a hidden layer composed of a fully connected network. It can be composed of a multi-complex neural network model that obtains the expected solar power generation amount as an output through a neural network. have.

In this way, the prediction unit of the analysis unit 310 calculates the present value of power generation revenue using a plurality of prediction means, and substantially more information is considered together. A more specific process of generating the net present value by the analysis unit 310 will be described with reference to FIG. 6.

6 is a conceptual diagram illustrating a specific operation method of an analysis unit according to an embodiment of the present invention. Since the operation method of the illustrated analysis unit 310 and the used influencing factors (collection information, etc.) may be partially changed according to the actual configuration or design conditions, the illustrated configuration is limited to this specific configuration, although it relates to one of the preferred methods. It doesn't work.

Looking at the illustrated configuration, first, the excellent control operation cost calculation unit of the analysis unit 310 checks the flow rate and power used by the data collection unit as internal measurement data based on the water charge and the electric charge, and provides water spray for cooling. Calculate the required operating costs.

The power generation revenue present value calculation unit checks the actual generation amount from instantaneous power information collected as internal measurement data by the data collection unit, and outside temperature information, module temperature information, solar radiation information, and solar altitude information collected as internal measurement data by the data collection unit Using the module temperature prediction units 311 and 312, the predicted next module temperature and the next predicted module temperature are calculated. Here, the end point of the sprinkling effect is a point when the actual module temperature reaches the predicted module temperature, and can be confirmed by comparing the measured module temperature information with the predicted module temperature. If there is a module that is cooled by sprinkling and a reference module that is not cooled, it is possible to further confirm whether or not the sprinkling effect is maintained through the temperature difference between them.

In addition, the predicted power generation amount can be obtained through the power generation amount predicting unit using the pollution level of the panel, the temperature of the module, the amount of solar radiation, and the solar altitude information, and the module temperature predicting unit 311, 312 predicts the module temperature as the temperature of the module applied to Temperature can be used.

On the other hand, the generation revenue present value calculation unit predicts the REC price-related information collected as external data from the data collection unit through the REC price prediction unit 313 to calculate a new renewable energy supply certificate (REC) estimated price, and the data collection unit The SMP price is checked from the system sales price (SMP) information of generated power by time, which is collected as external data from, and based on the predicted power generation amount and actual power generation amount, the estimated amount of additional power generated by cooling according to spraying and the corresponding economic value is calculated. do.

The net present value generator generates the net present value by subtracting the required cost from the additional gain by taking into account the present value of the power generation revenue including the economic value that can be obtained additionally by cooling and the excellent control operation cost required for such cooling.

This net present value is provided to the reinforcement learning execution unit 320 as compensation information, and the reinforcement learning execution unit 320 checks the actual measured values such as module temperature information, insolation information, and humidity information as status information, and this compensation information is A method of spraying related control (pump operation, valve opening, spraying time, spraying pressure, nozzle selection, etc.) is determined so as to be maximum and provided to the controller 330 as optimal control information. Meanwhile, the reinforcement learning execution unit 320 refers to information on the control of the control unit 330 performed in the past (information collected as power plant operation data by the data collection unit) as past excellent control data.

7 is a conceptual diagram for explaining a method of calculating the power generation profit due to cooling of the solar power module according to an embodiment of the present invention, and the profit obtained by cooling the panel of the solar power module is an increase in solar power generation amount due to cooling. Calculate the REC price and the SMP price, and calculate the water consumption and power consumption as cost, the difference becomes the power generation profit. In the graph shown, the shaded portion is the total power generation increased by panel cooling according to water spraying, and when this increased total power generation increase (PGI: Power Generation Increase) is multiplied by the REC price or SMP price (PGI*REC&SMP=Profit), profit Can be obtained.

If there is a water flow rate among the used flow rates, water tax (WT) can be calculated by multiplying the water rate standard information, and the electricity bill (ET) can be obtained by multiplying the power consumption required for watering through the pump with the electricity rate standard information. It becomes the operating cost (WT+ET=Expense).

This excellent control operation cost refers to the cost incurred from the time when spraying starts (CST) to the time when spraying ends (CET) in the graph, and the increased total power generation (PGI) is the time when spraying starts (CST). It occurs until the point at which the sprinkling effect ends (CEET). The time point CEET at which the sprinkling effect ends may be defined as a time point at which the predicted module temperature MTPM(t) is equal to or lower than the actual module temperature RMT(t).

This can be seen from FIG. 7B that when the watering effect occurs, the actual module temperature is lower than the predicted module temperature. If the actual module temperature is equal to or higher than the predicted module temperature, it can be seen that the watering effect is terminated. have.

As described above, the analysis unit 310 of the optimization unit 300 predicts the correct compensation information and provides it to the reinforcement learning execution unit 320, and control spraying through the control unit 330 while checking the actual photovoltaic power generation status information. As the process of performing is repeated, the control information provided to the control unit 330 is optimized so that the net profit corresponding to the compensation increases by the A3C deep reinforcement learning.

However, the reinforcement learning execution unit 320 applied to the solar power module optimal cooling device according to the embodiment of the present invention performs learning in a real environment rather than a virtual environment, and the season changes due to the characteristics of solar power generation. This takes place and the local characteristics are strongly reflected, so the contents learned for a certain period are mostly similar environments, resulting in a problem that the learning data is limited or biased.

In the case of A3C deep reinforcement learning applied to the present invention, it was originally developed for a digital game, and if necessary, a virtual subagent can be duplicated as desired so that each agent can independently play the game and have different experiences. Although it is possible to prevent the training data from being biased by using the training data according to various experiences, there is a limit to improving the performance when applied to an actual limited environment as in the present invention.

Accordingly, the present invention extends such a reinforcement learning performing unit. That is, a separate global reinforcement learning execution unit is formed, and the learning information of the reinforcement learning execution units of individual solar power plants having different experiences is performed globally by using the reinforcement learning execution units configured in the actual solar power plant as subagents. Make it possible to reflect wealth.

8 is a conceptual diagram showing an expanded configuration of a solar power module optimal cooling device according to an embodiment of the present invention.

As shown, the main system 320_0 corresponds to the global reinforcement learning execution unit, and the sub-systems 320_1 to 302_n correspond to the reinforcement learning execution units of each solar power plant. The internal configurations of the reinforcement learning performing units (executing units and critics units) constituting the main system or sub-system correspond to the same neural network model.

Here, synchronization of the subsystem and the main system and asynchronous update of the global network are performed. For example, the reinforcement learning execution unit of an individual solar power plant corresponding to the subsystem 1 (320_1) learns its own experience and provides the learning information to the global reinforcement learning execution unit corresponding to the main system 320_0, and the main system (320_0) updates the neural network of the sub-system 1 (320_1) by reflecting the provided learning information to the neural network and synchronizing its own neural network with the neural network of the sub-system 1 (320_1). However, it does not synchronize with other subsystems.

If all the subsystems perform this synchronization process at the same time, only the size of the reinforcement learning execution unit is increased, and it is difficult to achieve the goal of learning multiple experiences in a neural network. Therefore, the process of synchronizing with the main system between each subsystem is performed asynchronously. For example, when the neural network of the sub-system 1 320_1 and the neural network of the main system 320_0 synchronize previously, the other sub-systems 320_2 to 320_n are not synchronized. Thereafter, when the sub-system n (320_n) synchronizes with the main system (320_0), the main system (320_0) maintains the latest neural network in which all the learning information is reflected, but the sub-system 1 (320_1) still performs training with the previous version of the neural network. . In this way, since the version of the neural network used between each sub-system is different, learning information is generated while each experiencing different experiences through different neural networks, and the main system 320_0 reflects the learning information gathered from these various experiences, and As the subsystems are updated asynchronously, all systems reflect the learning of various experiences in the direction of achieving their goals, so that high performance can be provided within a relatively short time.

An actor of the reinforcement learning performing unit learning the optimal control policy may be designed as a neural network model as shown in FIG. 9.

As shown, the execution unit 321 includes the solar altitude (SAA _CST ), which is an influencing factor reflecting the dependence of season and time based on the start time of sprinkling control, among the data collected by the data collection unit, and insolation ( RR _CST ), module temperature measurement value (RMT _CST ) and humidity measurement value (H _CST ), which are factors influencing efficiency reduction, and water temperature measurement value of storage tank (SWT _CST ), which is an influence factor on module cooling effect, are collected as status information By using these as inputs (X, learning data), each control method to provide control signals for panel sprinkling through the optimal control policy modeled as a hidden layer composed of a fully connected network The fit is output as a probability value. As for the activation function of the neural network, the last hidden layer used the Softmax function to derive the probability value, and the others were designed using ReLU. Excluding the last hidden layer, the activation function can be changed according to the locality and data characteristics of the plant.

10 is a conceptual diagram showing a critic neural network model 322 configured in a reinforcement learning performing unit of an analysis unit according to an embodiment of the present invention.

As shown, the same state information as the execution unit 321: solar altitude (SAA _CST ), solar radiation (RR _CST ), module temperature measurement value (RMT _CST ), humidity measurement value (H _CST ), water temperature measurement value of the storage tank (SWT _CST ) as input (X, learning data) and the expected compensation value when the control unit controls it with a preset control method in the environment according to this state information, a hidden layer composed of a fully connected network ), and output through the neural network model for calculating values. At this time, the compensation value is determined based on the net present value calculated based on generation revenue (REC and SMP revenue based on power generation) and operating cost (water tax + electricity tax) so that it can be used for learning. Unlike the neural network of the execution unit 321, since the output value is an actual value rather than a probability value, a separate Softmax function is not used, and a ReLU function is used for the entire hidden layer.

11 is a flowchart illustrating an operation method of an optimal cooling device for a photovoltaic module according to an embodiment of the present invention.

First, the data collection unit collects external data on weather and electricity prices and generation costs, a plurality of generation data related to solar power generation, internal measurement data related to panel cooling, and power plant operation data for panel cooling control, and the data storage unit Save the collected data.

The analysis unit of the optimization department predicts the panel temperature of the solar power module in the short and long term, predicts the amount of generation by reflecting the predicted panel temperature, and predicts the REC price. All of these prediction processes are performed through a learning model through a neural network, and these prediction models may be added or partially omitted, and more accurate prediction becomes possible as learning is performed.

Thereafter, the analysis unit of the optimization unit calculates the present value of generation revenue by using the collected data stored in the data storage unit and the prediction information obtained through the prediction process, and calculates the operating cost according to the flow rate and power use required for stormwater control.

Then, the analysis unit of the optimization unit calculates the net present value by taking into account the present value of generation revenue and the operating cost of excellent control, and generates it as compensation information and provides it to the reinforcement learning execution unit.

The reinforcement learning performing unit uses the state information related to photovoltaic power generation collected in the data collection unit to generate a control signal for sprinkling solar power panels in the direction of increasing the compensation and the net present value as compensation information. Perform.

By generating optimal control information according to this reinforcement learning and providing it to the control unit, the control unit cools the photovoltaic power generation module through excellent control, and this control information is accumulated and learned so that the tank for watering is developed in a way that can gradually earn It controls, controls the flow rate, and controls which panel to use, which nozzle to spray, and how much the pressure is used.

Although not shown, the reinforcement learning execution unit can be updated with a reinforcement learning neural network learned through various experiences by interlocking with the global reinforcement learning execution unit, thereby greatly increasing its reliability.

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 any person with ordinary knowledge in the technical field to which the present invention pertains without departing from the gist of the present invention appended in the claims will be able to implement various modifications. .

100: data collection unit 110: external data collection unit
120: internal measurement data collection unit 130: internal operation data collection unit
200: data storage unit 210: distributed database server
220: relational database server 300: optimization unit
310: analysis unit 320: reinforcement learning execution unit
330: control unit 400: information providing unit
410: optimization information providing unit 420: user UX unit

Claims (10)

A data collection unit for collecting external data on weather, electricity prices and generation costs, a plurality of generation data related to solar power generation, internal measurement data related to panel cooling, and power plant operation data for panel cooling control;
A data storage unit for storing the information collected by the data collection unit in a distributed database or a relational database;
An analysis unit that calculates a present value of generation revenue using collected data stored in the data storage unit, calculates an operating cost required to provide water for cooling, and calculates a net present value therefrom;
A control unit for providing a control signal for spraying solar power panels;
Reinforcement of performing A3C-based in-depth reinforcement learning using the state information related to photovoltaic power generation collected in the data collection unit so that the net present value according to the analysis of the analysis unit is used as compensation information and the control unit is controlled in a direction to increase such compensation. Including a learning execution unit,
The data collection unit,
Renewable Energy Certificate (REC) price-related information collected as external data on electricity prices: at least among the total REC transaction volume, REC average price, REC starting price, REC closing price, REC reference price, REC high price, REC lowest data Collects multiple data,
The analysis unit,
A REC price prediction unit to which a neural network model for predicting a next REC price using data of total REC transaction volume, average REC price, REC start price, REC closing price, REC reference price, REC highest price, and REC lowest price at the present time as training data is applied;
In order to predict the panel temperature of the photovoltaic module due to cooling, at least the actual panel temperature, solar radiation, outdoor temperature and solar radiation data collected as internal measurement data through the data storage are used as learning data, and the next panel temperature is provided as a result. Short-term panel temperature prediction unit to which a neural network model is applied;
In order to predict the trend of changes in the panel temperature of the solar power module in the cooling and cooling stopped state, at least the short-term panel temperature predictor is used to predict the panel temperature, solar radiation angle, outdoor air temperature, and solar radiation data as learning data. A long-term panel temperature prediction unit to which a neural network model providing as a result is applied;
The panel image of the photovoltaic power generation module is learned through a neural network based on a convolutional neural network (CNN) to calculate the pollution level from the input panel image, and the instantaneous power prediction model neural network is constructed using the pollution level, solar radiation, module temperature, and solar altitude as training data. A generation amount prediction unit that obtains the estimated solar power generation amount as an output through; And
Renewable energy supply certificate (REC) price-related information that the data collection unit collects as external data on the electric power price, the REC estimated price predicted by the REC price prediction unit, and the system sales price of generated power by time (System Marginal Price: SMP), the actual generation amount according to the instantaneous power collected by the data collection unit as internal measurement data, the outside temperature, the module temperature of the solar power generation panel predicted by the short-term and long-term panel temperature prediction unit, the solar radiation amount, Solar power generation module optimal cooling device using rain, characterized in that it comprises a generation profit present value calculation unit that calculates the generation profit using the estimated generation amount predicted using humidity.
delete delete delete The water spray control operating cost of claim 1, wherein the analysis unit checks the flow rate and power used, which are collected as internal measurement data by the data collection unit, based on a water charge and an electric charge, and calculates an operating cost required to provide water for cooling. Solar power module optimal cooling device using rain, characterized in that it comprises a calculation unit.
delete The method of claim 1, wherein the reinforcement learning performing unit
Among the data collected by the data collection unit, data set as an influence factor reflecting the dependence of season and time, an influence factor on power generation profit, an influence factor of lowering efficiency, and an influence factor on the module cooling effect are collected as state information, and the state An executor (actor) to which an optimal control policy neural network model is applied for learning information and outputting a suitability for each control scheme for providing a control signal for panel spraying by the control unit as a probability value;
When the control unit controls the controller in a preset control method in the environment according to the status information by inputting the same status information as the execution unit, the expected compensation value is based on the net present value calculated according to the generation revenue and operating cost. A solar power generation module optimal cooling device using rain, characterized in that it includes a critic to which a value calculation neural network model is applied to output an actual value.
The method of claim 1, further comprising a global reinforcement learning performing unit for collecting and reflecting the learning result learned by the reinforcement learning performing unit and synchronizing the neural network of the reinforcement learning performing unit with its own neural network, wherein the global reinforcement learning performing unit is By configuring a network with a plurality of other reinforcement learning execution units, the learning result is asynchronously collected from each of the plurality of reinforcement learning execution units, and synchronization is performed with the reinforcement learning execution unit providing the corresponding learning result. Solar power module optimal cooling device.
In the solar power module optimal cooling method using the solar power module optimal cooling device according to any one of claims 1, 5 and 7,
The data collection unit collects external data on weather and electricity prices and generation costs, a plurality of generation data related to solar power generation, internal measurement data related to panel cooling, and power plant operation data for panel cooling control, and the data storage unit collects the data. A data collection and storage step of storing the generated data;
An analysis step of calculating, by an analysis unit, a present value of generation revenue using the collected data stored in the data storage unit, calculating an operating cost required to provide water for cooling, and calculating a net present value therefrom;
The reinforcement learning performing unit uses the state information related to photovoltaic power collected in the data collection unit so that the net present value obtained in the analysis step is used as compensation information and a control signal for sprinkling the solar power generation panel is generated in a direction that increases the compensation. A solar power module optimal cooling method using rainwater comprising the step of performing reinforcement learning to perform A3C-based deep reinforcement learning.
The method of claim 9, wherein performing the reinforcement learning
Influence factors that reflect seasonal and time dependence among the data collected by the data collection unit by the actor to which the optimal control policy neural network model is applied, factors that affect power generation profits, factors that affect efficiency decline, and influence on module cooling effects Collecting data set as a factor as state information, learning the state information, and outputting a suitability of each control method as a probability value for providing a control signal for panel sprinkling;
When the critic to which the value estimation neural network model is applied receives the same status information as the execution unit as input, and the control unit controls the control in a preset control method in the environment according to the status information, the expected compensation value is generated and operated Optimal cooling method for solar power modules using rain, comprising the step of outputting an actual value based on a net present value calculated according to cost.

Images