A method for detecting hot spots in photovoltaic arrays

Abstract

The present invention belongs to the technical field of photovoltaic system fault diagnosis, and specifically relates to a method for detecting hot spots in photovoltaic arrays. In view of the shortcomings of existing photovoltaic system hot spot detection methods that rely only on a single parameter and have low accuracy, the present invention adopts the following technical solutions: a method for detecting hot spots in photovoltaic arrays, comprising: step S1, using current, voltage, power and temperature as features, and the fault degree of the photovoltaic panel as a label, to construct a data set including a training set and a test set; step S2, building a neural network classifier, using an optimizer for optimization, and using the training set for training; step S3, collecting data and uploading it to a management system, and judging whether the photovoltaic array meets the warning rules according to preset warning rules; step S4, the management system pushes the warning information to relevant parties. The beneficial effect of the present invention is that compared with the existing method that only analyzes and judges based on current or temperature, the degree of hot spot fault of the photovoltaic panel can be judged more accurately.

Description

Technical Field

[0001] The present invention belongs to the technical field of photovoltaic system fault diagnosis, and in particular relates to a photovoltaic array hot spot detection method. Background Art

[0002] Under certain conditions, blocked or defective areas in a series branch of a photovoltaic module in a power-generating state can act as loads, consuming energy generated by other areas and causing localized overheating. This phenomenon is known as the "hot spot effect" of photovoltaic modules. This can reduce the module's output power and cause permanent damage, such as localized burnout (dark spots), melted solder joints, and aging of packaging materials. This directly impacts the module's output power and service life.

[0003] Existing methods for detecting hot spots in photovoltaic systems mainly include infrared thermal imaging and photovoltaic module IV curve scanning. The infrared thermal imaging method, based on the principle that photovoltaic cells exhibit significant temperature differences under different operating conditions, uses an infrared thermal imager to capture thermal images to identify hot spot faults. Fault detection can be performed using methods such as handheld detection, fixed-point setup, and drone-mounted infrared thermal imagers. While accurate hot spot location is possible, the equipment procurement and maintenance costs are high, the system is highly dependent on the performance of the imaging equipment, and it is susceptible to weather conditions. It requires a considerable amount of labor, is time-consuming, and has low efficiency. The IV curve scanning method requires significant hardware and labor investment, resulting in low detection efficiency.

[0004] In order to solve the cost and efficiency problems, a method of detecting hot spots based on current has emerged. For example: The Chinese invention patent application with publication number CN110620551A discloses a hot spot detection method, including: obtaining the current values ​​of all photovoltaic strings in the photovoltaic array in real time; determining the current average value of each photovoltaic string in a predetermined length time period based on the current value of each photovoltaic string obtained; determining the current average value of each photovoltaic string combination in the predetermined length time period based on the current value of each photovoltaic string obtained; determining whether a hot spot fault occurs in each photovoltaic string based on the current average value of each photovoltaic string in the predetermined length time period and the current average value of the photovoltaic string combination to which each photovoltaic string belongs in the predetermined length time period. The Chinese invention patent application with publication number CN108964606A discloses a method for detecting hot spot faults in a photovoltaic system, including the following steps: S1: establishing a theoretical current model for each string of each array in the photovoltaic system; S2: predicting the theoretical current output value of each string based on the theoretical current model;

[0005] S3: Real-time monitoring of the current output value of each string; S4: Pre-processing of the current output value of each string to screen out arrays that may have hot spots; S5: Determining whether there are hot spots in the screened arrays that may have hot spots to obtain a hot spot array; S6: Further determining the hot spot strings in the hot spot array; S7: Determining the location of the faulty component in the hot spot string.

[0006] The aforementioned method of detecting hot spots based on current offers advantages such as high feasibility and economical efficiency. However, relying solely on current as a single parameter leaves room for improvement in accuracy. Furthermore, data sharing is often time-sensitive, preventing staff from receiving timely feedback.

[0007] Alternatively, PV module temperature sampling devices can be installed to collect temperature data from various parts of the PV module surface, and the presence of hot spots can be detected based on temperature data alone or by comparing it with thermal imaging images. However, relying solely on temperature data suffers from the same low accuracy as current alone, and comparing temperature data with thermal imaging still carries the added cost of thermal imaging. Summary of the Invention

[0008] In order to address the shortcomings of existing photovoltaic system hot spot detection methods that rely only on a single parameter and have low accuracy, the present invention provides a photovoltaic array hot spot detection method that comprehensively analyzes multi-channel data of current, voltage, power, and temperature to improve the judgment accuracy.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: a photovoltaic array hot spot detection method, the photovoltaic array hot spot detection method comprising:

[0010] Step S1, using current, voltage, power, and temperature as input features and the fault degree of the photovoltaic panel as the output label, to construct a dataset including a training set and a test set;

[0011] Step S2: Build a neural network classifier, optimize it using an optimizer, train it using the training set, and continuously adjust it based on the test results of the test set to obtain the optimal model;

[0012] Step S3: Collect the current, voltage, power, and temperature data of the photovoltaic array during operation and upload them to the management system. The management system uses the model obtained in the above steps to perform inference and, based on the inference results and preset early warning rules, determines whether the photovoltaic array meets the early warning rules.

[0013] Step S4: When the warning rules are met, warning information is generated and the management system pushes the warning information to relevant parties.

[0014] The present invention's hot spot detection method for photovoltaic arrays comprehensively analyzes data from four channels: current, voltage, power, and temperature. Using machine learning algorithms, it establishes a multimodal model to more accurately determine the severity of hot spot failures in photovoltaic panels. The directly collected data are current, voltage, and temperature, while power is calculated from current and voltage.

[0015] As an improvement, in step S2, a neural network classifier is built based on PyTorch, and Adam is used as the optimizer for optimization.

[0016] As an improvement, the neural network classifier includes three layers of neurons: input layer, hidden layer and output layer. The input layer contains 4 neurons, which are used to input data of four channels: voltage, current, power and temperature respectively. The hidden layer contains 16 neurons for extracting features. The output layer contains 4 neurons for outputting the probabilities of four categories: normal, mild, moderate and severe.

[0017] As an improvement, the activation function of the hidden layer uses the Relu function, and the output layer is connected to the Softmax function to output the category probability.

[0018] As an improvement, during model training, all data are Z-score transformed to eliminate the influence of dimension. During model inference, the data to be predicted are first Z-score transformed using the mean μ and standard deviation σ of the above training samples, then input into the model for analysis and output of the predicted category.

[0019] The formula for Z-score transformation is:

[0020] As an improvement, in step S3, the early warning rules include the voltage, current, power, temperature of the photovoltaic module and the alarm threshold of the hot spot fault degree output by the model.

[0021] As an improvement, in step S3, the largest probability among the four output categories is taken as the basis for judgment.

[0022] As an improvement, in step S4, the early warning information is transmitted to the management system and each user terminal through 5G communication technology, so as to realize remote transmission of dynamic real-time monitoring data on site and remote interactive setting of working parameters; in step S4, the early warning information is sent by text message through the software APP to remind the staff to carry out maintenance.

[0023] As an improvement, step S5 is also included, which uses the 5G smart gateway technology architecture to transmit data to the cloud platform and display the data in the form of charts so that staff can view the photovoltaic panel data and status.

[0024] As an improvement, the deployment method of 5G smart gateway is as follows: the smart sensor is deployed on-site, the access method / network is Zigbee, and it is used to collect data; the 5G smart gateway is deployed on-site, the access method / network is 5G, and it is used for 5G networking, protocol transmission, data collection, and intelligent control; the communication gateway is deployed in the external network room, the access method / network is the Internet, and it is used for authentication, protocol forwarding, and security gateway; the communication service is deployed in the office room, the access method / network is the photovoltaic power station office network, and it is used for application communication; the application service is deployed in the office room, the access method / network is the photovoltaic power station office network, and it is used for monitoring data analysis and remote control.

[0025] The beneficial effects of the photovoltaic array hot spot detection method of the present invention are: comprehensive analysis of the four channel data of current, voltage, power, and temperature, and establishment of a multimodal model based on a machine learning algorithm. Compared with the existing methods that analyze and judge based on current or temperature alone, the degree of hot spot failure of the photovoltaic panel can be judged more accurately. Furthermore, a neural network classifier is built based on PyTorch, and Adam is used as the optimizer for optimization. The activation function of the hidden layer adopts the Relu function, and the output layer is connected to the Softmax function for category probability output, which can obtain more accurate judgment results. The current, voltage, and temperature data required for detection are directly collected by the sensor, the model has a small amount of calculation and is fast to calculate; infrared image data is only required in the model creation stage. After the model is established, there is no need to take infrared images during the detection process, thereby making the detection faster and more convenient. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a photovoltaic array hot spot detection method according to a first embodiment of the present invention.

[0027] Figure 2 This is a structural diagram of a neural network model of a photovoltaic array hot spot detection method according to a first embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following is an explanation and description of the technical solutions of the embodiments of the present invention in conjunction with the drawings of the embodiments of the present invention, but the following embodiments are only preferred embodiments of the present invention and are not exhaustive. Based on the embodiments in the implementation mode, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention.

[0029] A photovoltaic array hot spot detection method, the photovoltaic array hot spot detection method comprising:

[0030] Step S1, using current, voltage, power, and temperature as input features and the fault degree of the photovoltaic panel as the output label, to construct a dataset including a training set and a test set;

[0031] Step S2: Build a neural network classifier, optimize it using an optimizer, train it using the training set, and continuously adjust it based on the test results of the test set to obtain the optimal model;

[0032] Step S3: Collect the current, voltage, power, and temperature data of the photovoltaic array during operation and upload them to the management system. The management system uses the model obtained in the above steps to perform inference and, based on the inference results and preset early warning rules, determines whether the photovoltaic array meets the early warning rules.

[0033] Step S4: When the warning rules are met, warning information is generated and the management system pushes the warning information to relevant parties.

[0034] The photovoltaic array hot spot detection method of the present invention comprehensively analyzes data from four channels, including current, voltage, power, and temperature, and establishes a multimodal model based on a machine learning algorithm. Compared with the existing method that only analyzes and judges based on current, it can more accurately judge the degree of hot spot failure of photovoltaic panels.

[0035] Example 1

[0036] See also Figure 1 and Figure 2 , a photovoltaic array hot spot detection method according to a first embodiment of the present invention, the photovoltaic array hot spot detection method comprising:

[0037] Step S1, using current, voltage, power, and temperature as input features and the fault degree of the photovoltaic panel as the output label, to construct a dataset including a training set and a test set;

[0038] Step S2: Build a neural network classifier, optimize it using an optimizer, train it using the training set, and continuously adjust it based on the test results of the test set to obtain the optimal model;

[0039] Step S3: Collect the current, voltage, power, and temperature data of the photovoltaic array during operation and upload them to the management system. The management system uses the model obtained in the above steps to perform inference and, based on the inference results and preset early warning rules, determines whether the photovoltaic array meets the early warning rules.

[0040] Step S4: When the warning rules are met, warning information is generated and the management system pushes the warning information to relevant parties.

[0041] In this embodiment, in step S1, the fault degree of the photovoltaic panel (0 for normal, 1 for mild, 2 for medium, 3 for severe) is used as a label to construct a data set containing 1000 samples, of which 80% of the samples constitute the training set and the remaining 20% ​​of the samples constitute the test set. Specifically, voltage, current and temperature sensors are installed on a photovoltaic panel array and data is collected. At the same time, a drone is used on-site to collect hot spot effect data. Professionals judge the hot spot effect level according to the evaluation criteria (0 for normal, 1 for mild, 2 for medium, and 3 for severe). The data of the sensors on each photovoltaic panel and the hot spot effect level judged above are combined together as a piece of data, in which the voltage, current and temperature data are used as input values, and the hot spot effect level is used as a label (expected value). According to the above method, a total of 1000 pieces of data are repeatedly collected to construct a data set.

[0042] In this embodiment, in step S2, a neural network classifier is built based on PyTorch, using Adam as the optimizer for optimization and the Softmax function for normalization. The degree of hot spot failure of the photovoltaic module is determined based on the maximum category probability output by the model (for example, if the output probability distribution is normal 10%, mild 80%, moderate 5%, and severe 5%, then mild is used as the predicted degree of hot spot failure of the photovoltaic module). Training is performed using the training set in step 1, and hyperparameters are continuously adjusted based on the test results of the test set. Ultimately, the optimal model is used to determine the degree of hot spot failure of the photovoltaic module.

[0043] In this embodiment, in step S2, the structure of the neural network classifier used is as follows: it comprises 3 layers of neurons, namely an input layer, a hidden layer and an output layer. The input layer comprises 4 neurons, which are used to input data of the 4 channels of voltage, current, power and temperature respectively. The hidden layer comprises 16 neurons for extracting features. The output layer comprises 4 neurons, which are used to output the probabilities of the 4 categories of normal, mild, moderate and severe. The activation function of the hidden layer adopts the Relu function, and the output layer is connected to the Softmax function to output the category probability. During model training, all data are Z-score transformed to eliminate the influence of dimension. During model inference, the data to be predicted is first Z-score transformed using the mean μ and standard deviation σ of the above-mentioned training samples, and then input into the model for analysis to output the predicted category.

[0044] The Relu function formula is expressed as:

[0045] The formula for Z-score transformation is:

[0046] In this embodiment, in step S2, the model training method (process) is as follows: suppose the input data is vector X, the output is vector Y, the weight matrix between the input layer and the hidden layer is W, and the weight matrix between the hidden layer and the output layer is V.

[0047]

[0048] When the input layer feeds forward to the hidden layer, the input X and W are operated as follows to obtain the output value L1 of the hidden layer:

[0049] L1=relu(W·X+B w )

[0050]

[0051] Among them, relu is the activation function, B W is the bias value of the hidden layer.

[0052] Then L1 is combined with V to obtain the output value L2 of the output layer.

[0053] L2=relu(V·L1+B v )

[0054]

[0055] Among them, B v is the bias value of the output layer.

[0056] The output value L2 is normalized by the Softmax function to obtain the final result Y.

[0057] Y = softmax(L2)

[0058]

[0059] The cross entropy of the result Y and the expected value T is used as the loss function, and the gradient descent method is used to optimize the loss function. First, the gradient value of the loss function loss to the hidden layer output L1 is calculated, and the weight matrix V is adjusted according to the product of the gradient value and the learning rate to obtain a new value V'. Then, according to the error back propagation algorithm, the weight matrix W is adjusted according to the product of the gradient value and the learning rate to obtain a new value W'.

[0060]

[0061] The above steps are then repeated until the loss function reaches a minimum or the number of iterations reaches 2000. At this point, the model training is complete.

[0062] In this embodiment, in step S3, an early warning rule is set. The early warning rule can set alarm thresholds for the voltage, current, power, temperature of the photovoltaic module and the degree of hot spot failure output by the model. The management system determines whether to generate an early warning information based on the preset values ​​and actual values.

[0063] In this embodiment, in step S3, the model inference method (process) is as follows: when the model is inferred, the unknown data X new As the input value, it is input into the input layer of the model, and the forward operation is performed on X according to the above steps until the calculated value Y is obtained. new Then Y new This is the inference value of the unknown data. Assume that Y new =[0.050.80 0.10 0.05], then according to the principle of maximum probability, the hot spot effect level of the photovoltaic panel is judged to be level 2 (mild). During the inference process, no loss function calculation and gradient descent method weight adjustment are performed.

[0064] In this embodiment, in step 4, the warning information is transmitted to the management system and each user terminal through 5G communication technology, realizing remote transmission of dynamic real-time monitoring data on site and remote interactive setting of working parameters. The warning information can also be sent via text messages via the software APP to remind staff to perform maintenance.

[0065] In this embodiment, in step 4, 5G-DNN-based photovoltaic panel monitoring can access the 5G network through the Data Network Name (DNN), and the monitoring data stream is connected to the communication network in 5G-DNN mode. A gateway platform containing a private SIM card is configured, with the gateway General Packet Radio Service (GPRS) support node network as the core network element. The gateway GPRS support node gateway configuration and specific interface are determined based on the communication and connection requirements of the photovoltaic panel data monitoring. The data then passes through the 5G base station and the operator's core network to achieve the above functions.

[0066] This embodiment also includes step 5: using the 5G smart gateway technology architecture to transmit data to the cloud platform and display the data in the form of charts, allowing staff to view the photovoltaic panel data and status at various times of the day. The 5G smart gateway integrates AI early warning, edge computing, data aggregation, and data compression functions. There is no need to upload the data center through the 5G network and then return the instructions. The processing process is completed at the local edge computing layer. The data is sent to the cloud platform via 5G encrypted transmission and displayed on the personal computer (PC) management platform and the handheld application platform.

[0067] The 5G smart gateway deployment method is as follows:

[0068] ① The smart sensor is deployed on-site and the access method / network is Zigbee for data collection;

[0069] ② The 5G smart gateway is deployed on-site, with a 5G access mode / network, and is used for 5G networking, protocol transmission, data collection, and intelligent control;

[0070] ③ The communication gateway is deployed in the external network room, with the access method / network being the Internet, and is used for authentication, protocol forwarding, and security gateway;

[0071] ④ The communication service is deployed in the office room, and the access method / network is the photovoltaic power station office network, which is used for application communication, etc.

[0072] ⑤ The application service is deployed in the office computer room, and the access method / network is the photovoltaic power station office network, which is used for monitoring data analysis, remote control, etc.

[0073] The following describes a specific application example.

[0074] At a certain photovoltaic power station, the photovoltaic panel array consists of six strings, each containing 20 modules (panels), for a total of 120 photovoltaic panels. Each panel is equipped with independent voltage, current, and temperature sensors with detection ranges of 0-110V, 0-10A, and -30-500°C, respectively. The sensors communicate with the edge computing device via a serial port. The edge computing device uses a Raspberry Pi 4B as its motherboard, with computing resources configured with 8GB of memory, a 1.5GHz quad-core CPU, and 16GB of flash memory. The edge computing device software is developed in Java. The edge computing device runs an algorithm model built with the PyTorch deep learning framework. This model is a three-layer error back-propagation neural network that calculates the panel's hot spot fault severity based on real-time voltage, current, power (calculated from voltage and current), and temperature data transmitted by the sensors.

[0075] For example, a photovoltaic panel sensor measured an output voltage of 36V, a current of 0.5A, and a surface temperature of 105°C. The neural network calculated the probabilities of categories 1 to 4 (normal to severe) as 0.05, 0.10, 0.50, and 0.35, respectively. The category with the highest probability is considered moderate hot spot effect. See Table 1 for other data examples.

[0076] Table 1 Photovoltaic panel sensor data and hot spot effect identification example data

[0077]

[0078] The edge computing device is connected to a 5G communication module and pushes locally detected sensor data from each component, along with the estimated severity of hot spot faults, to the cloud management platform via HTTP. The cloud management platform is based on a client / server architecture. The backend server software is developed using Java / Springboot / SpringMVC, the database uses MySQL, and the front-end web page is developed using Vue.js / ElementUI to provide user interaction services. User terminals include browser-based web pages on personal computers and mobile applications such as Android apps, iOS apps, and WeChat mini-programs. These user terminals download PV panel data and warning information from the backend management platform via HTTP.

[0079] The beneficial effects of the photovoltaic array hot spot detection method of the first embodiment of the present invention are: evaluating the degree of hot spot failure of the photovoltaic panel based on current, voltage, power and temperature detection, and comprehensively analyzing the four channel data of current, voltage, power, temperature, etc., which can more accurately judge the degree of hot spot failure of the photovoltaic panel than the existing method that only analyzes and judges based on current; constructing a multimodal hot spot recognition model based on deep learning, and using this model to analyze the multi-channel status data collected by sensors on the photovoltaic panel array to evaluate the degree of hot spot failure of the photovoltaic panel; constructing early warning rules, and pushing early warning information and fault degree classification to the management system and each user terminal, dynamic real-time monitoring of data, and remote interactive setting of working parameters; relying on 5G network and cloud platform to deploy the method at the edge to realize the deployment and application of the method and intelligent early warning; using 5G transmission module to realize real-time sharing of data, staff can receive relevant early warning information in time and find the problem based on relevant data; it has the functions of on-site dynamic real-time monitoring of data, remote interactive setting of working parameters, early warning classification and push, historical data cloud platform recording, etc.

[0080] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Those skilled in the art will understand that the present invention includes, but is not limited to, the contents described in the above specific embodiments. Any modifications that do not deviate from the functional and structural principles of the present invention are intended to be included within the scope of the claims.

Claims

1. A method for detecting hot spots in a photovoltaic array, characterized by: The photovoltaic array hot spot detection method comprises: Step S1, using current, voltage, power, and temperature as input features and the fault degree of the photovoltaic panel as the output label, to construct a dataset including a training set and a test set; Step S2: Build a neural network classifier, optimize it using an optimizer, train it using the training set, and continuously adjust it based on the test results of the test set to obtain the optimal model; Step S3: Collect the current, voltage, power, and temperature data of the photovoltaic array during operation and upload them to the management system. The management system uses the model obtained in the above steps to perform inference and, based on the inference results and preset early warning rules, determines whether the photovoltaic array meets the early warning rules. Step S4: When the warning rules are met, a warning message is generated and the management system pushes the warning message to relevant parties; In step S1, voltage, current, and temperature sensors are installed on the photovoltaic panel array to collect data. Hot spot effect data is collected on-site using a drone. Professionals determine the hot spot effect level based on evaluation criteria. The sensor data on each photovoltaic panel and the determined hot spot effect level are combined into a single piece of data, with the voltage, current, and temperature data serving as input values ​​and the hot spot effect level serving as a label. In step S2, a neural network classifier is built based on PyTorch and optimized using Adam as the optimizer. The neural network classifier includes three layers of neurons: an input layer, a hidden layer, and an output layer. The input layer contains four neurons, which are used to input data from four channels: voltage, current, power, and temperature. The hidden layer contains 16 neurons for feature extraction. The output layer contains four neurons for outputting the probabilities of four categories: normal, mild, moderate, and severe.

2. A photovoltaic array hot spot detection method according to claim 1, characterized in that: In step S2, the activation function of the hidden layer adopts the Relu function, and the output layer is connected to the Softmax function to output the category probability. The Relu function is expressed as:

3. A photovoltaic array hot spot detection method according to claim 2, characterized in that: In step S2, during model training, all data are Z-score transformed to eliminate the influence of dimension. During model inference, the data to be predicted are first Z-score transformed using the mean μ and standard deviation σ of the training samples, then input into the model for analysis and output of the predicted category. The formula for Z-score transformation is:

4. A photovoltaic array hot spot detection method according to claim 3, characterized in that: In step S2, let the input data be vector X, the output data be vector Y, the weight matrix between the input layer and the hidden layer be W, and the weight matrix between the hidden layer and the output layer be V, which can be expressed as follows: When the input layer feeds forward to the hidden layer, the input X and W are operated as follows to obtain the output value L1 of the hidden layer: L1=train(W·X+B w ) Among them, relu is the activation function, B W is the bias value of the hidden layer; Then, L1 is combined with V to obtain the output value L2 of the output layer: L2=train(V·L1+B v ) Among them, B v is the bias value of the output layer, and the output value L2 is normalized by the Softmax function to obtain the final result Y: Y = softmax(L2) The result Y and the expected value T are used to calculate the cross entropy as the loss function, and the gradient descent method is used to optimize the loss function. First, the gradient value of the loss function loss to the hidden layer output L1 is calculated, and the weight matrix V is adjusted according to the product of the gradient value and the learning rate to obtain a new value V'. Then, according to the error back propagation algorithm, the weight matrix W is adjusted according to the product of the gradient value and the learning rate to obtain a new value W': Repeat the above steps until the loss function reaches the minimum or the number of iterations reaches the preset number.

5. A photovoltaic array hot spot detection method according to claim 4, characterized in that: In step S3, when the model is inferring, the unknown data X new As the input value, it is input into the input layer of the model, and the forward operation is performed on X according to the above steps until the calculated value Y is obtained. new , then Y new That is, it is the inference value of unknown data, and no loss function calculation and gradient descent method are used to adjust the weights during the inference process.

6. The photovoltaic array hot spot detection method according to claim 1, characterized in that: In step S3, the early warning rules include the voltage, current, power, temperature of the photovoltaic module and the alarm threshold of the hot spot fault degree output by the model; in step S3, the largest probability of the four output categories is taken as the judgment basis.

7. The photovoltaic array hot spot detection method according to claim 1, characterized in that: In step S4, the warning information is transmitted to the management system and each user terminal through 5G communication technology, realizing remote transmission of dynamic real-time monitoring data on site and remote interactive setting of working parameters; in step S4, the warning information is sent via text messages via the software APP to remind staff to perform maintenance.

8. The photovoltaic array hot spot detection method according to claim 1, characterized in that: The method further includes step S5, wherein the data is transmitted to a cloud platform using a 5G smart gateway technology architecture, and the data is displayed in a chart format for staff to view the photovoltaic panel data and status; The 5G smart gateway deployment method is as follows: the smart sensor is deployed on-site, with the access method / network being Zigbee, for data collection; the 5G smart gateway is deployed on-site, with the access method / network being 5G, for 5G networking, protocol transmission, data collection, and intelligent control; the communication gateway is deployed in the external network room, with the access method / network being the Internet, for authentication, protocol forwarding, and security gateway. The communication service is deployed in the office computer room, and the access method / network is the photovoltaic power station office network, which is used for application communication. The application service is deployed in the office computer room, and the access method / network is the photovoltaic power station office network, which is used for monitoring data analysis and remote control.

Images