Mountain photovoltaic power generation data real-time processing method based on edge calculation

Abstract

The invention discloses a mountain photovoltaic power generation data real-time processing method based on edge calculation, belongs to the technical field of data processing, and adopts preprocessed historical mountain photovoltaic power generation data and data classification labels corresponding to the historical mountain photovoltaic power generation data to train a lightweight deep learning model. Acquiring a photovoltaic power generation anomaly detection model, deploying the photovoltaic power generation anomaly detection model to each edge calculation node, establishing connection between the mountain photovoltaic array and the edge calculation nodes, and collecting mountain photovoltaic power generation data of the mountain photovoltaic array according to a first preset data sampling frequency through the edge calculation nodes, according to the method, the mountain photovoltaic power generation data are acquired, and the photovoltaic power generation anomaly detection model is scheduled to identify the mountain photovoltaic power generation data and determine the photovoltaic power generation anomaly detection result, so that the real-time processing efficiency of the mountain photovoltaic power generation data is effectively improved, the monitoring of mountain photovoltaic power generation is more timely and accurate, and the normal operation of mountain photovoltaic power generation is ensured.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and particularly relates to a method for real-time processing of mountain photovoltaic power generation data based on edge computing. Background Art

[0002] With the transformation of the global energy structure, photovoltaic power generation, as an important part of clean energy, has been increasingly emphasized in its development. Mountain photovoltaic power generation is a way of utilizing clean energy that uses the mountain terrain to build a photovoltaic power station and convert solar energy into electrical energy. Its advantages lie in making full use of the mountain space without occupying arable land resources, and at the same time, the mountainous area has sufficient sunlight, which is conducive to improving the power generation efficiency. However, due to the complex terrain and changing environment of mountain photovoltaic power stations, higher requirements are put forward for the construction and maintenance of the power stations. In particular, the real-time processing and monitoring of power generation data have become the key to ensuring the stable operation of the power stations. However, traditional data processing methods often rely on a central server, suffering from problems such as data transmission delay, bandwidth limitation, and untimely monitoring of mountain photovoltaic power generation, making it difficult to meet the real-time requirements. Summary of the Invention

[0003] The present invention provides a method for real-time processing of mountain photovoltaic power generation data based on edge computing to solve the problems in the prior art that rely on a central server, including data transmission delay, bandwidth limitation, and untimely monitoring of mountain photovoltaic power generation.

[0004] A method for real-time processing of mountain photovoltaic power generation data based on edge computing includes:

[0005] Collecting historical mountain photovoltaic power generation data under various fault states and preprocessing the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data;

[0006] Constructing a classification label for each fault state corresponding to the historical mountain photovoltaic power generation data to obtain the data classification labels corresponding to the historical mountain photovoltaic power generation data;

[0007] Constructing a lightweight deep learning model and training the lightweight deep learning model with the preprocessed historical mountain photovoltaic power generation data and the data classification labels corresponding to the historical mountain photovoltaic power generation data to obtain a photovoltaic power generation anomaly detection model;

[0008] Deploying the photovoltaic power generation anomaly detection model to each edge computing node and establishing a connection between the mountain photovoltaic array and the edge computing node so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array;

[0009] The edge computing node collects the mountain photovoltaic power generation data of the mountain photovoltaic array according to the first preset data sampling frequency, and schedules the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data to determine the photovoltaic power generation anomaly detection result;

[0010] The edge computing node transmits the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server, and when the photovoltaic power generation anomaly detection result meets the preset condition, adjusts the first preset data sampling frequency to a larger second preset data sampling frequency to strengthen the data monitoring and processing degree.

[0011] Further, historical mountain photovoltaic power generation data in various fault states is collected, and the historical mountain photovoltaic power generation data is preprocessed to obtain the preprocessed historical mountain photovoltaic power generation data, including:

[0012] Collect the operating voltage data corresponding to the photovoltaic inverters in the mountain photovoltaic array in various fault states to obtain the first mountain photovoltaic power generation data; wherein, the first mountain photovoltaic power generation data is used to characterize the internal operating characteristics of the mountain photovoltaic array;

[0013] Collect the infrared image data corresponding to the mountain photovoltaic array in various fault states to obtain the second mountain photovoltaic power generation data; wherein, the second mountain photovoltaic power generation data is used to characterize the surface operating characteristics of the mountain photovoltaic array;

[0014] Take the first mountain photovoltaic power generation data and the second mountain photovoltaic power generation data together as the historical mountain photovoltaic power generation data, and preprocess the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data.

[0015] Further, preprocessing the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data, including:

[0016] Perform image enhancement processing on the historical mountain photovoltaic power generation data to obtain the historical mountain photovoltaic power generation data after image enhancement processing; wherein, only the second mountain photovoltaic power generation data is subjected to image enhancement processing;

[0017] Perform sample equalization processing on the historical mountain photovoltaic power generation data after image enhancement processing to obtain the historical mountain photovoltaic power generation data after sample equalization processing;

[0018] Perform image diversification processing on the historical mountain photovoltaic power generation data after sample equalization processing to obtain the preprocessed historical mountain photovoltaic power generation data.

[0019] Further, a lightweight deep learning model is constructed, and the preprocessed historical mountain photovoltaic power generation data and the corresponding data classification labels of the historical mountain photovoltaic power generation data are used to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model, including:

[0020] Construct a CNN-CNN-BP model to obtain a lightweight deep learning model;

[0021] Use the preprocessed historical mountain photovoltaic power generation data as the input of the lightweight deep learning model, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output of the lightweight deep learning model, and use a hyperparameter joint optimization algorithm to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model.

[0022] Further, use a hyperparameter joint optimization algorithm to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model, including:

[0023] For the first CNN model in the lightweight deep learning model, use the first mountain photovoltaic power generation data in the preprocessed historical mountain photovoltaic power generation data as the input, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use an intelligent optimization algorithm to train the first CNN model to obtain the trained first CNN model;

[0024] For the second CNN model in the lightweight deep learning model, use the second mountain photovoltaic power generation data in the preprocessed historical mountain photovoltaic power generation data as the input, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use an intelligent optimization algorithm to train the second CNN model to obtain the trained second CNN model;

[0025] Remove the classification layers of the first CNN model and the second CNN model to obtain a first feature extraction model and a second feature extraction model;

[0026] For the BP neural network model in the lightweight deep learning model, extract the features of the first mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data through the first feature extraction model to obtain the first feature; extract the features of the second mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data through the second feature extraction model to obtain the second feature; after fusing the first feature and the second feature, obtain a fused feature; use the fused feature as the input of the BP neural network model, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use an intelligent optimization algorithm to train the BP neural network model to obtain the trained BP neural network model to obtain a feature recognition model;

[0027] Construct a feature fusion layer, connect the outputs of the first feature extraction model and the second feature extraction model to the input of the feature fusion layer, and connect the output of the feature fusion layer to the input of the feature recognition model to obtain a photovoltaic power generation anomaly detection model; wherein, the feature fusion layer is used to fuse the first feature and the second feature.

[0028] Further, deploy the photovoltaic power generation anomaly detection model to each edge computing node, and establish a connection between the mountain photovoltaic array and the edge computing node so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array, including:

[0029] Deploy the photovoltaic power generation anomaly detection model to each edge computing node;

[0030] For any mountain photovoltaic array, determine the edge computing node closest to the mountain photovoltaic array to obtain the target edge computing node corresponding to each mountain photovoltaic array;

[0031] Associate the mountain photovoltaic array with the target edge computing node to obtain the mountain photovoltaic array corresponding to each target edge computing node;

[0032] For any target edge computing node, determine the load overrun situation of the target edge computing node; wherein, the load overrun situation includes that the number of connected mountain photovoltaic arrays exceeds a preset threshold or the number of connected mountain photovoltaic arrays does not exceed a preset threshold;

[0033] When there is a load overrun situation corresponding to the target edge computing node that the number of connected mountain photovoltaic arrays exceeds the preset threshold, perform load reallocation and establish a connection between the mountain photovoltaic array and the edge computing node so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array;

[0034] When the load overrun situations corresponding to all target edge computing nodes are that the number of connected mountain photovoltaic arrays does not exceed the preset threshold, establish a connection between the mountain photovoltaic array and the edge computing node according to the mountain photovoltaic array corresponding to the target edge computing node so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array.

[0035] Further, when there is a load overrun situation corresponding to the target edge computing node that the number of connected mountain photovoltaic arrays does not exceed the preset threshold, perform load reallocation and establish a connection between the mountain photovoltaic array and the edge computing node, including:

[0036] Put all target edge computing nodes into the pending loading list;

[0037] Retrieve the target edge computing nodes with a load overrun situation where the number of connected mountain photovoltaic arrays exceeds a preset threshold, and obtain the target edge computing nodes to be balanced;

[0038] For the target edge computing nodes to be balanced, obtain the mountain photovoltaic array with the farthest distance among the target edge computing nodes to be balanced, and obtain the mountain photovoltaic array to be separated;

[0039] For the target edge computing nodes to be balanced, repeatedly obtain the mountain photovoltaic array to be separated until the load overrun situation of the target edge computing nodes to be balanced is that the number of connected mountain photovoltaic arrays does not exceed the preset threshold, and obtain at least one mountain photovoltaic array to be separated;

[0040] Determine the target edge computing node closest to the mountain photovoltaic array to be separated in the list to be loaded, obtain the target edge computing node to be allocated, and allocate the mountain photovoltaic array to be separated to the target edge computing node to be allocated;

[0041] Judge whether there are target edge computing nodes with the number of connected mountain photovoltaic arrays exceeding the preset threshold in the list to be loaded. If so, return to the step of obtaining the target edge computing nodes to be balanced. Otherwise, complete the load reallocation process and establish the connection between the mountain photovoltaic array and the edge computing node.

[0042] Furthermore, the edge computing node collects the mountain photovoltaic power generation data of the mountain photovoltaic array according to the first preset data sampling frequency, and schedules the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data, and determines the photovoltaic power generation anomaly detection result, including:

[0043] The edge computing node collects the mountain photovoltaic power generation data of the mountain photovoltaic array according to the first preset data sampling frequency, and obtains the first mountain photovoltaic power generation data to be identified and the second mountain photovoltaic power generation data to be identified;

[0044] Use the first mountain photovoltaic power generation data to be identified as the input of the first feature extraction model in the photovoltaic power generation anomaly detection model, and use the second mountain photovoltaic power generation data to be identified as the input of the first feature extraction model, and obtain the output of the feature recognition model in the photovoltaic power generation anomaly detection model to obtain the photovoltaic power generation anomaly detection result.

[0045] Furthermore, the edge computing node transmits the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server, and when the photovoltaic power generation anomaly detection result meets the preset conditions, adjusts the first preset data sampling frequency to a larger second preset data sampling frequency, including:

[0046] Through data encryption technology, the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection results are transmitted to the central server through the edge computing nodes, so that the central server can uniformly manage the mountain photovoltaic power generation data;

[0047] Judge whether the photovoltaic power generation anomaly detection result is a non-normal category. If so, it is determined that the preset conditions are met; otherwise, it is determined that the preset conditions are not met;

[0048] When the photovoltaic power generation anomaly detection result meets the preset conditions, adjust the data sampling frequency to the second preset data sampling frequency, and continue to perform real-time processing on the mountain photovoltaic power generation data.

[0049] Furthermore, it also includes: transmitting the photovoltaic power generation anomaly detection result to the device designated by the staff.

[0050] A real-time processing method for mountain photovoltaic power generation data based on edge computing provided by the present invention uses the historical mountain photovoltaic power generation data after preprocessing and the corresponding data classification labels of the historical mountain photovoltaic power generation data to train a lightweight deep learning model, obtain a photovoltaic power generation anomaly detection model, and deploy the photovoltaic power generation anomaly detection model to each edge computing node, establish a connection between the mountain photovoltaic array and the edge computing node, collect the mountain photovoltaic power generation data of the mountain photovoltaic array by the edge computing node according to the first preset data sampling frequency, and schedule the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data, determine the photovoltaic power generation anomaly detection result, effectively improving the real-time processing efficiency of the mountain photovoltaic power generation data, making the monitoring of mountain photovoltaic power generation more timely and accurate, and ensuring the normal operation of mountain photovoltaic power generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings here are incorporated into the specification and form a part of this specification, showing the embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0052] Figure 1 It is a flowchart of a real-time processing method for mountain photovoltaic power generation data based on edge computing provided by an embodiment of the present invention.

[0053] Through the above accompanying drawings, the clear embodiments of the present invention have been shown, and there will be more detailed descriptions later. These drawings and the text description are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of apparatuses and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0055] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0056] As Figure 1 shown, an embodiment of the present invention provides a method for real-time processing of mountain photovoltaic power generation data based on edge computing, including:

[0057] S1. Collect historical mountain photovoltaic power generation data under various fault states, and preprocess the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data;

[0058] The historical mountain photovoltaic power generation data may include internal data and external data of the mountain photovoltaic power generation array, so as to achieve a better recognition effect and improve the monitoring accuracy of mountain photovoltaic power generation. Then, preprocess the historical mountain photovoltaic power generation data to make the data easier to learn, thereby indirectly improving the subsequent data processing accuracy.

[0059] S2. Construct a classification label for each fault state corresponding to the historical mountain photovoltaic power generation data to obtain the data classification labels corresponding to the historical mountain photovoltaic power generation data;

[0060] Each data classification label represents a fault state. In addition, a classification label corresponding to the normal state needs to be added, so as to identify whether the mountain photovoltaic power generation array is in an abnormal state according to the mountain photovoltaic power generation data.

[0061] S3. Construct a lightweight deep learning model, and train the lightweight deep learning model using the preprocessed historical mountain photovoltaic power generation data and the data classification labels corresponding to the historical mountain photovoltaic power generation data to obtain a photovoltaic power generation anomaly detection model;

[0062] The lightweight deep learning model can adopt some conventional deep learning models to learn the historical mountain photovoltaic power generation data. The lightweight deep learning model can be trained using intelligent optimization algorithms (such as particle swarm optimization algorithm or gradient descent optimization algorithm) to obtain a photovoltaic power generation anomaly detection model.

[0063] S4. Deploy the photovoltaic power generation anomaly detection model to each edge computing node, and establish a connection between the mountain photovoltaic array and the edge computing node, so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array;

[0064] The main purpose of establishing the connection between the mountain photovoltaic array and the edge computing node is to connect the sensors that collect the mountain photovoltaic power generation data, so as to facilitate data acquisition and real-time data processing.

[0065] S5. The edge computing node collects the mountain photovoltaic power generation data of the mountain photovoltaic array according to the first preset data sampling frequency, and schedules the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data to determine the photovoltaic power generation anomaly detection result;

[0066] The first preset data sampling frequency is the data sampling frequency under the normal operation state. The larger the data sampling frequency, the better the monitoring effect, but it will consume a large amount of network bandwidth; while under the normal operation state, a too large data sampling frequency is not required. Therefore, the first preset data sampling frequency can be set to some smaller frequencies, or can be set according to the actual needs of the staff.

[0067] S6. The edge computing node transmits the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server, and when the photovoltaic power generation anomaly detection result meets the preset conditions, adjusts the first preset data sampling frequency to a larger second preset data sampling frequency to strengthen the data monitoring and processing degree.

[0068] When a fault occurs, the monitoring of the mountain photovoltaic power generation array should be strengthened. Therefore, the data sampling frequency can be adjusted to a larger second preset data sampling frequency to enhance the monitoring effect. Therefore, the second preset data sampling frequency can be set to twice or more times the first preset data sampling frequency.

[0069] A real-time processing method for mountain photovoltaic power generation data based on edge computing provided by an embodiment of the present invention utilizes edge processing technology and deep learning technology, greatly improving the processing efficiency of mountain photovoltaic power generation data and reducing the requirement for network bandwidth.

[0070] In the embodiment of the present invention, historical mountain photovoltaic power generation data in various fault states is collected, and the historical mountain photovoltaic power generation data is preprocessed to obtain the preprocessed historical mountain photovoltaic power generation data, including:

[0071] Collect the operating voltage data corresponding to the photovoltaic inverter in the mountain photovoltaic array in various fault states to obtain the first mountain photovoltaic power generation data; wherein, the first mountain photovoltaic power generation data is used to characterize the internal operating characteristics of the mountain photovoltaic array;

[0072] For example, during the normal operation of a photovoltaic power generation system, it is described that the IGBTs in the photovoltaic inverter conduct according to the bridge inverter rule, and the DC-side current is a stable direct current, and the output terminal is three-phase voltage and current. When an open-circuit fault occurs in the inverter, one or more IGBTs (Insulated Gate Bipolar Transistors) are open-circuited. To simulate the open-circuit fault of IGBTs in a photovoltaic inverter system, the conduction of IGBTs can be controlled through a control circuit to simulate the working state of IGBTs in the inverter circuit. Assuming that the sampling frequency is set to 5000 Hz, and then by controlling the gate voltage of each IGBT, conduction and opening / closing are achieved, and sensors are used to sample at equal time intervals to obtain DC-side current signals under various different conditions, that is, the first mountain photovoltaic power generation data.

[0073] Collect the infrared image data corresponding to the mountain photovoltaic array under various fault states to obtain the second mountain photovoltaic power generation data; among them, the second mountain photovoltaic power generation data is used to characterize the surface operation characteristics of the mountain photovoltaic array;

[0074] Many photovoltaic power stations focus on the faults that occur in photovoltaic panels. This is because the occurrence of faults will lead to consequences such as local short circuits or fires, causing great losses to the entire power system. These faults can generally be divided into physical faults, environmental faults, and electrical faults, etc. Physical faults are usually caused by the surface cracking or damage of photovoltaic panels. Photovoltaic panels are often installed in suburban areas, desert areas, or places with extremely harsh environments. After long-term exposure to wind, sun, weather with large temperature differences, rainstorms, or heavy snow, and then being directly irradiated by sunlight, it is extremely easy to cause the surface of the photovoltaic panel to crack. Over time, it will cause serious consequences such as water ingress inside the photovoltaic panel, line damage, reduced output power at the cracked part, circuit short circuit, and even fire. Electrical faults are generally caused by short circuits or open circuits in photovoltaic panels. Once a short circuit or open circuit occurs in a photovoltaic panel, it will lead to local high temperature and inability to dissipate heat, thus generating hot spots. If not dealt with in time, the hot spots will gradually become larger and burn the entire photovoltaic panel. If the short circuit fault is serious, the internal components of the photovoltaic panel are damaged, resulting in overheating of the photovoltaic panel. If not repaired in time, it will cause a short circuit in the entire photovoltaic system, leading to the paralysis of the power supply system.

[0075] Therefore, collecting the second mountain photovoltaic power generation data can not only identify external faults but also assist in identifying internal faults, further improving the accuracy of fault identification. Similarly, the first mountain photovoltaic power generation data can assist in identifying external faults. The combined identification of the two can greatly improve the data processing efficiency and accuracy, realizing strong monitoring of the mountain photovoltaic power generation array.

[0076] In an embodiment of the present invention, external infrared image data and internal operating voltage data are used for joint recognition, which can identify various faults and effectively improve the accuracy of data processing.

[0077] The first mountain photovoltaic power generation data and the second mountain photovoltaic power generation data are jointly used as historical mountain photovoltaic power generation data, and the historical mountain photovoltaic power generation data is preprocessed to obtain the preprocessed historical mountain photovoltaic power generation data.

[0078] In an embodiment of the present invention, preprocessing the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data includes:

[0079] Performing image enhancement processing on the historical mountain photovoltaic power generation data to obtain the historical mountain photovoltaic power generation data after image enhancement processing; among them, only the second mountain photovoltaic power generation data is subjected to image enhancement processing;

[0080] In the actual process of image generation and transmission, due to problems with the acquisition environment or acquisition equipment, the collected images may have random noise, which will cause the images to become blurred during imaging and the overall quality to decline. This random noise comes from various aspects, such as the noise of the camera sensor, the influence of environmental light, and the interference during signal transmission. These noises will introduce additional detail changes in the image, making the image blurred or distorted. To cope with the influence of these random noises, it is usually necessary to adopt some image enhancement techniques to increase the useful information in the image and filter out redundant information, so that the blurred parts in the photo become clearer or more obvious picture features are formed. Currently, common image enhancement techniques include: spatial domain method and frequency domain method.

[0081] Performing sample equalization processing on the historical mountain photovoltaic power generation data after image enhancement processing to obtain the historical mountain photovoltaic power generation data after sample equalization processing;

[0082] During the actual sample collection process, it is impossible for all fault samples to be the same in number, resulting in an imbalance between the numbers of different types of samples. Therefore, the fault detection model will tend to learn the features of other types, and has weak recognition and classification capabilities for these fault samples with a small number, resulting in an overfitting phenomenon, causing the model to be unable to effectively identify and process rare fault types in photovoltaic panel fault detection, thus affecting the accuracy and generalization of model detection.

[0083] Therefore, it is possible to copy the sample data with a small number or perform operations to artificially add samples to achieve various sample equalizations. Or delete some of the sample data with a large number to achieve equalization processing.

[0084] Perform image diversification processing on the historical mountain photovoltaic power generation data after sample equalization processing to obtain the historical mountain photovoltaic power generation data after preprocessing.

[0085] The sample data can be rotated to simulate data acquisition under different angles. Since the first mountain photovoltaic power generation data is voltage data, that is, voltage waveform data, this step does not process the first mountain photovoltaic power generation data and only processes the second mountain photovoltaic power generation data.

[0086] In the embodiment of the present invention, a lightweight deep learning model is constructed, and the lightweight deep learning model is trained using the historical mountain photovoltaic power generation data after preprocessing and the corresponding data classification labels of the historical mountain photovoltaic power generation data to obtain a photovoltaic power generation anomaly detection model, including:

[0087] Construct a CNN-CNN (Convolutional Neural Network)-BP (Back Propagation Neural Network) model to obtain a lightweight deep learning model;

[0088] Use the historical mountain photovoltaic power generation data after preprocessing as the input of the lightweight deep learning model, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output of the lightweight deep learning model, and use the hyperparameter joint optimization algorithm to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model.

[0089] The CNN-CNN-BP model provided by the embodiment of the present invention combines the Convolutional Neural Network (CNN) and the Back Propagation (BP) algorithm, showing excellent performance. Its automatic feature extraction ability greatly reduces the burden of manual feature engineering. The local connection and parameter sharing design significantly reduces the model complexity, improves the calculation efficiency, and reduces the risk of overfitting at the same time. Hierarchical feature learning enables the model to capture features at all levels from simple to complex. The translational invariance enhances the generalization ability of the model. The backpropagation mechanism continuously improves the accuracy of the model parameters through gradient descent optimization.

[0090] In the embodiment of the present invention, the hyperparameter joint optimization algorithm is used to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model, including:

[0091] For the first CNN model in the lightweight deep learning model, use the first mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data after preprocessing as the input, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use the intelligent optimization algorithm to train the first CNN model to obtain the trained first CNN model;

[0092] For the second CNN model in the lightweight deep learning model, using the second mountain photovoltaic power generation data in the preprocessed historical mountain photovoltaic power generation data as input and the corresponding data classification label of the historical mountain photovoltaic power generation data as the expected output, an intelligent optimization algorithm is used to train the second CNN model to obtain the trained second CNN model;

[0093] Remove the classification layers of the first CNN model and the second CNN model to obtain the first feature extraction model and the second feature extraction model;

[0094] For the BP neural network model in the lightweight deep learning model, extract the features of the first mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data through the first feature extraction model to obtain the first feature; extract the features of the second mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data through the second feature extraction model to obtain the second feature; after fusing the first feature and the second feature, obtain the fused feature (for example, sequentially splice the first feature and the second feature into a vector); use the fused feature as the input of the BP neural network model and the corresponding data classification label of the historical mountain photovoltaic power generation data as the expected output, and use an intelligent optimization algorithm to train the BP neural network model to obtain the trained BP neural network model, thus obtaining the feature recognition model;

[0095] Construct a feature fusion layer, connect the outputs of the first feature extraction model and the second feature extraction model to the input of the feature fusion layer, and connect the output of the feature fusion layer to the input of the feature recognition model to obtain the photovoltaic power generation anomaly detection model; among them, the feature fusion layer is used to fuse the first feature and the second feature.

[0096] The joint recognition method provided by the embodiments of the present invention can effectively perform joint recognition on the internal data and external data of the mountain photovoltaic power generation array, which helps to improve the monitoring accuracy of the mountain photovoltaic power generation array.

[0097] In the embodiments of the present invention, deploying the photovoltaic power generation anomaly detection model to each edge computing node and establishing a connection between the mountain photovoltaic array and the edge computing node so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array includes:

[0098] Deploy the photovoltaic power generation anomaly detection model to each edge computing node;

[0099] For any mountain photovoltaic array, determine the edge computing node closest to the mountain photovoltaic array to obtain the target edge computing node corresponding to each mountain photovoltaic array;

[0100] Associate the mountain photovoltaic array with the target edge computing nodes to obtain the mountain photovoltaic array corresponding to each target edge computing node;

[0101] For any one of the target edge computing nodes, determine the overload situation of the target edge computing node; wherein, the overload situation includes that the number of connected mountain photovoltaic arrays exceeds a preset threshold or the number of connected mountain photovoltaic arrays does not exceed a preset threshold;

[0102] When there is an overload situation corresponding to a target edge computing node that the number of connected mountain photovoltaic arrays exceeds a preset threshold, perform load reallocation and establish a connection between the mountain photovoltaic array and the edge computing node, so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array;

[0103] When the overload situation corresponding to all target edge computing nodes is that the number of connected mountain photovoltaic arrays does not exceed a preset threshold, establish a connection between the mountain photovoltaic array and the edge computing node according to the mountain photovoltaic array corresponding to the target edge computing node, so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array.

[0104] In the embodiment of the present invention, when there is an overload situation corresponding to a target edge computing node that the number of connected mountain photovoltaic arrays does not exceed a preset threshold, perform load reallocation and establish a connection between the mountain photovoltaic array and the edge computing node, including:

[0105] Put all target edge computing nodes into the waiting-to-be-loaded list;

[0106] Take out the target edge computing node with an overload situation that the number of connected mountain photovoltaic arrays exceeds a preset threshold to obtain the target edge computing node to be balanced;

[0107] For the target edge computing node to be balanced, obtain the mountain photovoltaic array that is the farthest away in the target edge computing node to be balanced to obtain the mountain photovoltaic array to be separated;

[0108] For the target edge computing node to be balanced, repeatedly obtain the mountain photovoltaic array to be separated until the overload situation of the target edge computing node to be balanced is that the number of connected mountain photovoltaic arrays does not exceed a preset threshold, and obtain at least one mountain photovoltaic array to be separated;

[0109] Determine the target edge computing node that is the closest to the mountain photovoltaic array to be separated in the waiting-to-be-loaded list to obtain the target edge computing node to be allocated, and allocate the mountain photovoltaic array to be separated to the target edge computing node to be allocated;

[0110] Determine whether there is a target edge computing node with the number of connected mountain photovoltaic arrays in the list to be loaded exceeding a preset threshold. If so, return the step of obtaining the target edge computing node to be balanced. Otherwise, complete the load reallocation process and establish a connection between the mountain photovoltaic array and the edge computing node.

[0111] The edge computing method provided by the embodiments of the present invention can not only effectively improve the data processing efficiency of the mountain photovoltaic array, but also ensure load balance, guarantee the processing efficiency of edge computing, avoid overload situations, and enable the mountain photovoltaic operation data at any time to be processed.

[0112] In the embodiments of the present invention, the edge computing node collects the mountain photovoltaic power generation data of the mountain photovoltaic array according to the first preset data sampling frequency, and schedules the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data to determine the photovoltaic power generation anomaly detection result, including:

[0113] The edge computing node collects the mountain photovoltaic power generation data of the mountain photovoltaic array according to the first preset data sampling frequency to obtain the first to-be-identified mountain photovoltaic power generation data and the second to-be-identified mountain photovoltaic power generation data;

[0114] Use the first to-be-identified mountain photovoltaic power generation data as the input of the first feature extraction model in the photovoltaic power generation anomaly detection model, and use the second to-be-identified mountain photovoltaic power generation data as the input of the first feature extraction model to obtain the output of the feature recognition model in the photovoltaic power generation anomaly detection model, and obtain the photovoltaic power generation anomaly detection result.

[0115] In the embodiments of the present invention, the edge computing node transmits the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server, and when the photovoltaic power generation anomaly detection result meets the preset conditions, adjusts the first preset data sampling frequency to a larger second preset data sampling frequency, including:

[0116] Use data encryption technology to transmit the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server through the edge computing node, so that the central server can uniformly manage the mountain photovoltaic power generation data;

[0117] Judge whether the photovoltaic power generation anomaly detection result is a non-normal category. If so, it is determined that the preset conditions are met. Otherwise, it is determined that the preset conditions are not met;

[0118] When the photovoltaic power generation anomaly detection result meets the preset conditions, adjust the data sampling frequency to the second preset data sampling frequency and continue to perform real-time processing on the mountain photovoltaic power generation data.

[0119] Optionally, when the photovoltaic power generation anomaly detection result meets a preset condition, an anomaly alarm may be generated and transmitted to the central server, so that the staff can view data, receive data, and view anomaly alarms through the software system.

[0120] In the embodiment of the present invention, it further includes: transmitting the photovoltaic power generation anomaly detection result to the device designated by the staff, so that the staff can timely understand the anomaly.

[0121] A real-time processing method for mountain photovoltaic power generation data based on edge computing provided by the present invention uses historical mountain photovoltaic power generation data after preprocessing and corresponding data classification labels of the historical mountain photovoltaic power generation data to train a lightweight deep learning model, obtains a photovoltaic power generation anomaly detection model, deploys the photovoltaic power generation anomaly detection model to each edge computing node, establishes a connection between the mountain photovoltaic array and the edge computing node, collects the mountain photovoltaic power generation data of the mountain photovoltaic array by the edge computing node according to the first preset data sampling frequency, and schedules the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data to determine the photovoltaic power generation anomaly detection result, effectively improving the real-time processing efficiency of the mountain photovoltaic power generation data, making the monitoring of the mountain photovoltaic power generation more timely and accurate, and ensuring the normal operation of the mountain photovoltaic power generation.

[0122] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0124] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the function specified in one process Figure 1 or more processes and / or blocks Figure 1 or more blocks.

[0125] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are performed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the function specified in one process Figure 1 or more processes and / or blocks Figure 1 or more blocks.

[0126] Those of ordinary skill in the art can understand that all or part of the steps in implementing the above facts and methods can be completed by instructing relevant hardware through a program. The involved program or the described program can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are introduced. The storage medium can be ROM / RAM, magnetic disk, optical disc, etc.

[0127] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A real-time processing method for mountain photovoltaic power generation data based on edge computing, characterized in that, Including: Collect historical mountain photovoltaic power generation data under various fault conditions, and preprocess the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data; Construct a classification label for each fault condition corresponding to the historical mountain photovoltaic power generation data to obtain the data classification label corresponding to the historical mountain photovoltaic power generation data; Construct a lightweight deep learning model, and use the preprocessed historical mountain photovoltaic power generation data and the data classification label corresponding to the historical mountain photovoltaic power generation data to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model; Deploy the photovoltaic power generation anomaly detection model to each edge computing node, and establish a connection between the mountain photovoltaic array and the edge computing node so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array; Collect the mountain photovoltaic power generation data of the mountain photovoltaic array by the edge computing node according to the first preset data sampling frequency, and schedule the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data to determine the photovoltaic power generation anomaly detection result; Transmit the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server through the edge computing node, and when the photovoltaic power generation anomaly detection result meets the preset condition, adjust the first preset data sampling frequency to a larger second preset data sampling frequency to strengthen the data monitoring and processing degree.

2. The real-time processing method for mountain photovoltaic power generation data based on edge computing according to claim 1, wherein Collect historical mountain photovoltaic power generation data under various fault conditions, and preprocess the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data, including: Collect the operating voltage data corresponding to the photovoltaic inverter in the mountain photovoltaic array under various fault conditions to obtain the first mountain photovoltaic power generation data; wherein, the first mountain photovoltaic power generation data is used to characterize the internal operating characteristics of the mountain photovoltaic array; Collect the infrared image data corresponding to the mountain photovoltaic array under various fault conditions to obtain the second mountain photovoltaic power generation data; wherein, the second mountain photovoltaic power generation data is used to characterize the surface operating characteristics of the mountain photovoltaic array; Use the first mountain photovoltaic power generation data and the second mountain photovoltaic power generation data together as the historical mountain photovoltaic power generation data, and preprocess the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data.

3. The real-time processing method for mountain photovoltaic power generation data based on edge computing according to claim 2, wherein, Preprocess the historical mountain photovoltaic power generation data to obtain the preprocessed historical mountain photovoltaic power generation data, including: Perform image enhancement processing on the historical mountain photovoltaic power generation data to obtain the historical mountain photovoltaic power generation data after image enhancement processing; wherein, only the second mountain photovoltaic power generation data is subjected to image enhancement processing; Perform sample equalization processing on the historical mountain photovoltaic power generation data after image enhancement processing to obtain the historical mountain photovoltaic power generation data after sample equalization processing; Perform image diversification processing on the historical mountain photovoltaic power generation data after sample equalization processing to obtain the preprocessed historical mountain photovoltaic power generation data.

4. The real-time processing method of mountain photovoltaic power generation data based on edge computing according to claim 3, characterized in that, Construct a lightweight deep learning model, and use the preprocessed historical mountain photovoltaic power generation data and the corresponding data classification labels of the historical mountain photovoltaic power generation data to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model, including: Construct a CNN-CNN-BP model to obtain a lightweight deep learning model; Use the preprocessed historical mountain photovoltaic power generation data as the input of the lightweight deep learning model, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output of the lightweight deep learning model, and use the hyperparameter joint optimization algorithm to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model.

5. The real-time processing method of mountain photovoltaic power generation data based on edge computing according to claim 4, characterized in that, Use the hyperparameter joint optimization algorithm to train the lightweight deep learning model to obtain a photovoltaic power generation anomaly detection model, including: For the first CNN model in the lightweight deep learning model, use the first mountain photovoltaic power generation data in the preprocessed historical mountain photovoltaic power generation data as the input, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use the intelligent optimization algorithm to train the first CNN model to obtain the trained first CNN model; For the second CNN model in the lightweight deep learning model, use the second mountain photovoltaic power generation data in the preprocessed historical mountain photovoltaic power generation data as the input, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use the intelligent optimization algorithm to train the second CNN model to obtain the trained second CNN model; Remove the classification layers of the first CNN model and the second CNN model to obtain a first feature extraction model and a second feature extraction model; For the BP neural network model in the lightweight deep learning model, extract the features of the first mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data through the first feature extraction model to obtain the first feature; extract the features of the second mountain photovoltaic power generation data in the historical mountain photovoltaic power generation data through the second feature extraction model to obtain the second feature; after fusing the first feature and the second feature, obtain the fused feature; use the fused feature as the input of the BP neural network model, use the corresponding data classification labels of the historical mountain photovoltaic power generation data as the expected output, and use the intelligent optimization algorithm to train the BP neural network model to obtain the trained BP neural network model to obtain a feature recognition model; Construct a feature fusion layer, connect the outputs of the first feature extraction model and the second feature extraction model to the input of the feature fusion layer, and connect the output of the feature fusion layer to the input of the feature recognition model to obtain a photovoltaic power generation anomaly detection model; wherein, the feature fusion layer is used to fuse the first feature and the second feature.

6. The real-time processing method for mountain photovoltaic power generation data based on edge computing according to claim 1, wherein Deploy the photovoltaic power generation anomaly detection model to each edge computing node, and establish a connection between the mountain photovoltaic array and the edge computing node, so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array, including: Deploy the photovoltaic power generation anomaly detection model to each edge computing node; For any mountain photovoltaic array, determine the edge computing node closest to the mountain photovoltaic array, and obtain the target edge computing node corresponding to each mountain photovoltaic array; Associate the mountain photovoltaic array with the target edge computing node to obtain the mountain photovoltaic array corresponding to each target edge computing node; For any target edge computing node, determine the load overrun situation of the target edge computing node; wherein, the load overrun situation includes that the number of connected mountain photovoltaic arrays exceeds a preset threshold or the number of connected mountain photovoltaic arrays does not exceed a preset threshold; When there is a load overrun situation corresponding to a target edge computing node that the number of connected mountain photovoltaic arrays exceeds a preset threshold, perform load reallocation and establish a connection between the mountain photovoltaic array and the edge computing node, so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array; When the load overrun situation corresponding to all target edge computing nodes is that the number of connected mountain photovoltaic arrays does not exceed a preset threshold, establish a connection between the mountain photovoltaic array and the edge computing node according to the mountain photovoltaic array corresponding to the target edge computing node, so that the edge computing node can obtain the mountain photovoltaic power generation data of the mountain photovoltaic array.

7. The real-time processing method of mountain photovoltaic power generation data based on edge computing according to claim 6, characterized in that, When there is a load overrun situation corresponding to a target edge computing node that the number of connected mountain photovoltaic arrays does not exceed a preset threshold, perform load reallocation and establish a connection between the mountain photovoltaic array and the edge computing node, including: Put all target edge computing nodes into the waiting-to-load list; Take out the target edge computing node with a load overrun situation that the number of connected mountain photovoltaic arrays exceeds a preset threshold to obtain the target edge computing node to be balanced; For the target edge computing node to be balanced, obtain the mountain photovoltaic array with the farthest distance in the target edge computing node to be balanced to obtain the mountain photovoltaic array to be separated; For the target edge computing node to be balanced, repeatedly obtain the mountain photovoltaic array to be separated until the load overrun situation of the target edge computing node to be balanced is that the number of connected mountain photovoltaic arrays does not exceed a preset threshold, and obtain at least one mountain photovoltaic array to be separated; Determine the target edge computing node closest to the mountain photovoltaic array to be separated in the waiting-to-load list to obtain the target edge computing node to be allocated, and allocate the mountain photovoltaic array to be separated to the target edge computing node to be allocated; Judge whether there is a target edge computing node with the number of connected mountain photovoltaic arrays exceeding the preset threshold in the waiting-to-load list. If so, return to the step of obtaining the target edge computing node to be balanced. Otherwise, complete the load reallocation process and establish a connection between the mountain photovoltaic array and the edge computing node.

8. The real-time processing method for mountain photovoltaic power generation data based on edge computing according to claim 5, characterized in that, Collect the mountain photovoltaic power generation data of the mountain photovoltaic array by the edge computing node at the first preset data sampling frequency, and schedule the photovoltaic power generation anomaly detection model to identify the mountain photovoltaic power generation data to determine the photovoltaic power generation anomaly detection result, including: Collect the mountain photovoltaic power generation data of the mountain photovoltaic array by the edge computing node at the first preset data sampling frequency to obtain the first mountain photovoltaic power generation data to be identified and the second mountain photovoltaic power generation data to be identified; Use the first mountain photovoltaic power generation data to be recognized as the input of the first feature extraction model in the photovoltaic power generation anomaly detection model, and use the second mountain photovoltaic power generation data to be recognized as the input of the first feature extraction model. Obtain the output of the feature recognition model in the photovoltaic power generation anomaly detection model to get the photovoltaic power generation anomaly detection result.

9. The real-time processing method for mountain photovoltaic power generation data based on edge computing according to claim 1, wherein, Transmit the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server through the edge computing node, and when the photovoltaic power generation anomaly detection result meets the preset conditions, adjust the first preset data sampling frequency to a larger second preset data sampling frequency, including: Transmit the mountain photovoltaic power generation data and the corresponding photovoltaic power generation anomaly detection result to the central server through the edge computing node by means of data encryption technology, so that the central server can uniformly manage the mountain photovoltaic power generation data; Judge whether the photovoltaic power generation anomaly detection result is an abnormal category. If so, it is determined that the preset conditions are met; otherwise, it is determined that the preset conditions are not met; When the photovoltaic power generation anomaly detection result meets the preset conditions, adjust the data sampling frequency to the second preset data sampling frequency, and continue to perform real-time processing on the mountain photovoltaic power generation data.

10. The real-time processing method for mountain photovoltaic power generation data based on edge computing according to claim 1, wherein, It also includes: Transmit the photovoltaic power generation anomaly detection result to the device designated by the staff.

Images