A method, system and storage medium for optimizing photovoltaic power supply in a computer room

Abstract

The present invention discloses a method, system and storage medium for optimizing photovoltaic power supply in a computer room. The method includes obtaining the solar radiation intensity and the angular image, temperature image and absorption coefficient of the photovoltaic panel; using a trained neural network model to perform feature extraction and target detection on the angular image to obtain a target detection frame and an average angular feature; calculating an average temperature feature based on the target detection frame and the temperature image; calculating an optimal efficiency angle by using a machine learning method according to the average angular feature, the average temperature feature, the absorption coefficient and the solar radiation intensity, and adjusting the photovoltaic panel based on the optimal efficiency angle. The present invention can accurately determine the optimal working angle of the photovoltaic panel without manual intervention, keep the photovoltaic system outputting the optimal efficiency at each angle, improve the energy collection efficiency and energy utilization rate of the photovoltaic system, and achieve sustainable power supply and maximized energy collection of the photovoltaic system. The present invention is applied to the field of artificial intelligence technology.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and in particular to a method, system and storage medium for optimizing photovoltaic power supply in a computer room. Background Art

[0002] Photovoltaic power generation is a technology that directly converts light energy into electrical energy by using the photovoltaic effect at the semiconductor interface, mainly consisting of three major parts: solar photovoltaic panels, controllers, and inverters. Related technologies usually adjust the angle of photovoltaic panels in a way based on preset values or manual intervention to change the energy output effect of the photovoltaic panels. However, this rigid adjustment method lacks real-time performance and intelligence, which makes the photovoltaic system unable to quickly respond to the dynamic changes in environmental light, restricts the energy collection efficiency and power supply sustainability of the photovoltaic system, and further causes the photovoltaic system to be difficult to maintain the optimal efficiency output at each angle, resulting in fluctuations in power supply efficiency. In addition, the temperature change of the photovoltaic panel has a significant negative impact on the efficiency of the photovoltaic panel. Especially in a high-temperature environment, the performance of the photovoltaic panel seriously deteriorates, and related technologies often do not fully consider the influence of environmental temperature factors, lacking an adjustment strategy for environmental temperature factors, making it difficult for the photovoltaic system to cope with the dynamic changes in environmental temperature. Summary of the Invention

[0003] An object of the present invention is to solve at least to some extent one of the technical problems existing in the prior art.

[0004] To this end, an object of the present invention is to provide a method, system and storage medium for optimizing photovoltaic power supply in a computer room.

[0005] In order to achieve the above technical object, the technical solutions adopted in the embodiments of the present invention include:

[0006] On the one hand, an embodiment of the present invention provides a method for optimizing photovoltaic power supply in a computer room, including the following steps:

[0007] Obtain the solar radiation intensity, as well as the angle image, temperature image and absorption coefficient of the photovoltaic panel;

[0008] Use the trained neural network model to perform feature extraction and target detection on the angle image of the photovoltaic panel to obtain the target detection frame and average angle feature of the photovoltaic panel;

[0009] Calculate the average temperature feature of the photovoltaic panel according to the target detection frame and temperature image of the photovoltaic panel;

[0010] According to the average angle feature, average temperature feature and absorption coefficient of the photovoltaic panel, as well as the solar radiation intensity, use machine learning methods to calculate the optimal efficiency angle of the photovoltaic panel;

[0011] Adjust the photovoltaic panel based on the optimal efficiency angle.

[0012] According to the method for optimizing photovoltaic power supply in a computer room according to the above embodiments of the present invention, the following additional technical features may also be included:

[0013] Further, in an embodiment of the present invention, the step of obtaining the absorption coefficient of the photovoltaic panel includes:

[0014] Collect the real-time light intensity of the photovoltaic panel and obtain the reference light intensity of the photovoltaic panel;

[0015] Calculate the absorption coefficient of the photovoltaic panel according to the real-time light intensity and the reference light intensity.

[0016] Further, in an embodiment of the present invention, before using the trained neural network model to perform feature extraction and target detection on the angle image of the photovoltaic panel, the method further includes the following steps:

[0017] Unify the pixel sizes of the angle image and the temperature image;

[0018] Perform data augmentation on the angle image, and the data augmentation processing methods at least include random rotation and mirror operations.

[0019] Further, in an embodiment of the present invention, the trained neural network model includes a first feature map extraction module, a second feature map extraction module, a third feature map extraction module, and a convolution module. Using the trained neural network model to perform feature extraction and target detection on the angle image of the photovoltaic panel includes:

[0020] Perform depthwise separable convolution processing and de-linearization processing on the angle image through the first feature map extraction module to obtain a first feature map;

[0021] Perform depthwise separable convolution processing and de-linearization processing on the first feature map through the second feature map extraction module to obtain a second feature map;

[0022] Perform depthwise separable convolution processing and de-linearization processing on the second feature map through the third feature map extraction module to obtain a third feature map;

[0023] Perform convolution operation on the third feature map through the convolution module to generate a target detection frame;

[0024] Wherein, the convolution module includes at least five convolutional layers.

[0025] Further, in an embodiment of the present invention, the first feature map extraction module includes a depthwise separable convolutional layer, and the second feature map extraction module and the third feature map extraction module both include two depthwise separable convolutional layers connected in sequence, and a ReLU activation function is provided behind each depthwise separable convolutional layer.

[0026] Further, in an embodiment of the present invention, the trained neural network model further includes a first angle feature extraction module, a second angle feature extraction module, a third angle feature extraction module, and a feature output layer. The use of the trained neural network model to perform feature extraction and target detection on the angle image of the photovoltaic panel further includes:

[0027] Extracting angle features from the first feature map through the first angle feature extraction module to obtain first angle features;

[0028] Extracting angle features from the second feature map through the second angle feature extraction module to obtain second angle features;

[0029] Extracting angle features from the third feature map through the third angle feature extraction module to obtain third angle features;

[0030] Calculating and obtaining average angle features through the feature output layer based on the first angle features, the second angle features, and the third angle features.

[0031] Further, in an embodiment of the present invention, the first angle feature extraction module, the second angle feature extraction module, and the third angle feature extraction module all include a binarization layer, an angle feature extraction layer, an angle feature operation layer, and an angle output layer connected in sequence, wherein,

[0032] The binarization layer is used to perform binarization processing on the feature map input to the first angle feature extraction module, the second angle feature extraction module, or the third angle feature extraction module to obtain a binarized feature map;

[0033] The angle feature extraction layer is used to perform multiple feature extractions on the binarized feature map by using an improved convolution operator to obtain multiple convolution feature values;

[0034] The angle feature operation layer is used to calculate an initial angle feature value corresponding to each convolution feature value according to each convolution feature value;

[0035] The angle output layer is used to calculate and obtain the first angle feature, the second angle feature, or the third angle feature according to multiple initial angle feature values and the size value of the binarized feature map.

[0036] Further, in an embodiment of the present invention, the steps of extracting features from the binarized feature map using an improved convolution operator to obtain convolution eigenvalues include:

[0037] Calculating the convolution eigenvalue according to the current position value of the improved convolution operator and the position value at the corresponding position in the binarized feature map.

[0038] Further, in an embodiment of the present invention, the calculating the convolution eigenvalue according to the current position value of the improved convolution operator and the position value at the corresponding position in the binarized feature map includes:

[0039] When the product of the second position value of the current position value of the improved convolution operator and the position value at the corresponding position in the binarized feature map is greater than 0, using the first position value of the current position value of the improved convolution operator as the convolution eigenvalue.

[0040] Further, in an embodiment of the present invention, the calculating the convolution eigenvalue according to the current position value of the improved convolution operator and the position value at the corresponding position in the binarized feature map further includes:

[0041] When the product of the second position value of the current position value of the improved convolution operator and the position value at the corresponding position in the binarized feature map is less than or equal to 0, using a null value as the convolution eigenvalue.

[0042] Further, in an embodiment of the present invention, the current position value of the improved convolution operator is the union of the coordinate value of the current position in the improved convolution operator and a first numerical value.

[0043] Further, in an embodiment of the present invention, the size of the improved convolution operator is 10×10.

[0044] Further, in an embodiment of the present invention, the steps of calculating the initial angle eigenvalue corresponding to each convolution eigenvalue according to each convolution eigenvalue include: calculating the initial angle eigenvalue corresponding to each convolution eigenvalue according to the position values included in the upper left corner positioning point and the lower right corner positioning point of each convolution eigenvalue.

[0045] Further, in an embodiment of the present invention, the calculating the average angle feature by the feature output layer based on the first angle feature, the second angle feature, and the third angle feature includes: obtaining the valid pixel points in the binarized feature map, and calculating the average angle feature according to the number of valid pixel points of the first feature map, the second feature map, and the third feature map, in combination with the first angle feature, the second angle feature, and the third angle feature.

[0046] Further, in an embodiment of the present invention, the pixel points with the pixel value of the first numerical value in the binarized feature map are used as valid pixel points.

[0047] Further, in an embodiment of the present invention, calculating the average temperature feature of the photovoltaic panel according to the target detection frame and the temperature image of the photovoltaic panel includes:

[0048] Cropping the temperature image according to the coordinate information of the target detection frame in the angle image;

[0049] Calculating the average temperature feature according to the number and pixel values of the pixel points of the cropped temperature image.

[0050] Further, in an embodiment of the present invention, before calculating the optimal efficiency angle of the photovoltaic panel by using a machine learning method according to the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel, the method further includes the following steps:

[0051] Normalizing the solar radiation intensity and the absorption coefficient of the photovoltaic panel.

[0052] Further, in an embodiment of the present invention, calculating the optimal efficiency angle of the photovoltaic panel by using a machine learning method according to the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel includes:

[0053] Calculating the optimal efficiency angle of the photovoltaic panel by using an SVM model according to the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel.

[0054] Further, in an embodiment of the present invention, a Gaussian kernel function is used as the kernel function of the SVM model.

[0055] On the other hand, an embodiment of the present invention provides a computer room photovoltaic power supply optimization system, including:

[0056] An acquisition module, configured to acquire the solar radiation intensity, and the angle image, temperature image and absorption coefficient of the photovoltaic panel;

[0057] An angle feature processing module, configured to perform feature extraction and target detection on the angle image of the photovoltaic panel by using a trained neural network model to obtain the target detection frame and the average angle feature of the photovoltaic panel;

[0058] A temperature feature processing module, configured to calculate the average temperature feature of the photovoltaic panel according to the target detection frame and the temperature image of the photovoltaic panel;

[0059] An angle calculation module, configured to calculate the optimal efficiency angle of a photovoltaic panel by using a machine learning method according to the average angle feature, average temperature feature, absorption coefficient of the photovoltaic panel, and solar radiation intensity;

[0060] An adjustment module, configured to adjust the photovoltaic panel based on the optimal efficiency angle.

[0061] In another aspect, an embodiment of the present invention provides a storage medium, in which a program executable by a processor is stored, and the program executable by the processor is used to implement the above-mentioned method for optimizing photovoltaic power supply in a computer room when executed by the processor.

[0062] The beneficial effects of the present invention are as follows: A method, system and storage medium for optimizing photovoltaic power supply in a computer room are provided. After obtaining the solar radiation intensity, angle image, temperature image and absorption coefficient of the photovoltaic panel, the trained neural network model is used to extract features and detect targets from the angle image to obtain the average angle feature of the photovoltaic panel. Then, the average temperature feature of the photovoltaic panel is obtained by combining the temperature image with the result of the target detection, i.e., the target detection frame. Then, the optimal efficiency angle of the photovoltaic panel is calculated by using a machine learning method in combination with the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity. Finally, the photovoltaic panel is adjusted according to the optimal efficiency angle. By fusing multi-dimensional data, the embodiment of the present invention comprehensively considers multiple factors such as light, panel angle and temperature, and can accurately determine the optimal efficiency angle of the photovoltaic panel without manual intervention, so that the photovoltaic panel can quickly adjust the working angle when the light intensity or temperature changes, thereby enabling the photovoltaic system to maintain the optimal efficiency output at each angle, effectively improving the energy collection efficiency and energy utilization rate of the photovoltaic system while reducing the burden and cost of manual maintenance, significantly reducing energy waste, and realizing sustainable power supply and maximum energy collection of the photovoltaic system.

[0063] Other features and advantages of the present application will be described in the following specification, and part of them will become obvious from the specification, or will be understood by implementing the present application. The objectives and other advantages of the present application can be achieved and obtained through the structures specifically pointed out in the specification, claims and drawings. Description of the Drawings

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following introduces the accompanying drawings related to the technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings below are only for conveniently and clearly presenting some embodiments of the technical solutions in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative efforts.

[0065] Figure 1 It is a flowchart of an optimization method for photovoltaic power supply in a computer room provided by the present invention;

[0066] Figure 2 It is a schematic structural diagram of a neural network model provided by the present invention;

[0067] Figure 3 It is another schematic structural diagram of a neural network model provided by the present invention;

[0068] Figure 4 It is a flowchart of target detection provided by the present invention;

[0069] Figure 5 It is a flowchart of angle feature extraction provided by the present invention;

[0070] Figure 6 It is a schematic diagram of the principle for calculating angle features provided by the present invention;

[0071] Figure 7 It is a schematic structural diagram of an improved convolution operator provided by the present invention;

[0072] Figure 8 It is a schematic diagram of the principle for calculating the optimal efficiency angle of a photovoltaic panel provided by the present invention;

[0073] Figure 9 It is a schematic diagram of the principle for photovoltaic power supply optimization provided by the present invention;

[0074] Figure 10 It is a structural diagram of a photovoltaic power supply optimization system for a computer room provided by the present invention. Detailed implementation manners

[0075] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0076] The present application will be further described below with reference to the accompanying drawings of the specification and specific embodiments. The described embodiments should not be regarded as a limitation of the present application. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0077] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments. However, it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0078] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which this application belongs. The terms used herein are for the purpose of describing embodiments of this application only and are not intended to limit this application.

[0079] Photovoltaic power generation is a technology that directly converts light energy into electrical energy by using the photovoltaic effect at the semiconductor interface. It mainly consists of three major parts: solar photovoltaic panels, controllers, and inverters. Related technologies usually adjust the angle of photovoltaic panels in a way based on preset values or manual intervention to change the energy output effect of the photovoltaic panels. However, this rigid adjustment method lacks real-time performance and intelligence, which makes the photovoltaic system unable to quickly respond to the dynamic changes of environmental light. The energy collection efficiency and power supply sustainability of the photovoltaic system are limited. As a result, it is difficult for the photovoltaic system to maintain the optimal efficiency output at each angle, leading to fluctuations in power supply efficiency. In addition, the temperature change of the photovoltaic panel has a significant negative impact on the efficiency of the photovoltaic panel. Especially in a high-temperature environment, the performance of the photovoltaic panel seriously deteriorates. However, related technologies often do not fully consider the influence of environmental temperature factors and lack adjustment strategies for environmental temperature factors, making it difficult for the photovoltaic system to cope with the dynamic changes of environmental temperature.

[0080] Based on this, the embodiments of the present invention provide a method, system, and storage medium for optimizing photovoltaic power supply in a computer room. By real-time obtaining multi-dimensional data such as solar radiation intensity, angle image, temperature image, and absorption coefficient of the photovoltaic panel, an improved deep learning algorithm is introduced to extract features and perform object detection on the angle image to obtain the average angle feature of the photovoltaic panel. Then, based on the result of object detection, that is, the object detection frame, the average temperature feature of the photovoltaic panel is obtained in combination with the temperature image. Then, by combining the average angle feature, average temperature feature, absorption coefficient, and solar radiation intensity, the optimal efficiency angle of the photovoltaic panel is calculated using machine learning methods. Finally, the photovoltaic panel is adjusted according to the optimal efficiency angle. The embodiments of the present invention can accurately determine the optimal efficiency angle of the photovoltaic panel without manual intervention under the condition of comprehensively considering multiple factors such as light, panel angle, and temperature, realizing the dynamic adjustment of the photovoltaic panel, enabling the photovoltaic panel to adapt to different light environment conditions and temperature environment conditions, and significantly improving the energy collection efficiency.

[0081] The following further elaborates on the embodiments of this application in conjunction with the accompanying drawings.

[0082] First, an implementation step of a method for optimizing photovoltaic power supply in a computer room proposed by the embodiments of the present invention will be described in detail below.

[0083] An optimization method for photovoltaic power supply in a computer room provided by an embodiment of the present invention can be applied to a terminal, a server, or software running on a terminal or a server. The terminal can be a tablet computer, a laptop computer, a desktop computer, etc., but is not limited thereto. The server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. In addition, the server can also be a node server in a blockchain network, but is not limited thereto. Among them, blockchain is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithms.

[0084] Referring to Figure 1 , Figure 1 is a flowchart of an optimization method for photovoltaic power supply in a computer room provided by the present invention. An optimization method for photovoltaic power supply in a computer room provided by an embodiment of the present invention may include but is not limited to the following steps:

[0085] S101, obtaining the solar radiation intensity and the angular image, temperature image, and absorption coefficient of the photovoltaic panel.

[0086] In this step, the solar radiation intensity and the key data of the photovoltaic panel are obtained through an acquisition module. The key data of the photovoltaic panel may include but is not limited to the angular image, temperature image, and absorption coefficient of the photovoltaic panel.

[0087] Optionally, the acquisition module may include but is not limited to a drone acquisition module and a parameter calculation module. In an embodiment of the present invention, the drone acquisition module is used to collect the solar radiation intensity, the angular image, temperature image, and real-time illumination intensity of the photovoltaic panel, and the parameter calculation module processes the real-time illumination intensity of the photovoltaic panel to obtain the absorption coefficient of the photovoltaic panel.

[0088] S102, using the trained neural network model to perform feature extraction and target detection on the angular image of the photovoltaic panel to obtain the target detection frame and average angular feature of the photovoltaic panel.

[0089] In this step, in order to improve the optimization efficiency, the trained neural network model can be used to perform feature extraction and target detection on the angular image of the photovoltaic panel, and then the target detection frame and average angular feature of the photovoltaic panel are obtained.

[0090] It should be noted that the present invention does not specifically limit the specific network structure of the neural network model, and those skilled in the art can flexibly select according to actual needs. For example, a convolutional neural network, a recurrent neural network, etc. can be selected.

[0091] Exemplarily, MobileNet is used as the base network to construct a neural network model, and the neural network model is trained using the sample angle images of the labeled photovoltaic panels and the corresponding labeling results to obtain a trained neural network model.

[0092] It can be understood that MobileNet is a lightweight deep neural network model architecture. It adopts a series of technologies such as depthwise separable convolution, enabling the network structure to efficiently perform computer vision tasks such as image classification and object detection under limited computing resources. This network structure can reduce the computational and memory requirements of the model while maintaining high model performance, and is suitable for mobile devices and embedded systems, etc.

[0093] S103, Calculate the average temperature feature of the photovoltaic panel based on the target detection frame and temperature image of the photovoltaic panel.

[0094] In the previous step S102, target detection is performed on the angle image of the photovoltaic panel through the trained neural network model to obtain the target detection frame of the photovoltaic panel. In this step, the average temperature feature of the photovoltaic panel is calculated using the result of this target detection, i.e., the target detection frame of the photovoltaic panel, and the temperature image of the photovoltaic panel.

[0095] S104, Calculate the optimal efficiency angle of the photovoltaic panel using machine learning methods based on the average angle feature, average temperature feature, absorption coefficient, and solar radiation intensity of the photovoltaic panel.

[0096] Optionally, the machine learning method can be the support vector machine algorithm (Support vector machine, SVM). Those skilled in the art can understand that other machine learning algorithms such as the decision tree algorithm (Decision Tree, DT), random forest algorithm (Random Forest, RF), and logistic regression algorithm (Logistic Regression, LR) are also equally applicable, and only need to be adjusted accordingly.

[0097] S105, Adjust the photovoltaic panel based on the optimal efficiency angle.

[0098] In this step, the angle of the photovoltaic panel is adjusted using the optimal efficiency angle obtained in the previous step S105, so that under the current light conditions and temperature conditions, the photovoltaic panel can achieve the optimal efficiency output at each angle, ensuring the power supply sustainability and maximum energy collection of the photovoltaic system.

[0099] In some embodiments of the present invention, the UAV acquisition module is equipped with an angular image acquisition unit, an infrared image acquisition unit, a radiation intensity acquisition unit, and an illumination intensity acquisition unit.

[0100] Optionally, the angular image acquisition unit can be a camera or other devices for acquiring images such as an image sensor (Charge-coupled Device, CCD). The infrared image acquisition unit can be a thermal infrared camera or other devices for detecting temperature changes or acquiring temperature images such as a thermal imager. The radiation intensity acquisition unit can be a light sensor or other devices for detecting solar radiation intensity such as a pyranometer or a solar tracker. The illumination intensity acquisition unit can be a spectroradiometer or other devices for measuring illumination intensity such as a light sensor.

[0101] The implementation process of obtaining the solar radiation intensity, the angular image, the temperature image, and the absorption coefficient of the photovoltaic panel in step S101 may include but is not limited to the following steps:

[0102] S1011, when the UAV reaches the acquisition position, collect the solar radiation intensity, the angular image, the temperature image, and the real-time illumination intensity of the photovoltaic panel through the UAV acquisition module.

[0103] Optionally, the upper right front of the photovoltaic panel is used as the acquisition position.

[0104] In this step, the angular image of the photovoltaic panel is collected through the angular image acquisition unit, the temperature image of the photovoltaic panel under the current temperature condition is collected through the infrared image acquisition unit, the real-time illumination intensity of the photovoltaic panel under the current illumination condition is collected through the illumination intensity acquisition unit, and the solar radiation intensity under the current illumination condition is collected through the radiation intensity acquisition unit.

[0105] S1012, obtain the reference illumination intensity of the photovoltaic panel.

[0106] It should be noted that the reference illumination intensity of the photovoltaic panel is defined as the illumination intensity of the preset reference surface of the UAV acquisition module under the current illumination condition.

[0107] S1013, calculate the absorption coefficient of the photovoltaic panel according to the real-time illumination intensity and the reference illumination intensity.

[0108] It should be noted that the absorption coefficient of the photovoltaic panel is defined as the absorption performance of the acquisition position of the photovoltaic panel relative to the preset reference surface of the UAV acquisition module.

[0109] Optionally, the absorption coefficient of the photovoltaic panel is calculated by the following formula (1) in the embodiments of the present invention:

[0110]

[0111] In the above formula (1), θ represents the absorption coefficient of the photovoltaic panel, and Li Sp represents the real-time illumination intensity of the photovoltaic panel, and Li Ref represents the reference illumination intensity of the photovoltaic panel.

[0112] In some embodiments of the present invention, before using the trained neural network model to extract features and detect targets from the angle image of the photovoltaic panel in step S102, a method for optimizing photovoltaic power supply in a computer room provided by the embodiments of the present invention may further include the following steps:

[0113] Unify the pixel sizes of the angle image and the temperature image.

[0114] In this step, before extracting features and detecting targets from the angle image, it is necessary to modify the pixel sizes of the angle image and the temperature image of the photovoltaic panel. More specifically, the embodiments of the present invention uniformly modify the pixel sizes of the angle image and the temperature image of the photovoltaic panel to 648×648.

[0115] It should be noted that the present invention does not specifically limit the method for modifying the pixel size of the image, and those skilled in the art can flexibly select according to actual needs. For example, interpolation algorithms, non-uniform scaling, uniform scaling, etc. can be selected to modify the pixel size of the image. Among them, the interpolation algorithm can be bilinear, based on pixel region relationship, Lanczos interpolation, etc.

[0116] Perform data augmentation on the angle image.

[0117] Optionally, the data augmentation processing methods at least include random rotation and mirror operations, and may also include one or more of adding Gaussian noise, shearing, random translation and scaling, randomly changing the image brightness, hue adjustment, motion blur, random cropping, and vertical flipping. The embodiments of the present invention do not specifically limit this.

[0118] In this step, before inputting the angle image into the trained neural network model, performing data augmentation on the angle image can not only effectively avoid the occurrence of overfitting of the neural network model, but also enhance the robustness and generalization ability of the neural network model, and improve the detection accuracy of the neural network model in the target detection task.

[0119] In some embodiments of the present invention, before calculating the optimal efficiency angle of the photovoltaic panel by using machine learning methods based on the average angle feature, average temperature feature, absorption coefficient, and solar radiation intensity of the photovoltaic panel in step S104, a method for optimizing photovoltaic power supply in a computer room provided by the embodiments of the present invention may further include the following steps:

[0120] Normalize the solar radiation intensity and the absorption coefficient of the photovoltaic panel.

[0121] In this step, since the solar radiation intensity and the absorption coefficient of the photovoltaic panel belong to different data characteristics, it is necessary to normalize the solar radiation intensity and the absorption coefficient of the photovoltaic panel, so that both the solar radiation intensity and the absorption coefficient of the photovoltaic panel are restricted within a certain fixed range, facilitating subsequent data processing and analysis.

[0122] In some embodiments of the present invention, for the angle recognition task of the photovoltaic panel, the embodiments of the present invention propose an improved neural network model based on MobileNet, aiming to improve the detection rate of the photovoltaic panel and the feature extraction rate of the angle feature, and at the same time improve the accuracy of object detection and feature extraction. Refer to Figure 2 , Figure 2 is a schematic structural diagram of the neural network model provided by the present invention. The neural network model 100 proposed in the embodiments of the present invention mainly includes an object detection branch structure 110 and an angle feature extraction branch structure 120. Through multiple feature map extraction modules of the object detection branch structure 110, depthwise separable convolution processing and de-linearization processing are performed on the angle image to obtain the object detection box of the photovoltaic panel, thereby realizing the object detection of the photovoltaic panel; and in the process of object detection of the angle image, the angle feature extraction branch structure 120 is used to extract and calculate the angle features of the output of each feature map extraction module respectively, thereby obtaining the average angle feature of the photovoltaic panel.

[0123] Specifically, the object detection branch structure 110 in the neural network model 100 proposed in the embodiments of the present invention mainly includes:

[0124] The first feature map extraction module 111 is used to perform depthwise separable convolution processing and de-linearization processing on the angle image input to the first feature map extraction module 111 to generate a first feature map;

[0125] The second feature map extraction module 112 is used to perform depthwise separable convolution processing and de-linearization processing on the first feature map output by the first feature map extraction module 111 to generate a second feature map;

[0126] The third feature map extraction module 113 is used to perform depthwise separable convolution processing and de-linearization processing on the second feature map output by the second feature map extraction module 112 to generate a third feature map;

[0127] The convolution module 114 is used to perform convolution operation on the third feature map output by the third feature map extraction module 113 to generate an object detection box.

[0128] In the embodiment of the present invention, in the target detection branch structure 110, the angular image is subjected to multiple depthwise separable convolution processes and de-linearization processes in sequence through the first feature map extraction module 111, the second feature map extraction module 112, and the third feature map extraction module 113, and finally, a convolution operation is performed on the output of the third feature map extraction module 113 through the convolution module 114, thereby obtaining the target detection box of the photovoltaic panel and realizing the target detection of the photovoltaic panel. Among them, the result of the target detection of the photovoltaic panel will be combined with the temperature image in the subsequent steps to calculate the average temperature feature of the photovoltaic panel.

[0129] Specifically, the angular feature extraction branch structure 120 in the neural network model 100 proposed in the embodiment of the present invention mainly includes:

[0130] The first angular feature extraction module 121 is used to extract the angular feature from the first feature map output by the first feature map extraction module 111 to obtain the first angular feature;

[0131] The second angular feature extraction module 122 is used to extract the angular feature from the second feature map output by the second feature map extraction module 112 to obtain the second angular feature;

[0132] The third angular feature extraction module 123 is used to extract the angular feature from the third feature map output by the third feature map extraction module 113 to obtain the third angular feature;

[0133] The feature output layer 124 is used to calculate the average angular feature according to the first angular feature, the second angular feature, and the third angular feature.

[0134] In the embodiment of the present invention, in the angular feature extraction branch structure 120, the input end of the first angular feature extraction module 121 is connected to the output end of the first feature map extraction module 111, the input end of the second angular feature extraction module 122 is connected to the output end of the second feature map extraction module 112, and the input end of the third angular feature extraction module 123 is connected to the output end of the third feature map extraction module 113. The first angular feature extraction module 121 extracts the angular feature from the output of the first feature map extraction module 111, the second angular feature extraction module 122 extracts the angular feature from the output of the second feature map extraction module 112, and the third angular feature extraction module 123 extracts the angular feature from the output of the third feature map extraction module 113. Finally, the feature output layer 124 synthesizes the extraction results of the three angular features to calculate the average angular feature.

[0135] In some embodiments of the present invention, refer to Figure 3 , Figure 3It is another structural schematic diagram of the neural network model provided by the present invention. In the object detection branch structure 110 of the neural network model 100 provided in the embodiments of the present invention, the first feature map extraction module 111 includes a depthwise separable convolutional layer conv1, and both the second feature map extraction module 112 and the third feature map extraction module 113 include two depthwise separable convolutional layers conv1 connected in sequence. The convolutional module 114 includes at least five convolutional layers conv2.

[0136] It should be noted that depthwise separable convolution is a decomposable convolution operation. Depthwise separable convolution can be decomposed into two smaller operations: depthwise convolution and pointwise convolution. Depthwise convolution is used to perform convolution operations on each channel of the data without changing the depth of the data, and pointwise convolution is used to increase and decrease the dimension of the data. Generally, when performing depthwise separable convolution processing, depthwise convolution is first performed on the data, and then pointwise convolution is performed on the data after depthwise convolution processing. Depthwise separable convolution can effectively reduce the number of parameters and the amount of computation of the neural network model 100.

[0137] Optionally, the size of the convolution kernel of each convolutional layer conv2 is 3×3, and the padding of each convolutional layer conv2 is 1. Those skilled in the art can understand that the padding of the convolutional layer refers to filling elements on both sides of the input height and width, and the padding of the convolutional layer conv2 is one of the hyperparameters of the convolutional layer.

[0138] More specifically, an activation function is provided behind each depthwise separable convolutional layer conv1.

[0139] Optionally, the activation function is the ReLU activation function.

[0140] In the embodiments of the present invention, the depthwise separable convolutional layer conv1 is used to perform depthwise separable convolution operations on the input feature map, and the activation function is used to perform non-linear processing on the output of each depthwise separable convolutional layer conv1. While realizing feature extraction, it can effectively reduce the number of parameters and the computational cost of the neural network model 100, enhance the expression ability of the neural network model 100, and help to accelerate the running speed of the neural network model 100. In addition, in the embodiments of the present invention, the result of object detection is output through at least five convolutional layers conv2, which can ensure that the size of the angular image to which the object detection box belongs remains unchanged.

[0141] In some embodiments of the present invention, referring to Figure 3 and Figure 4 , Figure 4It is the target detection flowchart of the angle image of the photovoltaic panel provided by the present invention. In step S102, the implementation process of using the trained neural network model 100 to extract features and detect targets from the angle image of the photovoltaic panel may include but is not limited to the following steps:

[0142] A1, perform depthwise separable convolution processing and de-linearization processing on the angle image through the first feature map extraction module to obtain the first feature map.

[0143] In this step, perform depthwise separable convolution processing on the angle image through the single-layer depthwise separable convolution layer conv1 in the first feature map extraction module 111, and perform de-linearization processing on the convolved angle image through the activation function set behind the depthwise separable convolution layer conv1, thereby obtaining the first feature map.

[0144] A2, perform depthwise separable convolution processing and de-linearization processing on the first feature map through the second feature map extraction module to obtain the second feature map.

[0145] In this step, perform depthwise separable convolution processing on the first feature map through the two-layer depthwise separable convolution layer conv1 in the second feature map extraction module 112, and perform de-linearization processing on the output of the depthwise separable convolution layer conv1 through the activation function set behind the depthwise separable convolution layer conv1, and finally obtain the second feature map.

[0146] A3, perform depthwise separable convolution processing and de-linearization processing on the second feature map through the third feature map extraction module to obtain the third feature map.

[0147] In this step, perform depthwise separable convolution processing on the second feature map through the two-layer depthwise separable convolution layer conv1 in the third feature map extraction module 113, and perform de-linearization processing on the output of the depthwise separable convolution layer conv1 through the activation function set behind the depthwise separable convolution layer conv1, and finally obtain the third feature map.

[0148] A4, perform convolution operation on the third feature map through the convolution module to generate the target detection box.

[0149] In this step, perform convolution operation on the third feature map through at least five convolution layers conv2 to obtain the target detection box of the photovoltaic panel.

[0150] In some embodiments of the present invention, refer to Figure 3 and Figure 5 , Figure 5It is a flowchart for extracting angular features of the angular image of a photovoltaic panel provided by the present invention. In step S102, the implementation process of using the trained neural network model 100 to extract features and detect targets from the angular image of the photovoltaic panel may further include, but is not limited to, the following steps:

[0151] B1. Extract angular features from the first feature map through the first angular feature extraction module to obtain the first angular feature;

[0152] B2. Extract angular features from the second feature map through the second angular feature extraction module to obtain the second angular feature;

[0153] B3. Extract angular features from the third feature map through the third angular feature extraction module to obtain the third angular feature;

[0154] B4. Calculate the average angular feature based on the first angular feature, the second angular feature, and the third angular feature through the feature output layer.

[0155] In some embodiments of the present invention, referring to Figure 3 and Figure 6 , Figure 6 is a schematic diagram for calculating angular features. In the angular feature extraction branch structure 120 of the neural network model 100 provided in the embodiments of the present invention, the first angular feature extraction module 121, the second angular feature extraction module 122, and the third angular feature extraction module 123 all include a binarization layer Binar1, an angular feature extraction layer con3, an angular feature operation layer calculate1, and an angular output layer output1 that are connected in sequence. Among them:

[0156] The function of the binarization layer Binar1 includes binarizing the feature map input to the first angular feature extraction module 121, the second angular feature extraction module 122, or the third angular feature extraction module 123 to obtain a binarized feature map.

[0157] It can be understood that binarization processing means that by setting a threshold, the original field greater than the threshold is configured as 1, and the original field less than or equal to the threshold is configured as 0. In the embodiments of the present invention, by binarizing the feature map input to the first angular feature extraction module 121, the second angular feature extraction module 122, or the third angular feature extraction module 123, the noise data in the feature map can be effectively eliminated.

[0158] The function of the angular feature extraction layer con3 includes using an improved convolution operator to perform multiple feature extractions on the binarized feature map to obtain multiple convolution feature values.

[0159] Specifically, in the embodiments of the present invention, by improving the convolution operator, the improved convolution operator is more sensitive to the high-order angular features in the angular image, which helps to extract the high-order angular features in the angular image.

[0160] Optionally, the number of times of feature extraction of the angular feature extraction layer con3 satisfies where H represents the height of the binarized feature map, W represents the width of the binarized feature map, K×K represents the size of the improved convolution operator, and P h represents the padding in the height direction of the binarized feature map, and P w represents the padding in the width direction of the binarized feature map, and S represents the stride of the improved convolution operator.

[0161] In the embodiments of the present invention, the size of the improved convolution operator is 10×10, that is, K is 10, the padding in the height direction of the binarized feature map and the padding in the width direction of the binarized feature map are both 0, that is, P h and P w are both 0, and the stride of the improved convolution operator is 1, that is, S is 1. Then when the size of the binarized feature map is 648×648, the number of times the angular feature extraction layer con3 needs to perform feature extraction on the binarized feature map is 639×639.

[0162] The function of the angular feature operation layer calculate1 includes calculating the initial angular feature value corresponding to each convolution feature value according to each convolution feature value.

[0163] Specifically, in the embodiments of the present invention, through the angular feature operation layer calculate1, each convolution feature value is further operated and processed in sequence, and multiple initial angular feature values can be obtained, and each initial angular feature value corresponds to each convolution feature value.

[0164] Optionally, the number of initial angular feature values is equal to the number of times of feature extraction of the angular feature extraction layer con3. For example, when the size of the binarized feature map is 648×648 and the number of times of feature extraction of the angular feature extraction layer con3 is 639×639, the number of initial angular feature values is also 639×639.

[0165] The function of the angular output layer output1 includes calculating the first angular feature, the second angular feature or the third angular feature according to the multiple initial angular feature values and the size value of the binarized feature map.

[0166] Specifically, in the embodiments of the present invention, the method of taking the average value can be adopted to integrate multiple initial angular feature values to generate the final angular feature.

[0167] Further, referring toFigure 7 , Figure 7 : is a schematic diagram of the structure of the improved convolution operator provided by the present invention. The size of the improved convolution operator provided by the embodiment of the present invention is 10×10. The current point value of the improved convolution operator is the union of the coordinate value of the current point in the improved convolution operator and the first value. Optionally, the first value is 1. For example, the upper left corner of the improved convolution operator is Oc 1，1 [(0,0),1],Oc 1，1 It represents the value contained in the first row and first column of the improved convolution operator. For example, the lower right corner of the improved convolution operator is Oc 10，10 [(10, 10), 1], Oc 10，10 It represents the value contained in the 10th row and 10th column of the improved convolution operator.

[0168] Furthermore, in the angle feature extraction layer con3 of any angle feature extraction module, the improved convolution operator is used to extract features from the binarized feature map, and the steps of obtaining the convolution feature value mainly include:

[0169] The convolution feature value is calculated based on the current point value of the improved convolution operator and the point value of the corresponding position in the binary feature map.

[0170] In this step, the convolution feature value is calculated by using the current point value of the improved convolution operator and the point value of the corresponding position in the binarized feature map, combined with the positional product formula.

[0171] Optionally, the positional product formula is as shown in the following formula (2):

[0172]

[0173] In the above formula (2), δ i,j Represents the point value of the i-th row and j-th column in the convolution feature value, OC i，j [0] represents the first value of the point value of the i-th row and j-th column in the improved convolution operator, OC i,j [1] represents the second bit value of the point value in the i-th row and j-th column in the improved convolution operator, Empty represents the empty value, and FeatureMap represents the feature map after binarization.

[0174] As can be seen from the above formula (2), when the product of the second bit value of the point position value in the i-th row and j-th column of the improved convolution operator and the point position value at the corresponding position in the binarized feature map is greater than 0, record the first bit value of the point position value in the i-th row and j-th column of the improved convolution operator as the convolution feature value at this time. When the product of the second bit value of the point position value in the i-th row and j-th column of the improved convolution operator and the point position value at the corresponding position in the binarized feature map is less than or equal to 0, record the convolution feature value at this time as a null value.

[0175] Furthermore, according to the current point position value of the improved convolution operator and the point position value at the corresponding position in the binarized feature map, the steps of calculating the convolution feature value mainly include:

[0176] When the product of the second bit value of the current point position value of the improved convolution operator and the point position value at the corresponding position in the binarized feature map is greater than 0, use the first bit value of the current point position value of the improved convolution operator as the convolution feature value;

[0177] When the product of the second bit value of the current point position value of the improved convolution operator and the point position value at the corresponding position in the binarized feature map is less than or equal to 0, use a null value as the convolution feature value.

[0178] Further, in the angle feature operation layer calculate1 of any angle feature extraction module, the steps of calculating the initial angle feature value corresponding to each convolution feature value according to each convolution feature value mainly include:

[0179] Calculate the initial angle feature value corresponding to each convolution feature value according to the point position value included in the upper left corner positioning point of each convolution feature value and the point position value included in the lower right corner positioning point of each convolution feature value.

[0180] In this step, when the improved convolution operator performs a single convolution calculation on the upper left corner of the binarized feature map, the corresponding convolution feature value can be obtained. Then, using the photovoltaic module angle calculation formula, combined with the point position value included in the upper left corner positioning point of each convolution feature value and the point position value included in the lower right corner positioning point of each convolution feature value, calculate the initial angle feature value corresponding to each convolution feature value.

[0181] Optionally, the photovoltaic module angle calculation formula is as shown in the following formulas (3) to (6):

[0182] x a ,y a =i,j,Min(δ i，j [0]+δ i，j [1]), i, j = 1, 2, 3,..., 10 (3)

[0183] In the above formula (3), xa , y a represents the value contained in the upper-left corner positioning point of the convolution eigenvalue extracted when performing convolution using the improved convolution operator for the a-th time, δ i，j [0] represents the first value of the point position value at the i-th row and j-th column of the convolution eigenvalue extracted when performing convolution using the improved convolution operator for the a-th time, δ i，j [1] represents the second value of the point position value at the i-th row and j-th column of the convolution eigenvalue extracted when performing convolution using the improved convolution operator for the a-th time.

[0184] In the embodiment of the present invention, the bright spot existing in the upper left corner of the convolution eigenvalue can be extracted through formula (3). The bright spot is also called a positioning point, and the positioning point can be understood as a component of the photovoltaic panel. For example, referring to Figure 7 , the value x contained in the bright spot existing in the upper left corner of the convolution eigenvalue can be extracted through the operation of formula (3) a , y a = 0, 0.

[0185] X a , Y a = I, J, Max(δ I，J [0] + δ I，J [1]), I, J = 1, 2, 3,..., 10 (4)

[0186] In the above formula (4), X a , Y a represents the value contained in the lower-right corner positioning point of the convolution eigenvalue extracted when performing convolution using the improved convolution operator for the a-th time. δ I，J [0] represents the first value of the point position value at the I-th row and J-th column of the convolution eigenvalue extracted when performing convolution using the improved convolution operator for the a-th time, δ I，J [1] represents the second value of the point position value at the I-th row and J-th column of the convolution eigenvalue extracted when performing convolution using the improved convolution operator for the a-th time.

[0187] In the embodiment of the present invention, the bright spot existing in the lower right corner of the convolution eigenvalue can be extracted through formula (4). The bright spot is also called a positioning point, and the positioning point can be understood as a component of the photovoltaic panel. For example, referring to Figure 7 , the value X contained in the bright spot existing in the lower right corner of the convolution eigenvalue can be extracted through the operation of formula (4) a , Y a = 10, 7.

[0188] δ i，j , δ I，J ≠ Empty (5)

[0189] In the above formula (5), δi,j represents the position value at the i-th row and j-th column of the convolution eigenvalue extracted during convolution using the improved convolution operator, δ I，J represents the position value at the I-th row and J-th column of the convolution eigenvalue extracted during convolution using the improved convolution operator.

[0190] In the embodiments of the present invention, formula (5) represents not calculating the convolution eigenvalue with a null value.

[0191]

[0192] In the above formula (6), ω a represents the initial angle eigenvalue corresponding to the convolution eigenvalue extracted during the a-th convolution using the improved convolution operator.

[0193] In the embodiments of the present invention, the characteristic slope can be calculated by combining the value included in the lower right positioning point of the convolution eigenvalue and the value included in the upper left positioning point of the convolution eigenvalue through formula (6), and the characteristic slope is used as the initial angle eigenvalue.

[0194] Optionally, since the pixel size of the image input to the neural network model 100 is 648×648, the angle feature extraction layer con3 needs to perform 639×639 feature extraction on the binarized feature map, and the number of initial angle eigenvalues is 639×639. The value range of the parameter a in the foregoing formulas (3) to (4) and formula (6) can be a = 1, 2,..., 639×639.

[0195] Furthermore, in the angle output layer output1 of any angle feature extraction module, the embodiments of the present invention adopt the method of taking the average to integrate multiple initial angle eigenvalues to generate the final angle feature, as shown in the following formula (7):

[0196]

[0197] In the above formula (7), N represents the number of times of feature extraction by the angle feature extraction layer con3. In the embodiments of the present invention, N is defined as 639×639; ω avg-i represents the i-th angle feature output by the i-th angle feature extraction module in the angle feature extraction branch structure 120, where i = 1, 2, 3.

[0198] In some embodiments of the present invention, referring to Figure 3 , in the feature output layer 124 of the angle feature extraction branch structure 120, the implementation process of calculating the average angle feature according to the first angle feature, the second angle feature, and the third angle feature may include but is not limited to the following steps:

[0199] Obtain the valid pixel points in the binarized feature map, and calculate the average angle feature based on the number of valid pixel points in the first feature map, the second feature map, and the third feature map, in combination with the first angle feature, the second angle feature, and the third angle feature.

[0200] It should be noted that the pixel points with the pixel value being the first numerical value in the binarized feature map are used as the valid pixel points.

[0201] Optionally, the first numerical value is 1.

[0202] In the embodiment of the present invention, in the angle feature extraction branch structure 120, the first angle feature extraction module 121 located at the front of the neural network model 100, the second angle feature extraction module 122 located in the middle of the neural network model 100, and the third angle feature extraction module 123 located at the end of the neural network model 100 are used to extract the angle features, and then the first angle feature, the second angle feature, and the third angle feature can be obtained. Since there is a large difference in the number of valid pixel points of the first angle feature, the second angle feature, and the third angle feature, in the embodiment of the present invention, the feature output layer 124 uses the pixel threshold formula to perform an average calculation on the three angle features, and then obtains the average angle feature of the photovoltaic panel.

[0203] Optionally, the pixel threshold formula satisfies the following formula (8):

[0204]

[0205] In the above formula (8), ω last represents the average angle feature of the photovoltaic panel, ω avg-1 represents the first angle feature, ω <> avg-2 represents the second angle feature, ω avg-3 represents the third angle feature, n1 represents the number of valid pixel points of the first angle feature, n2 represents the number of valid pixel points of the second angle feature, n3 represents the number of valid pixel points of the third angle feature, M represents the total number of valid pixel points of the first angle feature, the second angle feature, and the third angle feature, and M = n1 + n2 + n3.

[0206] In some embodiments of the present invention, in step S103, calculating the average temperature feature of the photovoltaic panel based on the target detection frame and the temperature image of the photovoltaic panel may include but is not limited to the following steps:

[0207] S1031, crop the temperature image according to the coordinate information of the target detection frame in the angle image.

[0208] In this step, the coordinate information of the target detection box of the photovoltaic panel in the angular image is obtained, and the temperature image of the photovoltaic panel is directly cropped according to this coordinate information. Among them, the temperature image represents the temperatures at different positions in the form of pixel values.

[0209] S1032. Calculate the average temperature feature based on the number of pixel points and pixel values of the cropped temperature image.

[0210] In this step, for the cropped temperature image, the average temperature feature of the photovoltaic panel is calculated by taking the average value and combining the number of pixel points and pixel values of the cropped temperature image.

[0211] Optionally, the average temperature feature of the photovoltaic panel is calculated according to the number of pixel points and pixel values of the cropped temperature image through the following formula (9):

[0212]

[0213] In the above formula (9), Tem represents the average temperature feature of the photovoltaic panel, m represents the number of pixel points of the cropped temperature image, and pixel i represents the pixel value of the i-th pixel point of the cropped temperature image.

[0214] In some embodiments of the present invention, referring to Figure 8 , Figure 8 is the schematic diagram of calculating the optimal efficiency angle of the photovoltaic panel provided by the present invention. In step S104, according to the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel, calculating the optimal efficiency angle of the photovoltaic panel by using the machine learning method may include but is not limited to the following steps:

[0215] Calculate the optimal efficiency angle of the photovoltaic panel by using the SVM model according to the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel.

[0216] In this step, the machine learning model is pre-trained to obtain an SVM model for predicting the optimal efficiency angle of the photovoltaic panel. Then, the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel are used as the input of the SVM model, and the average angle feature, average temperature feature, absorption coefficient and solar radiation intensity of the photovoltaic panel are regressively fitted through the SVM model, and then the optimal efficiency angle of the photovoltaic panel is predicted, and the optimal efficiency angle of the photovoltaic panel satisfies the maximization of energy collection.

[0217] Optionally, the steps of training the machine learning model in the embodiments of the present invention to obtain an SVM model for predicting the optimal efficiency angle of the photovoltaic panel are as follows:

[0218] First, construct a machine learning model with SVM as the basic architecture, and obtain the sample key data of the sample solar radiation intensity and the photovoltaic panel, as well as the corresponding labeled results as the angle data set.

[0219] It should be noted that the sample key data of the photovoltaic panel includes the sample average angle feature, the sample average temperature feature, and the sample absorption coefficient of the photovoltaic panel. The labeled result corresponding to the sample solar radiation intensity and the sample key data of the photovoltaic panel is the sample optimal efficiency angle corresponding to the sample solar radiation intensity and the sample key data of the photovoltaic panel.

[0220] Then, divide the angle data set into a training set and a test set according to a preset ratio. The training set is used to train the machine learning model, and the test set is used to test the performance of the trained machine learning model.

[0221] Optionally, the preset ratio is 5:5.

[0222] After that, use the training set to train the machine learning model, and use the test set to perform a performance test on the trained machine learning model.

[0223] Specifically, use the sample solar radiation intensity and the sample key data of the photovoltaic panel as the input of the machine learning model, use the sample optimal efficiency angle as the learning target of the machine learning model, and use the training set to train the machine learning model, so that the machine learning model learns the correlation between the sample solar radiation intensity and the sample key data of the photovoltaic panel and the sample optimal efficiency angle, and then obtain the trained machine learning model. After that, use the test set to perform a performance test on the trained machine learning model to obtain the performance test result of the trained machine learning model.

[0224] Optionally, the hyperparameters of the machine learning model provided by the embodiments of the present invention include: using the mean squared error loss function as the loss function of the SVM model, and using the Gaussian kernel function as the kernel function of the SVM model. Those skilled in the art can understand that other kernel functions such as linear kernel function, polynomial kernel function, Laplace kernel function, Sigmoid kernel function, etc., and other loss functions such as logarithmic loss function, absolute value loss function, etc. are also applicable to the SVM model. The embodiments of the present invention do not make specific limitations on this.

[0225] Optionally, the performance test metrics of the machine learning model provided by the embodiments of the present invention may include but are not limited to error rate, accuracy, precision, recall, Receiver Operating Characteristic (ROC) curve, etc.

[0226] Finally, select the machine learning model whose performance test result meets the expected requirements as the output of the SVM model.

[0227] Optionally, the embodiments of the present invention use accuracy as a performance test index for the machine learning model. When the accuracy is greater than the required threshold, it is regarded that the performance test result of the machine learning model meets the expected requirements, and the machine learning model whose performance test result meets the expected requirements is selected as the output of the SVM model. Those skilled in the art can understand that other performance test indexes can be selected according to actual needs to test the performance of the machine learning model, and the expected requirements for the model performance can be established according to the selected performance test indexes, so as to realize the performance judgment of the machine learning model. The present invention does not make specific limitations on this.

[0228] Based on the above embodiments, referring to Figure 3 and Figure 9 , Figure 9 is the schematic diagram of the photovoltaic power supply optimization provided by the present invention. The implementation process and implementation principle of a method for optimizing the photovoltaic power supply of a computer room provided by the embodiments of the present invention will be described below.

[0229] The first step, data acquisition: When the drone reaches the acquisition position, the angle image of the photovoltaic panel is collected by the angle image acquisition unit, the temperature image of the photovoltaic panel under the current temperature condition is collected by the infrared image acquisition unit, the real-time light intensity of the photovoltaic panel under the current light condition is collected by the light intensity acquisition unit, and the solar radiation intensity under the current light condition is collected by the radiation intensity acquisition unit. Then, based on the real-time light intensity of the photovoltaic panel under the current light condition and the light intensity of the preset reference surface of the drone acquisition module, the absorption performance of the acquisition position of the photovoltaic panel relative to the preset reference surface of the drone acquisition module is calculated, that is, the absorption coefficient of the photovoltaic panel.

[0230] The second step, data preprocessing: The pixel sizes of the angle image and the temperature image of the photovoltaic panel are unified into the same pixel size, and the solar radiation intensity and the absorption coefficient of the photovoltaic panel are normalized. In particular, data augmentation is performed on the angle image of the photovoltaic panel.

[0231] The third step, extraction of the average angle feature and the average temperature feature of the photovoltaic panel:

[0232] For the extraction of the average angle feature of the photovoltaic panel, depthwise separable convolution processing and de-linearization processing are performed on the angle image through multiple feature map extraction modules of the target detection branch structure 110 in the trained neural network model 100, so as to realize the target detection of the photovoltaic panel. During the process of target detection of the angle image, the angle feature extraction branch structure 120 is used to extract and calculate the angle features of the output of each feature map extraction module respectively, so as to obtain the average angle feature of the photovoltaic panel.

[0233] For the extraction of the average temperature characteristics of the photovoltaic panel, the angle image is subjected to depthwise separable convolution processing and de-linearization processing through multiple feature map extraction modules of the object detection branch structure 110 in the trained neural network model 100 to obtain the object detection frame of the photovoltaic panel, and the average temperature characteristics of the photovoltaic panel are calculated in combination with the temperature image.

[0234] Step 4, generation of the optimal efficiency angle of the photovoltaic panel:

[0235] According to the average temperature characteristics, average angle characteristics, absorption coefficient and solar radiation intensity of the photovoltaic panel, the SVM model is used to fit the optimal efficiency angle of the photovoltaic panel to obtain the optimal efficiency angle of the photovoltaic panel.

[0236] Step 5, adjust the photovoltaic panel through the optimal efficiency angle of the photovoltaic panel.

[0237] In summary, a method for optimizing photovoltaic power supply in a computer room provided by an embodiment of the present invention, through the fusion of multi-dimensional data, comprehensively considers multiple factors such as light, panel angle and temperature, and can accurately determine the optimal efficiency angle of the photovoltaic panel without manual intervention, so that the photovoltaic panel can quickly adjust the working angle when the light intensity or temperature changes, and then enables the photovoltaic system to maintain the optimal efficiency output at each angle, effectively improving the energy collection efficiency and energy utilization rate of the photovoltaic system while reducing the burden and cost of manual maintenance, significantly reducing energy waste, and realizing sustainable power supply and maximum energy collection of the photovoltaic system.

[0238] In addition, referring to Figure 10 , Figure 10 is the structural diagram of a photovoltaic power supply optimization system in a computer room provided by the present invention. An embodiment of the present invention also provides a photovoltaic power supply optimization system in a computer room, including:

[0239] An acquisition module 201, configured to acquire the solar radiation intensity and the angle image, temperature image and absorption coefficient of the photovoltaic panel;

[0240] An angle feature processing module 202, configured to perform feature extraction and object detection on the angle image of the photovoltaic panel by using the trained neural network model 100 to obtain the object detection frame and average angle characteristics of the photovoltaic panel;

[0241] A temperature feature processing module 203, configured to calculate the average temperature characteristics of the photovoltaic panel according to the object detection frame and temperature image of the photovoltaic panel;

[0242] An angle calculation module 204, configured to calculate the optimal efficiency angle of the photovoltaic panel by using a machine learning method according to the average angle characteristics, average temperature characteristics, absorption coefficient and solar radiation intensity of the photovoltaic panel;

[0243] Adjustment module 205, configured to adjust the photovoltaic panel based on the optimal efficiency angle.

[0244] The content in the above method embodiments is applicable to the system embodiments of the present invention. The functions specifically implemented in the system embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0245] The embodiment of the present invention further provides a computer-readable storage medium, which stores a program executable by a processor. The program executable by the processor is used to execute an optimization method for photovoltaic power supply in a computer room described above when executed by the processor.

[0246] Similarly, the content in the above method embodiments is applicable to the storage medium embodiments of the present invention. The functions specifically implemented in the storage medium embodiments of the present invention are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0247] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and purposes of the present invention. The scope of the present invention is defined by the claims and their equivalents.

[0248] The above is a specific description of the preferred embodiments of the present invention, but the present invention is not limited to the described embodiments. Those skilled in the art can also make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A method for optimizing photovoltaic power supply in a computer room, characterized in that, It includes the following steps: Obtain the solar radiation intensity, as well as the angular image, temperature image, and absorption coefficient of the photovoltaic panel; Use the trained neural network model to perform feature extraction and object detection on the angular image of the photovoltaic panel to obtain the object detection frame and average angular feature of the photovoltaic panel; Calculate the average temperature feature of the photovoltaic panel based on the object detection frame and temperature image of the photovoltaic panel; Calculate the optimal efficiency angle of the photovoltaic panel by using machine learning based on the average angular feature, average temperature feature, absorption coefficient, and solar radiation intensity of the photovoltaic panel; Adjust the photovoltaic panel based on the optimal efficiency angle; Among them, the trained neural network model includes a first feature map extraction module, a second feature map extraction module, a third feature map extraction module, and a convolution module. The use of the trained neural network model to perform feature extraction and object detection on the angular image of the photovoltaic panel includes: Perform depthwise separable convolution processing and de-linearization processing on the angular image through the first feature map extraction module to obtain a first feature map; Perform depthwise separable convolution processing and de-linearization processing on the first feature map through the second feature map extraction module to obtain a second feature map; Perform depthwise separable convolution processing and de-linearization processing on the second feature map through the third feature map extraction module to obtain a third feature map; Perform convolution operation on the third feature map through the convolution module to generate an object detection frame; Among them, the convolution module includes at least five convolutional layers; Among them, the trained neural network model further includes a first angular feature extraction module, a second angular feature extraction module, a third angular feature extraction module, and a feature output layer. The use of the trained neural network model to perform feature extraction and object detection on the angular image of the photovoltaic panel further includes: Extract the angular feature from the first feature map through the first angular feature extraction module to obtain a first angular feature; Extract the angular feature from the second feature map through the second angular feature extraction module to obtain a second angular feature; Extract the angular feature from the third feature map through the third angular feature extraction module to obtain a third angular feature; Calculate the average angular feature through the feature output layer based on the first angular feature, the second angular feature, and the third angular feature.

2. The optimization method for photovoltaic power supply in a computer room according to claim 1, wherein The first feature map extraction module includes one depthwise separable convolutional layer, and the second feature map extraction module and the third feature map extraction module both include two depthwise separable convolutional layers connected in sequence, and a ReLU activation function is provided behind each depthwise separable convolutional layer.

3. The optimization method for photovoltaic power supply in a computer room according to claim 1, characterized in that, The first angular feature extraction module, the second angular feature extraction module, and the third angular feature extraction module all include a binarization layer, an angular feature extraction layer, an angular feature operation layer, and an angular output layer connected in sequence. Among them, The binarization layer is used to perform binarization processing on the feature map input to the first angular feature extraction module, the second angular feature extraction module, or the third angular feature extraction module to obtain a binarized feature map; The angle feature extraction layer is used to perform multiple feature extractions on the binarized feature map using an improved convolution operator to obtain multiple convolution feature values; The angle feature operation layer is used to calculate an initial angle feature value corresponding to each convolution feature value according to each convolution feature value; The angle output layer is used to calculate the first angle feature, the second angle feature or the third angle feature according to multiple initial angle feature values and the size value of the binarized feature map.

4. The optimization method for photovoltaic power supply in a computer room according to claim 3, wherein The step of calculating an initial angle feature value corresponding to each convolution feature value according to each convolution feature value includes: calculating an initial angle feature value corresponding to each convolution feature value according to the point values included in the upper left corner positioning point of each convolution feature value and the point values included in the lower right corner positioning point of each convolution feature value.

5. The optimization method for photovoltaic power supply in a computer room according to claim 3, characterized in that The calculating, by the feature output layer, an average angle feature based on the first angle feature, the second angle feature and the third angle feature includes: obtaining valid pixel points in the binarized feature map, and calculating an average angle feature according to the number of valid pixel points in the first feature map, the second feature map and the third feature map, in combination with the first angle feature, the second angle feature and the third angle feature.

6. The optimization method for photovoltaic power supply in a computer room according to claim 1, characterized in that The calculating the average temperature feature of the photovoltaic panel according to the target detection frame and the temperature image of the photovoltaic panel includes: cropping the temperature image according to the coordinate information of the target detection frame in the angle image; calculating an average temperature feature according to the number and pixel values of the pixel points of the cropped temperature image.

7. An optimized system for photovoltaic power supply in a computer room, characterized in that, including: an acquisition module, configured to acquire the solar radiation intensity, as well as the angle image, the temperature image and the absorption coefficient of the photovoltaic panel; an angle feature processing module, configured to perform feature extraction and target detection on the angle image of the photovoltaic panel by using a trained neural network model to obtain a target detection frame and an average angle feature of the photovoltaic panel; a temperature feature processing module, configured to calculate an average temperature feature of the photovoltaic panel according to the target detection frame and the temperature image of the photovoltaic panel; an angle calculation module, configured to calculate the optimal efficiency angle of the photovoltaic panel by using a machine learning method according to the average angle feature, the average temperature feature and the absorption coefficient of the photovoltaic panel, as well as the solar radiation intensity; an adjustment module, configured to adjust the photovoltaic panel based on the optimal efficiency angle; [[ID=~14]]wherein, the trained neural network model includes a first feature map extraction module, a second feature map extraction module, a third feature map extraction module and a convolution module, and the performing feature extraction and target detection on the angle image of the photovoltaic panel by using the trained neural network model includes: performing depthwise separable convolution processing and de-linearization processing on the angle image through the first feature map extraction module to obtain a first feature map; performing depthwise separable convolution processing and de-linearization processing on the first feature map through the second feature map extraction module to obtain a second feature map; performing depthwise separable convolution processing and de-linearization processing on the second feature map through the third feature map extraction module to obtain a third feature map; Performing a convolution operation on the third feature map through the convolution module to generate a target detection box; wherein the convolution module includes at least five convolutional layers; wherein the trained neural network model further includes a first angle feature extraction module, a second angle feature extraction module, a third angle feature extraction module, and a feature output layer, and the use of the trained neural network model to perform feature extraction and target detection on the angle image of the photovoltaic panel further includes: Extracting angle features from the first feature map through the first angle feature extraction module to obtain first angle features; Extracting angle features from the second feature map through the second angle feature extraction module to obtain second angle features; Extracting angle features from the third feature map through the third angle feature extraction module to obtain third angle features; Calculating an average angle feature through the feature output layer based on the first angle feature, the second angle feature, and the third angle feature.

8. A computer-readable storage medium storing a program executable by a processor, characterized in that, The program executable by the processor, when executed by the processor, is used to implement a method for optimizing photovoltaic power supply in a computer room as described in any one of claims 1 to 6.

Images