A photovoltaic intelligent spraying method and system based on the Internet of Things

Abstract

The application provides a photovoltaic intelligent spraying method and system based on the Internet of Things, and relates to the technical field of photovoltaic power generation. The method comprises the following steps: in a test process, acquiring self-state test data and surrounding environment test data of a test photovoltaic panel at multiple test time points; obtaining a trained power generation data prediction model; acquiring self-state test data, surrounding environment test data and spraying water quantity data of the test photovoltaic panel before and after multiple spraying tests; obtaining a trained spraying control model; collecting self-state data and surrounding environment data of the photovoltaic panel in a process of generating power by using the photovoltaic panel; determining whether spraying is needed; and determining an optimized spraying water quantity. According to the application, whether spraying is needed can be determined based on the self-state data and the surrounding environment data of the photovoltaic panel, and the optimized spraying water quantity is determined. The accuracy and pertinence of the spraying water quantity optimization are improved, and the efficiency of spraying cooling and photovoltaic power generation is improved.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and in particular to a photovoltaic intelligent spraying method and system based on the Internet of Things. Background Technology

[0002] Photovoltaic power generation, as a clean energy source, is widely used globally. During operation, when the surface temperature of photovoltaic panels rises, their power generation efficiency drops significantly. In such cases, water spraying is necessary to cool the panels. Current technologies typically rely on the experience of operators to determine whether water spraying is needed and to decide the appropriate water volume. This human judgment is inherently random, leading to inaccurate water volume determination and low spraying efficiency. It is difficult to optimize the water volume by considering the condition of the photovoltaic panels and the surrounding environment, resulting in low photovoltaic power generation efficiency.

[0003] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a photovoltaic intelligent spraying method and system based on the Internet of Things, which can solve the technical problem that it is difficult to determine the optimal spraying water volume by combining the condition of the photovoltaic panel itself and the surrounding environment.

[0005] According to a first aspect of the present invention, a photovoltaic smart sprinkler method based on the Internet of Things is provided, comprising:

[0006] During the test, at multiple test moments, test data on the photovoltaic panel's own status and the surrounding environment were acquired. The surrounding environment test data included the irradiance received by the photovoltaic panel, the ambient temperature of the environment where the photovoltaic panel was located, the ambient humidity, the wind speed, and the angle between the wind direction and the photovoltaic panel's orientation. The self-status test data included the photovoltaic panel's own temperature and the photovoltaic panel's power generation.

[0007] Based on the test data of the photovoltaic panel's own condition and the test data of the surrounding environment, the power generation data prediction model is trained to obtain the trained power generation data prediction model.

[0008] Before and after multiple spray tests, test data on the photovoltaic panel's own condition, the surrounding environment, and the spray water volume were obtained respectively.

[0009] Based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, the spray control model is trained to obtain the trained spray control model.

[0010] During the process of generating electricity using photovoltaic panels, data on the photovoltaic panels' own status and the surrounding environment are collected. The surrounding environment data includes the irradiance data received by the test photovoltaic panels, the ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the orientation of the photovoltaic panels. The photovoltaic panels' own status data includes the photovoltaic panels' own temperature data and the photovoltaic panels' power generation. The photovoltaic panels are of the same model as the test photovoltaic panels.

[0011] Based on its own status data, surrounding environment data, and the trained power generation data prediction model, it is determined whether spraying is necessary.

[0012] If spraying is required, the optimal spraying water volume is determined based on its own status data, surrounding environment data, and the trained spraying control model.

[0013] According to the present invention, obtaining a trained power generation data prediction model includes:

[0014] Based on irradiance test data and its own temperature test data, information on the test of directly influencing factors was obtained;

[0015] Based on ambient temperature test data, wind speed test data, and test angle, information on indirect influencing factors was obtained.

[0016] Obtain the test information of indirect influencing factors from the (i-k+1)th test time to the ith test time and the test information of direct influencing factors at the ith test time, where i and k are positive integers;

[0017] By using the self-attention mechanism of the power generation data prediction model, the test information of indirect influencing factors from the i-k+1 test time to the i test time is processed to obtain the comprehensive environmental impact information from the i-k+1 test time to the i test time.

[0018] By splicing the comprehensive environmental impact information at the i-th test time with the direct influencing factor test information at the i-th test time, the power generation efficiency impact information at the i-th test time is obtained.

[0019] Input the power generation efficiency impact information at the i-th test time into the first multilayer sensing network layer of the power generation data prediction model to obtain the predicted power generation from the i-th test time to the (i+j)-th time, where j is a positive integer;

[0020] The comprehensive environmental impact information from the (i-k+1)th test time to the ith test time is input into the 1D convolutional sub-model of the power generation data prediction model to obtain the predicted comprehensive environmental impact information from the (i+1)th test time to the (i+j)th test time.

[0021] Based on the predicted power generation, power generation test data, and predicted comprehensive environmental impact information from the i-th test time to the (i+j)-th test time, the loss function of the power generation data prediction model is determined.

[0022] Based on the loss function of the power generation data prediction model, the power generation data prediction model is trained to obtain the trained power generation data prediction model.

[0023] According to the present invention, determining the loss function of a power generation data prediction model includes:

[0024] Based on the predicted power generation and power generation test data from the i-th time to the (i+j)-th test time, determine the power generation data loss function;

[0025] The comprehensive environmental impact information from the i-th test time to the (i+j)-th test time is input into the second multilayer sensing network layer to obtain the predicted environmental temperature, predicted wind speed and predicted angle from the i-th test time to the (i+j)-th test time.

[0026] Based on the predicted ambient temperature, predicted wind speed, predicted angle, ambient temperature test data, wind speed test data, and test angle from the i-th test time to the (i+j)-th test time, determine the prediction loss function for influencing factors;

[0027] Based on the power generation data loss function and the influencing factor prediction loss function, the loss function of the power generation data prediction model is obtained.

[0028] According to the present invention, determining the prediction loss function for influencing factors includes:

[0029] According to the formula

[0030] ,

[0031] Determine the influencing factors and predict the loss function ,in, Let be the predicted ambient temperature at the (i+s)th test time. Let be the predicted wind speed at the (i+s)th test time. Let be the predicted angle at the (i+s)th test time. For the reason , and The vector formed This refers to the ambient temperature test data at the (i+s)th test time. The wind speed test data is for the (i+s)th test time. Let be the angle between the test points at the (i+s)th test time. For the reason , and The vectors formed by s and j are sim, where s ≤ j and s is a positive integer.

[0032] According to the present invention, obtaining a trained sprinkler control model includes:

[0033] Based on the surrounding environment test data before the spray test, determine the irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data and test angle, and based on the self-state test data before the spray test, determine the self-temperature test data.

[0034] Based on the irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, test angle and self-temperature test data before the spray test, the cooling effect information is determined;

[0035] Based on the temperature test data before and after the spray test, determine the cooling result information;

[0036] Based on the information on the impact and results of cooling, spraying prediction information is obtained;

[0037] Based on the sprinkler prediction information and sprinkler water volume data, the loss function of the sprinkler control model is determined;

[0038] The sprinkler control model is trained based on the loss function of the sprinkler control model to obtain the trained sprinkler control model.

[0039] According to the present invention, determining the loss function of the sprinkler control model includes:

[0040] According to the formula

[0041] ,

[0042] Determine the loss function of the sprinkler control model. ,in, For sprinkler forecast information, For spray water volume data, denoted as the maximum water volume for a single spray cycle, and max is the function that takes the maximum value.

[0043] According to the present invention, determining whether spraying is required includes:

[0044] By inputting its own state data and surrounding environment data into the trained power generation data prediction model, the expected power generation is obtained.

[0045] If the temperature data is higher than or equal to the temperature threshold, or if the relative difference between the expected power generation and the actual power generation is greater than the preset error threshold, then spraying is required.

[0046] According to the present invention, determining the optimal spray water volume includes:

[0047] The target's own temperature is determined based on the predicted model of power generation data after training and the surrounding environmental data;

[0048] Input the surrounding environmental data, the target's own temperature data, and the target's own temperature into the trained sprinkler control model to determine the first candidate sprinkler water volume.

[0049] The second candidate spray volume is determined based on the relative difference between the expected power generation and the power generation and the maximum single spray volume.

[0050] The maximum value of the first and second candidate spray water volumes is determined as the optimal spray water volume.

[0051] According to the present invention, determining the temperature of a target itself includes:

[0052] The simulated value of its own temperature data and the surrounding environment data are input into the trained power generation data prediction model to obtain the predicted power generation value at multiple future times. The simulated value of its own temperature data includes a value equal to its own temperature data, and the simulated value of its own temperature data is higher than the temperature of the spray water.

[0053] By fitting the predicted power generation value with multiple future time points, a regular function of power generation versus time can be obtained;

[0054] By integrating the regularity function, the power generation corresponding to the simulated value of its own temperature data is obtained;

[0055] According to the formula

[0056] ,

[0057] To achieve the maximum cooling efficiency E, where, The power generation corresponding to the simulated value of the xth self-temperature data. The amount of electricity generated is equal to the value corresponding to its own temperature data. Let x be the simulated value of its own temperature data. The value is equal to its own temperature data, N is the number of simulated values ​​of its own temperature data, max is the maximum value function, x≤N, and x and N are both positive integers;

[0058] The minimum value between the simulated value of the self-temperature data corresponding to the maximum cooling efficiency and the self-temperature threshold is determined as the target self-temperature.

[0059] According to a second aspect of the present invention, an Internet of Things-based photovoltaic smart sprinkler system is provided, comprising:

[0060] The first acquisition module acquires test data of the photovoltaic panel's own status and test data of the surrounding environment at multiple test moments during the test process. The test data of the surrounding environment includes test data of the irradiance received by the photovoltaic panel, test data of the ambient temperature of the environment where the photovoltaic panel is located, test data of the ambient humidity, test data of the wind speed, and test angle between the wind direction and the orientation of the photovoltaic panel. The test data of the photovoltaic panel's own status includes test data of the photovoltaic panel's own temperature and test data of the photovoltaic panel's power generation.

[0061] The first training module trains the power generation data prediction model based on the test data of the photovoltaic panel's own state and the test data of the surrounding environment, and obtains the trained power generation data prediction model.

[0062] The second acquisition module acquires test data on the photovoltaic panel's own status, test data on the surrounding environment, and data on the amount of water sprayed before and after multiple spray tests.

[0063] The second training module trains the spray control model based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, and obtains the trained spray control model.

[0064] The data acquisition module collects data on the photovoltaic panel's own status and the surrounding environment during the photovoltaic panel power generation process. The surrounding environment data includes the irradiance data received by the test photovoltaic panel, the ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the photovoltaic panel's orientation. The photovoltaic panel's own status data includes the photovoltaic panel's own temperature data and the photovoltaic panel's power generation capacity. The photovoltaic panel is the same model as the test photovoltaic panel.

[0065] The judgment module determines whether spraying is necessary based on its own status data, surrounding environmental data, and the trained power generation data prediction model.

[0066] The sprinkler water volume optimization module determines the optimal sprinkler water volume based on its own status data, surrounding environment data, and the trained sprinkler control model if sprinkler operation is required.

[0067] By adopting the above technical solution, the present invention can achieve the following technical effects:

[0068] According to this invention, test data on the photovoltaic panel's own state and surrounding environment can be acquired at multiple test moments to obtain a trained power generation data prediction model. Furthermore, test data on the photovoltaic panel's own state, surrounding environment, and spray water volume can be acquired before and after multiple spray tests to obtain a trained spray control model. During the photovoltaic panel power generation process, data on the photovoltaic panel's own state and surrounding environment are collected to determine whether spraying is necessary and to optimize the spray water volume. Combining the photovoltaic panel's own state and the surrounding environment to determine the optimal spray water volume improves the accuracy and relevance of spray water volume optimization, thereby increasing the efficiency of spray cooling and photovoltaic power generation. Moreover, during the testing process, test data on the photovoltaic panel's own state and surrounding environment can be acquired at multiple test moments under various seasons and conditions, providing basic data for training the power generation data prediction model. When training the power generation data prediction model, the power generation data loss function and the influencing factor prediction loss function can be determined based on the test data of the photovoltaic panel's own state and the surrounding environment. These are then weighted and summed to obtain the loss function of the power generation data prediction model, which is then used to train the model. Considering that indirect influencing factor test information from historical moments only affects indirect influencing factor test information from future moments, an influencing factor prediction loss function is determined based on a self-attention mechanism and 1D convolution processing. This function assists in training the power generation data prediction model, improving the accuracy and comprehensiveness of the training and enhancing the model's performance. Similarly, when training the sprinkler control model, the loss function can be determined based on cooling impact information and cooling result information. The sprinkler control model is then trained to obtain the trained model. Considering that the closer the spray water volume is to the maximum single spray water volume, the smaller the change in cooling efficiency, weights are assigned to the differences between different spray water volume data and spray prediction information. This determines the loss function of the spray control model, improving the accuracy and relevance of training and enhancing the performance of the spray control model. Furthermore, based on its own state data, surrounding environmental data, and the trained power generation data prediction model, it can be determined whether spraying is necessary. Based on its own state data, surrounding environmental data, and the trained spray control model, a first and second candidate spray water volume are determined, resulting in an optimized spray water volume. Considering that spraying must ensure that power generation reaches the desired high power output, the spray water volume is set based on the efficiency of power generation improvement caused by cooling. This improves the accuracy and relevance of spray water volume optimization, enhancing the efficiency of spray cooling and photovoltaic power generation.

[0069] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0071] Figure 1 A schematic flowchart of an IoT-based smart photovoltaic sprinkler method according to an embodiment of the present invention is shown as an example.

[0072] Figure 2 An exemplary flowchart of obtaining a trained power generation data prediction model according to an embodiment of the present invention is shown;

[0073] Figure 3 A block diagram of an Internet of Things-based photovoltaic smart sprinkler system according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0076] Figure 1 An exemplary flowchart of an IoT-based smart photovoltaic sprinkler system according to an embodiment of the present invention is shown, including:

[0077] Step S1: During the test, at multiple test moments, acquire test data on the photovoltaic panel's own status and test data on the surrounding environment. The surrounding environment test data includes test data on the irradiance received by the photovoltaic panel, test data on the ambient temperature of the environment where the photovoltaic panel is located, test data on the ambient humidity, test data on the wind speed, and test angle between the wind direction and the orientation of the photovoltaic panel. The test data on the photovoltaic panel's own status includes test data on the photovoltaic panel's own temperature and test data on the photovoltaic panel's power generation.

[0078] Step S2: Based on the test data of the photovoltaic panel's own state and the test data of the surrounding environment, train the power generation data prediction model to obtain the trained power generation data prediction model.

[0079] Step S3: Before and after multiple spray tests, acquire test data on the photovoltaic panel's own condition, test data on the surrounding environment, and data on the spray water volume.

[0080] Step S4: Based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, train the spray control model to obtain the trained spray control model.

[0081] Step S5: During the process of using photovoltaic panels to generate electricity, collect the photovoltaic panel's own status data and surrounding environment data. The surrounding environment data includes the irradiance data received by the test photovoltaic panel, the ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the photovoltaic panel's orientation. The photovoltaic panel's own status data includes the photovoltaic panel's own temperature data and the photovoltaic panel's power generation. The photovoltaic panel is the same model as the test photovoltaic panel.

[0082] Step S6: Determine whether spraying is necessary based on its own status data, surrounding environment data, and the trained power generation data prediction model.

[0083] Step S7: If spraying is required, determine the optimal spray water volume based on its own state data, surrounding environment data, and the trained spray control model.

[0084] The IoT-based photovoltaic intelligent sprinkler method and system according to embodiments of the present invention can acquire test data on the photovoltaic panel's own state and the surrounding environment at multiple test moments, thereby obtaining a trained power generation data prediction model. Furthermore, before and after multiple sprinkler tests, it acquires test data on the photovoltaic panel's own state, the surrounding environment, and the sprinkler water volume, thereby obtaining a trained sprinkler control model. During the photovoltaic panel power generation process, by collecting the photovoltaic panel's own state data and the surrounding environment data, it determines whether sprinklering is necessary and optimizes the sprinkler water volume. Combining the photovoltaic panel's own state and the surrounding environment to determine the optimal sprinkler water volume improves the accuracy and targeting of sprinkler water volume optimization, thereby increasing the efficiency of sprinkler cooling and photovoltaic power generation.

[0085] According to an embodiment of the present invention, in step S1, during the testing process, at multiple test moments, test data of the photovoltaic panel's own state and test data of the surrounding environment are acquired. The surrounding environment test data includes irradiance test data received by the photovoltaic panel, ambient temperature test data, ambient humidity test data, wind speed test data, and the test angle between the wind direction and the photovoltaic panel's orientation. The self-state test data includes the photovoltaic panel's own temperature test data and the photovoltaic panel's power generation test data. A radiation sensor (e.g., a total radiation meter, a direct radiation meter, etc.) is installed on the photovoltaic panel, with the sensor's sensing plane aligned with the photovoltaic panel's surface, to accurately detect the solar radiation intensity received by the photovoltaic panel, which is the irradiance test data. By using sensors located near the photovoltaic panel to detect meteorological data such as temperature, humidity, wind speed, and wind direction detected by the nearest meteorological monitoring station, and calculating the angle between the wind direction and the photovoltaic panel's orientation, the ambient temperature test data, ambient humidity test data, wind speed test data, and test angle can be obtained. The irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, and test angle can be used as environmental test data for the photovoltaic panel, describing the environment surrounding the panel. The temperature of the photovoltaic panel surface can be obtained using a temperature sensor (e.g., a patch temperature sensor, infrared temperature sensor, etc.) installed on the panel, which is the self-temperature test data. The power generation of the photovoltaic panel can be monitored in real time using instruments such as ammeters and voltmeters, which is the power generation test data. The self-temperature test data and power generation test data of the photovoltaic panel can be used as self-state test data for the panel, describing its own state. During the test, the self-state test data and surrounding environmental test data of the photovoltaic panel can be acquired at multiple test moments to train the power generation data prediction model and the sprinkler control model. This invention can acquire test data over a relatively long period of time (the testing process), such as test data from the past year or two, with short time intervals between adjacent test moments, such as one hour, half an hour, or ten minutes. This allows for the acquisition of test data on the photovoltaic panels' own condition and the surrounding environment under various seasons and conditions, providing comprehensive and sufficient data for training power generation data prediction models. This invention does not limit the duration of the testing process or the time interval between adjacent test moments.

[0086] In this way, test data on the photovoltaic panels' own condition and the surrounding environment can be obtained at multiple test points during the test process, under various seasons and conditions, providing basic data for training power generation data prediction models.

[0087] Figure 2A flowchart illustrating the process of obtaining a trained power generation data prediction model according to an embodiment of the present invention is shown.

[0088] According to an embodiment of the present invention, in step S2, the power generation data prediction model is trained based on the test data of the photovoltaic panel's own state and the test data of the surrounding environment to obtain the trained power generation data prediction model, including: step S21, obtaining direct influencing factor test information based on irradiance test data and its own temperature test data; step S22, obtaining indirect influencing factor test information based on ambient temperature test data, wind speed test data, and test angle; step S23, obtaining indirect influencing factor test information from the (i-k+1)th test time to the ith test time and direct influencing factor test information at the ith test time, where i and k are positive integers; step S24, processing the indirect influencing factor test information from the (i-k+1)th test time to the ith test time through the self-attention mechanism of the power generation data prediction model to obtain comprehensive environmental impact information from the (i-k+1)th test time to the ith test time; step S25, processing the indirect influencing factor test information from the ith test time to the ith test time using the self-attention mechanism of the power generation data prediction model to obtain comprehensive environmental impact information from the (i-k+1)th test time to the ith test time; Step S26: The comprehensive environmental impact information is concatenated with the direct influencing factor test information at time i to obtain the power generation efficiency impact information at time i; Step S27: The comprehensive environmental impact information at time i to time i is input into the first multilayer perceptual network layer of the power generation data prediction model to obtain the predicted power generation from time i to time i+j, where j is a positive integer; Step S28: The comprehensive environmental impact information from time i-k+1 to time i is input into the 1D convolutional sub-model of the power generation data prediction model to obtain the predicted comprehensive environmental impact information from time i+1 to time i+j; Step S29: The loss function of the power generation data prediction model is determined based on the predicted power generation from time i to time i+j, the power generation test data, and the predicted comprehensive environmental impact information; Step S20: The power generation data prediction model is trained based on the loss function of the power generation data prediction model to obtain the trained power generation data prediction model.

[0089] According to an embodiment of the present invention, in step S21, direct influencing factor test information is obtained based on irradiance test data and self-temperature test data. The irradiance test data and self-temperature test data are concatenated, and the vector obtained by the concatenation is subjected to dimensionality increase processing through a fully connected layer to obtain a higher-dimensional (e.g., 128-dimensional) vector, which is the direct influencing factor test information, which can describe the direct factors affecting the power generation efficiency of the tested photovoltaic panel (e.g., the photovoltaic panel's own temperature).

[0090] According to an embodiment of the present invention, in step S22, indirect influencing factor test information is obtained based on ambient temperature test data, wind speed test data, and test angle. Similar to obtaining direct influencing factor test information, the ambient temperature test data, wind speed test data, and test angle are concatenated, and the vector obtained by the concatenation is upgraded through a fully connected layer to obtain the indirect influencing factor test information, which can describe the indirect factors affecting the power generation efficiency of the tested photovoltaic panel (e.g., the influence of ambient temperature and humidity on the heat dissipation of the photovoltaic panel).

[0091] According to an embodiment of the present invention, in step S23, indirect influencing factor test information from the (i-k+1)th test time to the ith test time and direct influencing factor test information at the ith test time are obtained, where i and k are positive integers. During the multiple test times in the testing process, the time period from the (i-k+1)th test time to the ith test time is selected for training the power generation data prediction model, and indirect influencing factor test information from the (i-k+1)th test time to the ith test time and direct influencing factor test information at the ith test time are obtained. The value of k can be determined based on the time interval between two adjacent test times. The longer the time interval between two adjacent test times, the smaller the value of k. For example, when the time interval between two adjacent test times is 1 hour, k can be 72, ensuring that the indirect influencing factor test information from the (i-k+1)th test time to the ith test time is sufficient to reflect recent changes in the environmental conditions around the photovoltaic panel, thereby improving the accuracy of determining the power generation efficiency impact information at the ith time.

[0092] According to an embodiment of the present invention, in step S24, the indirect influencing factor test information from the (i-k+1)th test time to the ith test time is processed using the self-attention mechanism of the power generation data prediction model to obtain the comprehensive environmental impact information from the (i-k+1)th test time to the ith test time. Since the indirect influencing factor test information from historical times prior to the ith test time only affects the indirect influencing factor test information at the ith test time and does not affect the direct influencing factor test information at the ith test time, and the direct influencing factor test information from historical times does not affect the direct influencing factor test information at the ith test time, that is, neither the indirect nor direct influencing factor test information from historical times prior to the ith test time directly affects the power generation efficiency of the tested photovoltaic panel at the ith test time. However, the surrounding environment from historical times prior to the ith test time may affect the surrounding environment at the ith test time; in other words, the indirect influencing factor test information at the ith test time may be influenced by the indirect influencing factor test information from previous historical times. Therefore, by multiplying the Q-matrix and K-matrix in the self-attention mechanism of the power generation data prediction model with the indirect influencing factor test information from the (i-k+1)th test time to the ith test time, the indirect influencing factor test information can be projected into a more suitable attention space for computation. This yields the query vector and key-value vector corresponding to each indirect influencing factor test information from the (i-k+1)th test time to the ith test time. Since the Q-matrix and K-matrix have the same shape, the query vector and key-value vector also have the same dimension, thus allowing for a dot product operation between them. Furthermore, the query vector corresponding to the indirect influencing factor test information at the i-th test time is multiplied (i.e., a dot product operation) by the key-value vector corresponding to the indirect influencing factor test information at each test time (including the i-th test time), and then divided by the square root of the dimension of the key-value vector. After processing with an activation function (softmax activation function), the relevant weights corresponding to the indirect influencing factor test information at each test time can be obtained. These weights describe the correlation between the indirect influencing factor test information at each test time and the indirect influencing factor test information at the i-th test time, and also describe the correlation between the surrounding environmental conditions of the photovoltaic panel at each test time and the surrounding environmental conditions of the photovoltaic panel at the i-th test time. The larger the aforementioned relevant weights, the higher the correlation between the surrounding environmental conditions of the photovoltaic panel at the corresponding test time and the surrounding environmental conditions of the photovoltaic panel at the i-th test time, and the higher the importance. Of course, the Q matrix and K matrix are both learnable relevant weight matrices, and the parameters of the Q matrix and K matrix can be determined through training.

[0093] According to an embodiment of the present invention, the weight vectors corresponding to the test information of each indirect influencing factor are obtained by multiplying the test information of each indirect influencing factor by the V matrix of the self-attention mechanism. The V matrix is ​​a learnable weight matrix, and its parameters can be determined through training. Further, by using the relevant weights corresponding to the test information of each indirect influencing factor at each test time, a new vector is obtained, which is the comprehensive environmental impact information. For example, using the relevant weights between the i-th test time and other test times, the new vector corresponding to the i-th test time is weighted and summed with the new vectors corresponding to other test times to obtain the comprehensive environmental impact information of the i-th test time. This can integrate the features of the surrounding environment of the photovoltaic panel at each test time, describe the features of the surrounding environment of the photovoltaic panel at the i-th test time, and comprehensively describe the influence of multiple times on the surrounding environment at the i-th test time, improving the accuracy and comprehensiveness of the feature representation of the surrounding environment of the photovoltaic panel at the i-th test time. Based on the same processing method, the comprehensive environmental impact information of each test time from the (i-k+1)-th test time to the i-th test time can be obtained.

[0094] According to an embodiment of the present invention, in step S25, the comprehensive environmental impact information at the i-th test time is concatenated with the direct influencing factor test information at the i-th test time to obtain the power generation efficiency impact information at the i-th test time. The comprehensive environmental impact information at the i-th test time is concatenated with the direct influencing factor test information at the i-th test time, and then subjected to dimensionality reduction (e.g., from 256 dimensions to 128 dimensions) through a fully connected layer to obtain a dimensionality-reduced vector, which is the power generation efficiency impact information at the i-th test time. This vector can describe the power generation efficiency impact at the i-th test time from both direct factors and environmental factors, making the characteristic expression of the power generation efficiency impact more comprehensive and accurate.

[0095] According to an embodiment of the present invention, in step S26, the power generation efficiency impact information at the i-th test time is input into the first multilayer perceptron layer of the power generation data prediction model to obtain the predicted power generation from the i-th test time to the (i+j)-th time, where j is a positive integer. The power generation efficiency impact information at the i-th test time is input into the first multilayer perceptron layer of the power generation data prediction model (e.g., including fully connected layers and activation layers, where the activation layers are processed using the ReLU activation function) for processing. The output neurons of the last layer are set to j, thus mapping the power generation efficiency impact information to the predicted power generation from the i-th test time to the (i+j)-th time, i.e., predicting the power generation at the next j-th time. j is a positive integer parameter configured based on the time interval between two adjacent test times and the duration of the spraying effect, for example, 4, 5, 6, etc.

[0096] According to an embodiment of the present invention, in step S27, the comprehensive environmental impact information from the (i-k+1)th test time to the ith test time is input into the 1D convolutional sub-model of the power generation data prediction model to obtain the predicted comprehensive environmental impact information from the (i+1)th test time to the (i+j)th test time. The comprehensive environmental impact information from the (i-k+1)th test time to the ith test time is arranged and combined in chronological order to obtain a time series of comprehensive environmental impact information. Further, the comprehensive environmental impact information is convolved using the 1D convolutional sub-model of the power generation data prediction model to obtain the output vector corresponding to the (i+1)th test time to the (i+j)th test time, which is the predicted comprehensive environmental impact information. Through the above convolutional processing, when predicting the comprehensive environmental impact information from the (i+1)th to the (i+j)th test time, features of the comprehensive environmental impact information from previous test times can be extracted. This allows the predicted comprehensive environmental impact information to describe the trend of changes in the surrounding environment of the photovoltaic panels over the past test times, thereby more accurately predicting future environmental change trends. This information can then be compared and fed back with the comprehensive environmental impact information calculated from the (i+1)th to the (i+j)th test times, thus assisting in the training of the power generation data prediction model.

[0097] According to an embodiment of the present invention, in step S28, the loss function of the power generation data prediction model is determined based on the predicted power generation, power generation test data, and predicted comprehensive environmental impact information from the i-th test time to the i+j-th test time. This includes: determining the power generation data loss function based on the predicted power generation and power generation test data from the i-th test time to the i+j-th test time; inputting the predicted comprehensive environmental impact information from the i-th test time to the i+j-th test time into the second multilayer sensing network layer to obtain the predicted ambient temperature, predicted wind speed, and predicted angle from the i-th test time to the i+j-th test time; determining the influencing factor prediction loss function based on the predicted ambient temperature, predicted wind speed, predicted angle, ambient temperature test data, wind speed test data, and test angle from the i-th test time to the i+j-th test time; and obtaining the loss function of the power generation data prediction model based on the power generation data loss function and the influencing factor prediction loss function.

[0098] According to an embodiment of the present invention, a power generation data loss function is determined based on the predicted power generation from time i to time i+j and the power generation test data. The average absolute error between the predicted power generation from time i to time i+j and the power generation test data (i.e., the actual power generation) at the same time can be used as the power generation data loss function.

[0099] According to an embodiment of the present invention, the predicted comprehensive environmental impact information from the i-th test time to the (i+j)-th test time is input into the second multilayer sensing network layer to obtain the predicted environmental temperature, predicted wind speed, and predicted angle from the i-th test time to the (i+j)-th test time. The predicted comprehensive environmental impact information from the i-th test time to the (i+j)-th test time is then input into the second multilayer sensing network layer of the power generation data prediction model (e.g., including a fully connected layer and an activation layer, where the activation layer is a network layer processed using the ReLU activation function) for processing. This maps the predicted comprehensive environmental impact information from the i-th test time to the (i+j)-th test time to the predicted environmental temperature, predicted wind speed, and predicted angle corresponding to each test time between the i-th and (i+j)-th test times, thereby predicting the environmental temperature, wind speed, and the angle between the wind direction and the orientation of the test photovoltaic panel at future j-times.

[0100] According to an embodiment of the present invention, the prediction loss function for influencing factors is determined based on the predicted ambient temperature, predicted wind speed, predicted angle, ambient temperature test data, wind speed test data, and test angle from the i-th test time to the (i+j)-th test time. This includes: determining the prediction loss function for influencing factors according to formula (1). ,

[0101] (1),

[0102] in, Let be the predicted ambient temperature at the (i+s)th test time. Let be the predicted wind speed at the (i+s)th test time. Let be the predicted angle at the (i+s)th test time. For the reason , and The vector formed This refers to the ambient temperature test data at the (i+s)th test time. The wind speed test data is for the (i+s)th test time. Let be the angle between the test points at the (i+s)th test time. For the reason , and The vectors formed by s and j are sim, where s ≤ j and s is a positive integer.

[0103] According to an embodiment of the present invention, in formula (1), For the reason , and The vector formed can describe the predicted ambient temperature, predicted wind speed, and predicted angle at the (i+s)th test time. For the reason , and The vector formed can describe the ambient temperature test data, wind speed test data, and test angle at the i+s test time. In formula (1), This represents the similarity (e.g., cosine similarity) between the predicted ambient temperature, predicted wind speed, and predicted angle at the (i+s)-th test time and the ambient temperature test data, wind speed test data, and test angle. This can be represented by values ​​describing the deviation of the predicted ambient temperature, predicted wind speed, and predicted angle at the (i+s)th test time from the actual ambient temperature, wind speed, and angle test data. It can be considered as a value describing the deviation of the predicted influencing factors at the (i+s)th test time from the actual influencing factors. Therefore, It can represent the sum of the values ​​of the predicted influencing factors and the actual influencing factors corresponding to each test time from the i-th test time to the (i+j)-th test time. It can describe the total deviation of the predicted influencing factors from the actual influencing factors and can be used as the prediction loss function of the influencing factors.

[0104] According to an embodiment of the present invention, the loss function of the power generation data prediction model is obtained based on the power generation data loss function and the influencing factor prediction loss function. The loss function of the power generation data prediction model is obtained by performing a weighted summation on the power generation data loss function and the influencing factor prediction loss function.

[0105] According to an embodiment of the present invention, a power generation data prediction model is trained based on the loss function of the power generation data prediction model to obtain a trained power generation data prediction model. Backpropagation can be performed based on the loss function of the power generation data prediction model, and the parameters of the power generation data prediction model can be adjusted using the gradient descent method to train the model. After multiple training iterations (i.e., training using predicted ambient temperature, predicted wind speed, predicted angle, ambient temperature test data, wind speed test data, and test angle at multiple test times), training is complete, and a trained power generation data prediction model is obtained.

[0106] In this way, based on the test data of the photovoltaic panels' own state and the surrounding environment, the power generation data loss function and the influencing factor prediction loss function can be determined. Then, after weighted summation, the loss function of the power generation data prediction model is obtained, and the model is trained to produce the trained power generation data prediction model. Considering that indirect influencing factor test information from historical moments only affects indirect influencing factor test information from future moments, an influencing factor prediction loss function is determined based on a self-attention mechanism and 1D convolution processing. This function assists in training the power generation data prediction model, improving the accuracy and comprehensiveness of the training and enhancing the performance of the power generation data prediction model.

[0107] According to an embodiment of the present invention, in step S3, before and after multiple spray tests, test data on the photovoltaic panel's own state, test data on the surrounding environment, and spray water volume are acquired respectively. During the test, the moment when the spray test is to be conducted is defined as before the spray test, and the moment after the spray test, after the required cooling time, is defined as after the spray test. For example, if the required cooling time after the spray test is 10 minutes, and the moment when the spray test is to be conducted is 10:10, then 10:10 is before the spray test, and the moment 10 minutes after the spray test, i.e., 10:20, is after the spray test. Based on the same processing method, test data on the photovoltaic panel's own state and test data on the surrounding environment can be acquired before and after multiple spray tests, and the spray water flow rate before and after each spray test can be determined by a flow sensor (e.g., turbine flow meter, electromagnetic flow meter, etc.) installed on the water supply pipeline of the spray system, which is the spray water volume data, used to train the spray control model. The present invention does not limit the determination of before and after the spray test.

[0108] According to an embodiment of the present invention, in step S4, the spray control model is trained based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data to obtain the trained spray control model. This includes: determining irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, and test angle based on the surrounding environment test data before the spray test, and determining self-temperature test data based on the self-state test data before the spray test; determining cooling effect information based on the irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, test angle, and self-temperature test data before the spray test; determining cooling result information based on the self-temperature test data before and after the spray test; obtaining spray prediction information based on the cooling effect information and cooling result information; determining the loss function of the spray control model based on the spray prediction information and the spray water volume data; and training the spray control model based on the loss function of the spray control model to obtain the trained spray control model.

[0109] According to an embodiment of the present invention, irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, and test angle are determined based on the surrounding environmental test data before the spray test, and self-temperature test data are determined based on the self-state test data before the spray test. Similar to training the power generation data prediction model, irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, and test angle are obtained from the surrounding environmental test data before the spray test, and self-temperature test data is obtained from the self-state test data before the spray test.

[0110] According to an embodiment of the present invention, cooling effect information is determined based on irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, test angle, and self-temperature test data before the spray test. The irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, test angle, and self-temperature test data before the spray test are concatenated, and the resulting vector is then subjected to dimensionality upscaling through a fully connected layer to obtain a higher-dimensional (e.g., 128-dimensional) vector, which is the cooling effect information. During the spray test, the higher the self-temperature test data, the faster the sprayed water evaporates upon contact with the photovoltaic panel, resulting in higher cooling efficiency of the photovoltaic panel. Therefore, the cooling effect information can describe the factors affecting the cooling effect of the photovoltaic panel.

[0111] According to an embodiment of the present invention, cooling result information is determined based on the self-temperature test data before and after the spray test. Similar to obtaining cooling effect information, the self-temperature test data before and after the spray test are spliced ​​together, and the resulting vector is then subjected to dimensionality-upgrading processing through a fully connected layer to obtain the cooling result information, which describes the cooling effect of the photovoltaic panel after spraying.

[0112] According to an embodiment of the present invention, spray prediction information is obtained based on cooling effect information and cooling result information. The cooling effect information and cooling result information are concatenated and then processed by a fully connected layer for dimensionality reduction (e.g., from 256 dimensions to 128 dimensions) to obtain a dimensionality-reduced vector. The dimensionality-reduced vector is then input into the spray control model to output the spray prediction information, that is, the predicted amount of spray water required to achieve the cooling effect.

[0113] According to an embodiment of the present invention, the loss function of the sprinkler control model is determined based on sprinkler prediction information and sprinkler water volume data, including: determining the loss function of the sprinkler control model according to formula (2). ,

[0114] (2),

[0115] in, For sprinkler forecast information, For spray water volume data, denoted as the maximum water volume for a single spray cycle, and max is the function that takes the maximum value.

[0116] According to an embodiment of the present invention, in formula (2), This represents the absolute value of the difference between the sprinkler prediction information and the sprinkler water volume data, and can be considered as the gap between the sprinkler prediction information and the sprinkler water volume data. In formula (2), for and The maximum value in, if This indicates that the error between the sprinkler prediction information and the sprinkler water volume data is small, but the difference between the sprinkler water volume data and the maximum single sprinkler water volume is large. In this case... , This represents the relative difference between the spray water volume and the maximum single spray water volume. Since the photovoltaic panel's temperature decreases and the surrounding humidity increases as the spray water volume increases, although further cooling is possible, the cooling rate of the photovoltaic panel decreases. Therefore, the smaller the relative difference between the spray water volume and the maximum single spray water volume (i.e., the closer the spray water volume is to the maximum single spray water volume), the lower the relative difference between the two values. The smaller the volume of water sprayed, the less significant the improvement in cooling effect. For example, if the maximum single spray volume is 20 liters, the actual cooling effect of spraying 8 liters of water is significantly different from that of spraying 10 liters of water, while the cooling effect of spraying 18 liters of water is not significantly different from that of spraying 20 liters of water. In this case, [the following text is incomplete and requires further context: "smaller", "spraying", "20 liters ... As a weight, it can describe the importance of the difference between the sprinkler prediction information and the sprinkler water volume data in this case. The larger the sprinkler water volume data, the smaller the value of the above weight, and the closer the sprinkler water volume is to the maximum single sprinkler water volume. The smaller the error between the spray prediction information and the spray water volume data, the less impact it has on the cooling effect. Therefore, a lower weight is assigned. Conversely, the smaller the spray water volume data, the more significant the impact of even a small deviation on the cooling effect when using that data. Larger, meaning higher weights are set. On the other hand, if If this is the case, it means that the error between the current sprinkler prediction information and the sprinkler water volume data is too large. In this case, [the following will be implemented / redirected]. As a weight, a larger weight is set to increase the training intensity.

[0117] According to an embodiment of the present invention, the sprinkler control model is trained based on the loss function of the sprinkler control model to obtain a trained sprinkler control model. Backpropagation can be performed based on the loss function of the sprinkler control model, and the parameters of the sprinkler control model can be adjusted using the gradient descent method to train the sprinkler control model. After multiple training iterations (i.e., training using sprinkler prediction information and sprinkler water volume data before and after multiple sprinkler tests), training is complete, and a trained sprinkler control model is obtained. This allows the trained sprinkler control model to output lower sprinkler prediction information, achieving a cooling effect while reducing the amount of sprinkler water used, effectively improving the efficiency of sprinkler cooling.

[0118] In this way, the loss function of the sprinkler control model can be determined based on information about the cooling effect and the cooling result, and the sprinkler control model can be trained to obtain the trained sprinkler control model. Considering that the closer the sprinkler water volume is to the maximum single sprinkler water volume, the smaller the change in cooling efficiency, weights are assigned to the differences between different sprinkler water volume data and sprinkler prediction information, thereby determining the loss function of the sprinkler control model. This improves the accuracy and relevance of the training and enhances the performance of the sprinkler control model.

[0119] According to an embodiment of the present invention, in step S5, during the process of generating electricity using a photovoltaic panel, data on the photovoltaic panel's own status and data on the surrounding environment are collected. The surrounding environment data includes irradiance data received by the test photovoltaic panel, ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the photovoltaic panel's orientation. The photovoltaic panel's own status data includes its own temperature data and the photovoltaic panel's power generation. The photovoltaic panel is of the same model as the test photovoltaic panel.

[0120] According to an embodiment of the present invention, in step S6, determining whether spraying is required based on self-state data, surrounding environment data, and a trained power generation data prediction model includes: inputting self-state data and surrounding environment data into the trained power generation data prediction model to obtain the expected power generation; if the self-temperature data is higher than or equal to the self-temperature threshold, or the relative difference between the expected power generation and the power generation is greater than a preset error threshold, then it is determined that spraying is required.

[0121] According to an embodiment of the present invention, the user's own state data and surrounding environment data are input into a trained power generation data prediction model to obtain the expected power generation. Similarly, by inputting the user's own state data and surrounding environment data into the trained power generation data prediction model, the target power generation that the photovoltaic panel should achieve can be obtained, which is the expected power generation.

[0122] According to an embodiment of the present invention, if the self-temperature data is higher than or equal to the self-temperature threshold, or the relative difference between the expected power generation and the actual power generation is greater than a preset error threshold, it is determined that spraying is required. The ratio of the difference between the expected power generation and the actual power generation to the expected power generation is the relative difference between the expected power generation and the actual power generation. The self-temperature threshold and the preset error threshold can be determined based on the temperature change characteristics of the photovoltaic panel, the parameters of the photovoltaic panel, etc. If the self-temperature data is higher than or equal to the self-temperature threshold (e.g., 45°C), or the relative difference between the expected power generation and the actual power generation is greater than the preset error threshold (e.g., 0.1), it can be considered that the temperature of the photovoltaic panel surface is too high, which may reduce the photovoltaic power generation efficiency or even damage the photovoltaic panel, or the difference between the target power generation that the photovoltaic panel should achieve and the actual power generation is too large, and the photovoltaic panel may be affected by other factors (e.g., too much dust on the photovoltaic panel, or other stains blocking the light). Therefore, it is necessary to spray the photovoltaic panel to cool it down, so as to protect the photovoltaic panel or increase the power generation, thereby improving the efficiency of photovoltaic power generation.

[0123] According to an embodiment of the present invention, in step S7, if spraying is required, the optimal spraying water volume is determined based on the self-state data, surrounding environment data, and the trained spraying control model, including: determining the target's own temperature based on the trained power generation data prediction model and surrounding environment data; inputting the surrounding environment data, the self-temperature data, and the target's own temperature into the trained spraying control model to determine a first candidate spraying water volume; determining a second candidate spraying water volume based on the relative difference between the expected power generation and the power generation and the maximum single spraying water volume; and determining the maximum value of the first candidate spraying water volume and the second candidate spraying water volume as the optimal spraying water volume.

[0124] According to an embodiment of the present invention, the target's own temperature is determined based on the trained power generation data prediction model and surrounding environmental data, including: inputting the simulated value of the target's own temperature data and the surrounding environmental data into the trained power generation data prediction model to obtain predicted power generation values ​​for multiple future time periods, wherein the simulated value of the target's own temperature data includes a value equal to the target's own temperature data, and the simulated value of the target's own temperature data is higher than the spray water temperature; fitting the predicted power generation values ​​with the multiple future time periods to obtain a regular function of power generation versus time; integrating the regular function to obtain the power generation corresponding to the simulated value of the target's own temperature data; and obtaining the maximum cooling efficiency E according to formula (3).

[0125] (3),

[0126] in, The power generation corresponding to the simulated value of the xth self-temperature data. The amount of electricity generated is equal to the value corresponding to its own temperature data. Let x be the simulated value of its own temperature data. The value is equal to the self-temperature data, N is the number of simulated values ​​of the self-temperature data, max is the maximum value function, x≤N, and x and N are both positive integers; the minimum value between the simulated value of the self-temperature data corresponding to the maximum cooling efficiency and the self-temperature threshold is determined as the target self-temperature.

[0127] According to an embodiment of the present invention, simulated values ​​of self-temperature data and surrounding environmental data are input into a trained power generation data prediction model to obtain predicted power generation values ​​for multiple future time periods. The simulated values ​​of self-temperature data include values ​​equal to the self-temperature data, and these simulated values ​​are higher than the spray water temperature. Since the self-temperature data decreases after spraying but does not fall below the spray water temperature, the simulated values ​​of self-temperature data are less than or equal to the self-temperature data and always greater than the spray water temperature. For example, if the self-temperature data is 50 degrees Celsius and the spray water temperature is 15 degrees Celsius, the range of the simulated values ​​of self-temperature data is (15, 50). Multiple simulated values ​​of self-temperature data can be selected within this range, and each must include a value equal to the self-temperature data, i.e., 50 degrees Celsius. Inputting these simulated values ​​of self-temperature data into the trained power generation data prediction model outputs predicted power generation values ​​corresponding to the simulated values ​​of various self-temperature data for multiple future time periods. The number of output predicted power generation values ​​is the same as the number of output neurons in the last layer of the trained power generation data prediction model.

[0128] According to an embodiment of the present invention, the predicted power generation value is fitted with multiple future time points to obtain a regular function of power generation versus time. An equation to be fitted is set for the predicted power generation value versus time, and the fitting coefficients are solved using a fitting method based on the predicted power generation value corresponding to the simulated value of each temperature data point and multiple future time points to obtain the solution value. Further, the solution values ​​of multiple sets of fitting coefficients are substituted into the above-mentioned equation to be fitted, and the fitting coefficient corresponding to the minimized sum of squares of the residuals obtained after multiple substitutions is found as the solution value of the fitting coefficients. Substituting this value into the equation to be fitted yields the regular function of power generation versus time. Since the temperature of the photovoltaic panel gradually decreases as the spraying proceeds, the power generation gradually increases, but the rate of increase (i.e., the cooling efficiency) gradually decreases. Therefore, the regular function of power generation versus time is a function that gradually increases, and the slope of the function gradually decreases.

[0129] According to an embodiment of the present invention, the power generation corresponding to the simulated value of its own temperature data is obtained by integrating the regularity function. By integrating the regularity function (i.e., integrating the power generation over time), the predicted power generation corresponding to each simulated value of its own temperature data can be obtained, which is the power generation corresponding to the simulated value of its own temperature data.

[0130] According to an embodiment of the present invention, in formula (3), since the temperature of the sprayed product decreases and the power generation increases after spraying, therefore, and In formula (3), The difference between the power generation corresponding to the simulated value of the x-th self-temperature data and the power generation corresponding to the value equal to the self-temperature data can be considered as the increase in power generation when the self-temperature data decreases to the simulated value of the x-th self-temperature data. This represents the difference between a value equal to its own temperature data and the simulated value of the x-th own temperature data point, which can be considered as the decrease in its own temperature data. Therefore, This can be represented as the ratio of the increase in power generation to the decrease in its own temperature data. It can be considered as the increase in power generation caused by a unit decrease in its own temperature data when the temperature data decreases to the simulated value of the x-th self-temperature data point. Alternatively, it can be considered as the cooling efficiency corresponding to the decrease in its own temperature data to the simulated value of the x-th self-temperature data point. Therefore, It can represent the maximum cooling efficiency corresponding to the temperature data of itself being reduced to the simulated value of each of its own temperature data, that is, the maximum cooling efficiency E.

[0131] According to an embodiment of the present invention, the simulated value of the self-temperature data corresponding to the maximum cooling efficiency is determined as the target self-temperature. The simulated value of the self-temperature data corresponding to the maximum cooling efficiency is determined as a candidate value for the target self-temperature. The cooling efficiency is highest when the self-temperature data is reduced to this candidate value of the target self-temperature, which can improve cooling efficiency and reduce water waste while achieving the cooling effect. Furthermore, to ensure that the self-temperature of the photovoltaic panel does not exceed the self-temperature threshold after spray cooling, the smaller of the candidate value of the target self-temperature and the self-temperature threshold can be determined as the target self-temperature.

[0132] According to an embodiment of the present invention, ambient environmental data, the user's own temperature data, and the target's own temperature are input into a trained sprinkler control model to determine a first candidate sprinkler water volume. Similar to obtaining sprinkler prediction information, by inputting ambient environmental data, the user's own temperature data, and the target's own temperature into the trained sprinkler control model, sprinkler prediction information corresponding to the target's own temperature can be obtained, which is the first candidate sprinkler water volume.

[0133] According to an embodiment of the present invention, a second candidate spray water volume is determined based on the relative difference between the desired power generation and the actual power generation, and the maximum single spray water volume. The ratio of the difference between the desired power generation and the actual power generation to the desired power generation is the relative difference, which can be considered as the relative difference between the spray water volume required to increase the power generation to the desired power generation and the maximum single spray water volume. Furthermore, multiplying the relative difference by the maximum single spray water volume yields the spray water volume required to increase the power generation to the desired power generation, which is the second candidate spray water volume. For example, the relative difference can be considered as being caused by dirt blocking light, and the blocking ratio is the relative difference. In this example, the maximum single spray water volume can be used to clean the entire photovoltaic panel, and the second candidate spray water volume obtained by multiplying the relative difference by the maximum single spray water volume can be used to clean the dirt with the blocking ratio.

[0134] According to an embodiment of the present invention, the maximum value of the first candidate spray water volume and the second candidate spray water volume is determined as the optimized spray water volume. Since spraying must enable the power generation to reach the desired power output, the maximum value of the first candidate spray water volume and the second candidate spray water volume can be determined as the optimized spray water volume. Spraying with the above-mentioned optimized spray water volume can improve the cooling efficiency and reduce the waste of water resources while increasing the power generation to the desired power output.

[0135] In this way, based on its own state data, surrounding environmental data, and a trained power generation data prediction model, the system determines whether spraying is necessary. It then determines a first and second candidate spraying water volume based on the same data, leading to an optimized spraying water volume. This approach considers the requirement that spraying must ensure the desired high-power generation, and sets the spraying water volume based on the efficiency increase in power generation caused by cooling. This improves the accuracy and relevance of spraying water volume optimization, thereby enhancing the efficiency of both cooling and photovoltaic power generation.

[0136] The IoT-based photovoltaic intelligent sprinkler method and system according to embodiments of the present invention can acquire test data on the photovoltaic panel's own state and surrounding environment at multiple test moments, thereby obtaining a trained power generation data prediction model. Furthermore, before and after multiple sprinkler tests, it acquires test data on the photovoltaic panel's own state, surrounding environment, and sprinkler water volume, respectively, thereby obtaining a trained sprinkler control model. During the photovoltaic panel power generation process, the system collects the photovoltaic panel's own state data and surrounding environment data to determine whether sprinklering is necessary and to optimize the sprinkler water volume. Combining the photovoltaic panel's own state and the surrounding environment to determine the optimized sprinkler water volume improves the accuracy and targeting of sprinkler water volume optimization, thereby increasing the efficiency of sprinkler cooling and photovoltaic power generation. Moreover, during the testing process, at multiple test moments, it can acquire test data on the photovoltaic panel's own state and surrounding environment under various seasons and conditions, providing basic data for training the power generation data prediction model. When training the power generation data prediction model, the power generation data loss function and the influencing factor prediction loss function can be determined based on the test data of the photovoltaic panel's own state and the surrounding environment. These are then weighted and summed to obtain the loss function of the power generation data prediction model, which is then used to train the model. Considering that indirect influencing factor test information from historical moments only affects indirect influencing factor test information from future moments, an influencing factor prediction loss function is determined based on a self-attention mechanism and 1D convolution processing. This function assists in training the power generation data prediction model, improving the accuracy and comprehensiveness of the training and enhancing the model's performance. Similarly, when training the sprinkler control model, the loss function can be determined based on cooling impact information and cooling result information. The sprinkler control model is then trained to obtain the trained model. Considering that the closer the spray water volume is to the maximum single spray water volume, the smaller the change in cooling efficiency, weights are assigned to the differences between different spray water volume data and spray prediction information. This determines the loss function of the spray control model, improving the accuracy and relevance of training and enhancing the performance of the spray control model. Furthermore, based on its own state data, surrounding environmental data, and the trained power generation data prediction model, it can be determined whether spraying is necessary. Based on its own state data, surrounding environmental data, and the trained spray control model, a first and second candidate spray water volume are determined, resulting in an optimized spray water volume. Considering that spraying must ensure that power generation reaches the desired high power output, the spray water volume is set based on the efficiency of power generation improvement caused by cooling. This improves the accuracy and relevance of spray water volume optimization, enhancing the efficiency of spray cooling and photovoltaic power generation.

[0137] Figure 3An exemplary block diagram of an IoT-based smart photovoltaic sprinkler system according to an embodiment of the present invention is shown, comprising:

[0138] The first acquisition module acquires test data of the photovoltaic panel's own status and test data of the surrounding environment at multiple test moments during the test process. The test data of the surrounding environment includes test data of the irradiance received by the photovoltaic panel, test data of the ambient temperature of the environment where the photovoltaic panel is located, test data of the ambient humidity, test data of the wind speed, and test angle between the wind direction and the orientation of the photovoltaic panel. The test data of the photovoltaic panel's own status includes test data of the photovoltaic panel's own temperature and test data of the photovoltaic panel's power generation.

[0139] The first training module trains the power generation data prediction model based on the test data of the photovoltaic panel's own state and the test data of the surrounding environment, and obtains the trained power generation data prediction model.

[0140] The second acquisition module acquires test data on the photovoltaic panel's own status, test data on the surrounding environment, and data on the amount of water sprayed before and after multiple spray tests.

[0141] The second training module trains the spray control model based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, and obtains the trained spray control model.

[0142] The data acquisition module collects data on the photovoltaic panel's own status and the surrounding environment during the photovoltaic panel power generation process. The surrounding environment data includes the irradiance data received by the test photovoltaic panel, the ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the photovoltaic panel's orientation. The photovoltaic panel's own status data includes the photovoltaic panel's own temperature data and the photovoltaic panel's power generation capacity. The photovoltaic panel is the same model as the test photovoltaic panel.

[0143] The judgment module determines whether spraying is necessary based on its own status data, surrounding environmental data, and the trained power generation data prediction model.

[0144] The sprinkler water volume optimization module determines the optimal sprinkler water volume based on its own status data, surrounding environment data, and the trained sprinkler control model if sprinkler operation is required.

[0145] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0146] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A photovoltaic intelligent sprinkler system based on the Internet of Things, characterized in that, include: During the test, at multiple test moments, test data on the photovoltaic panel's own status and the surrounding environment were acquired. The surrounding environment test data included the irradiance received by the photovoltaic panel, the ambient temperature of the environment where the photovoltaic panel was located, the ambient humidity, the wind speed, and the angle between the wind direction and the photovoltaic panel's orientation. The self-status test data included the photovoltaic panel's own temperature and the photovoltaic panel's power generation. Based on the test data of the photovoltaic panel's own condition and the test data of the surrounding environment, the power generation data prediction model is trained to obtain the trained power generation data prediction model. Before and after multiple spray tests, test data on the photovoltaic panel's own condition, the surrounding environment, and the spray water volume were obtained respectively. Based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, the spray control model is trained to obtain the trained spray control model. During the process of generating electricity using photovoltaic panels, data on the photovoltaic panels' own status and the surrounding environment are collected. The surrounding environment data includes the irradiance data received by the test photovoltaic panels, the ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the orientation of the photovoltaic panels. The photovoltaic panels' own status data includes the photovoltaic panels' own temperature data and the photovoltaic panels' power generation. The photovoltaic panels are of the same model as the test photovoltaic panels. Based on its own status data, surrounding environment data, and the trained power generation data prediction model, it is determined whether spraying is necessary. If spraying is required, the optimal spraying water volume is determined based on its own status data, surrounding environment data, and the trained spraying control model. Based on the test data of the photovoltaic panels' own condition and the test data of the surrounding environment, the power generation data prediction model is trained to obtain the trained power generation data prediction model, including: Based on irradiance test data and its own temperature test data, information on the test of directly influencing factors was obtained; Based on ambient temperature test data, wind speed test data, and test angle, information on indirect influencing factors was obtained. Obtain the test information of indirect influencing factors from the (i-k+1)th test time to the ith test time and the test information of direct influencing factors at the ith test time, where i and k are positive integers; By using the self-attention mechanism of the power generation data prediction model, the test information of indirect influencing factors from the (i-k+1)th test time to the ith test time is processed to obtain the comprehensive environmental impact information from the (i-k+1)th test time to the ith test time. By splicing the comprehensive environmental impact information at the i-th test time with the direct influencing factor test information at the i-th test time, the power generation efficiency impact information at the i-th test time is obtained. Input the power generation efficiency impact information at the i-th test time into the first multilayer sensing network layer of the power generation data prediction model to obtain the predicted power generation from the i-th test time to the (i+j)-th time, where j is a positive integer; The comprehensive environmental impact information from the (i-k+1)th test time to the ith test time is input into the 1D convolutional sub-model of the power generation data prediction model to obtain the predicted comprehensive environmental impact information from the (i+1)th test time to the (i+j)th test time. Based on the predicted power generation, power generation test data, and predicted comprehensive environmental impact information from the i-th test time to the (i+j)-th test time, the loss function of the power generation data prediction model is determined. Based on the loss function of the power generation data prediction model, the power generation data prediction model is trained to obtain the trained power generation data prediction model.

2. The IoT-based intelligent photovoltaic sprinkler method according to claim 1, characterized in that, Based on the predicted power generation, power generation test data, and predicted comprehensive environmental impact information from the i-th to the (i+j)-th test time, the loss function of the power generation data prediction model is determined, including: Based on the predicted power generation and power generation test data from the i-th time to the (i+j)-th test time, determine the power generation data loss function; The comprehensive environmental impact information from the i-th test time to the (i+j)-th test time is input into the second multilayer sensing network layer to obtain the predicted environmental temperature, predicted wind speed and predicted angle from the i-th test time to the (i+j)-th test time. Based on the predicted ambient temperature, predicted wind speed, predicted angle, ambient temperature test data, wind speed test data, and test angle from the i-th test time to the (i+j)-th test time, determine the prediction loss function for influencing factors; Based on the power generation data loss function and the influencing factor prediction loss function, the loss function of the power generation data prediction model is obtained.

3. The IoT-based intelligent photovoltaic sprinkler method according to claim 2, characterized in that, Based on the predicted ambient temperature, predicted wind speed, predicted angle, ambient temperature test data, wind speed test data, and test angle from the i-th test time to the (i+j-th)-th test time, determine the prediction loss function for influencing factors, including: According to the formula ， Determine the influencing factors and predict the loss function ,in, Let be the predicted ambient temperature at the (i+s)th test time. Let be the predicted wind speed at the (i+s)th test time. Let be the predicted angle at the (i+s)th test time. For the reason , and The vector formed This refers to the ambient temperature test data at the (i+s)th test time. The wind speed test data is for the (i+s)th test time. Let be the angle between the test points at the (i+s)th test time. For the reason , and The vectors formed by s and j are sim, where s ≤ j and s is a positive integer.

4. The IoT-based intelligent photovoltaic sprinkler method according to claim 1, characterized in that, Based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, the spray control model is trained to obtain the trained spray control model, including: Based on the surrounding environment test data before the spray test, determine the irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data and test angle, and based on the self-state test data before the spray test, determine the self-temperature test data. Based on the irradiance test data, ambient temperature test data, ambient humidity test data, wind speed test data, test angle and self-temperature test data before the spray test, the cooling effect information is determined; Based on the temperature test data before and after the spray test, determine the cooling result information; Based on the information on the impact and results of cooling, spraying prediction information is obtained; Based on the sprinkler prediction information and sprinkler water volume data, the loss function of the sprinkler control model is determined; Based on the loss function of the sprinkler control model, the sprinkler control model is trained to obtain the trained sprinkler control model.

5. The IoT-based intelligent photovoltaic sprinkler method according to claim 4, characterized in that, Based on sprinkler prediction information and sprinkler water volume data, the loss function of the sprinkler control model is determined, including: According to the formula ， Determine the loss function of the sprinkler control model. ,in, For sprinkler forecast information, For spray water volume data, denoted as the maximum water volume for a single spray cycle, and max is the function that takes the maximum value.

6. The IoT-based intelligent photovoltaic sprinkler method according to claim 1, characterized in that, Based on its own status data, surrounding environmental data, and the trained power generation data prediction model, it is determined whether spraying is necessary, including: By inputting its own state data and surrounding environment data into the trained power generation data prediction model, the expected power generation is obtained. If the temperature data is higher than or equal to the temperature threshold, or if the relative difference between the expected power generation and the actual power generation is greater than the preset error threshold, then spraying is required.

7. The IoT-based intelligent photovoltaic sprinkler method according to claim 6, characterized in that, If spraying is required, the optimal spray water volume is determined based on its own state data, surrounding environmental data, and the trained spray control model, including: The target's own temperature is determined based on the predicted model of power generation data after training and the surrounding environmental data; Input the surrounding environmental data, the target's own temperature data, and the target's own temperature into the trained sprinkler control model to determine the first candidate sprinkler water volume. The second candidate spray volume is determined based on the relative difference between the expected power generation and the power generation and the maximum single spray volume. The maximum value of the first and second candidate spray water volumes is determined as the optimal spray water volume.

8. The IoT-based intelligent photovoltaic sprinkler method according to claim 7, characterized in that, Based on the trained power generation data prediction model and surrounding environmental data, the target's own temperature is determined, including: The simulated value of its own temperature data and the surrounding environment data are input into the trained power generation data prediction model to obtain the predicted power generation value at multiple future times. The simulated value of its own temperature data includes a value equal to its own temperature data, and the simulated value of its own temperature data is higher than the temperature of the spray water. By fitting the predicted power generation value with multiple future time points, a regular function of power generation versus time can be obtained; By integrating the regularity function, the power generation corresponding to the simulated value of its own temperature data is obtained; According to the formula ， To achieve the maximum cooling efficiency E, where, The power generation corresponding to the simulated value of the xth self-temperature data. The amount of electricity generated is equal to the value corresponding to its own temperature data. Let x be the simulated value of its own temperature data. The value is equal to its own temperature data, N is the number of simulated values ​​of its own temperature data, max is the maximum value function, x≤N, and x and N are both positive integers; The minimum value between the simulated value of the self-temperature data corresponding to the maximum cooling efficiency and the self-temperature threshold is determined as the target self-temperature.

9. An Internet of Things-based photovoltaic intelligent sprinkler system for performing the method as described in any one of claims 1-8, characterized in that, include: The first acquisition module acquires test data of the photovoltaic panel's own status and test data of the surrounding environment at multiple test moments during the test process. The test data of the surrounding environment includes test data of the irradiance received by the photovoltaic panel, test data of the ambient temperature of the environment where the photovoltaic panel is located, test data of the ambient humidity, test data of the wind speed, and test angle between the wind direction and the orientation of the photovoltaic panel. The test data of the photovoltaic panel's own status includes test data of the photovoltaic panel's own temperature and test data of the photovoltaic panel's power generation. The first training module trains the power generation data prediction model based on the test data of the photovoltaic panel's own state and the test data of the surrounding environment, and obtains the trained power generation data prediction model. The second acquisition module acquires test data on the photovoltaic panel's own status, test data on the surrounding environment, and data on the amount of water sprayed before and after multiple spray tests. The second training module trains the spray control model based on the self-state test data before and after the spray test, the surrounding environment test data before the spray test, and the spray water volume data, and obtains the trained spray control model. The data acquisition module collects data on the photovoltaic panel's own status and the surrounding environment during the photovoltaic panel power generation process. The surrounding environment data includes the irradiance data received by the test photovoltaic panel, the ambient temperature data, ambient humidity data, wind speed data, and the angle between the wind direction and the photovoltaic panel's orientation. The photovoltaic panel's own status data includes the photovoltaic panel's own temperature data and the photovoltaic panel's power generation capacity. The photovoltaic panel is the same model as the test photovoltaic panel. The judgment module determines whether spraying is necessary based on its own status data, surrounding environmental data, and the trained power generation data prediction model. The sprinkler water volume optimization module determines the optimal sprinkler water volume based on its own status data, surrounding environment data, and the trained sprinkler control model if sprinkler operation is required.

Images