Photovoltaic station fire prevention method, system, equipment and medium

Abstract

The invention discloses a photovoltaic station fire prevention method, system, equipment and medium, and relates to the field of monitoring and early warning, and the method comprises the steps: arranging a sensor at a node of a photovoltaic module, collecting environment data in real time, and triggering a pre-alarm when the environment data exceeds a threshold value; after the pre-alarm is triggered, the photovoltaic module is accessed through a monitoring system to collect internal real-time data, the real-time data is processed based on a feature processing algorithm, and the processed real-time data is subjected to feature recognition and labeled; inputting the labeled data into a machine learning model of a monitoring platform to obtain a diagnosis result; and triggering an execution action based on the pre-alarm information and the diagnosis result. According to the invention, a closed-loop control mechanism from anomaly detection, reason identification, risk assessment, hierarchical response and fault isolation is formed, early identification of abnormal factors is realized, the timeliness and accuracy of early warning are improved, and the system can give out a pre-alarm when initial signals such as temperature and gas concentration changes appear to intervene in potential fire risks in advance.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and early warning, and in particular to a method, system, equipment and medium for preventing fires in photovoltaic stations. Background Art

[0002] During the actual operation of photovoltaic power plants, components such as PV module connectors, inverter cables, internal solder joints in junction boxes, and MC4 connectors often experience high-temperature melting or even burnout. These thermal failures are often caused by long-term high-load operation, excessive contact resistance, poor contact, or component aging. If these abnormal temperature rises go undetected, they can not only lead to reduced equipment efficiency and component failure, but can also easily cause sparks, arcs, and even fires, resulting in widespread system downtime, equipment damage, and even casualties, seriously threatening the operational safety and economic benefits of the power plant.

[0003] To address this issue, existing photovoltaic power plants generally use the following technical means for monitoring and prevention, including regular manual inspections: Operation and maintenance personnel periodically use tools such as infrared thermometers and handheld smoke detectors to inspect equipment within the station;

[0004] The SCADA system collects electrical parameters: real-time monitoring of basic operating data such as voltage, current, and power;

[0005] Install basic smoke detection devices: Install smoke detectors in some high-risk areas for local fire awareness.

[0006] However, these traditional methods still have many obvious drawbacks, including delayed response: manual inspections are infrequent and have long intervals, making it easy to miss the early stages of abnormal development, and it is often too late to detect them;

[0007] Insufficient monitoring dimensions: Traditional SCADA systems focus on electrical parameters and have limited monitoring capabilities for physical signs such as temperature rise and gas anomalies.

[0008] Strong deployment limitations: Ordinary smoke detectors lack intelligent analysis and positioning capabilities, resulting in high false alarm and missed alarm rates, making it difficult to achieve rapid response and accurate isolation.

[0009] Lack of linkage mechanism: Most existing systems are passive alarms and lack the integrated closed-loop processing capabilities of "monitoring-analysis-early warning-isolation". Manual intervention is still required after a fire occurs, posing a major hidden danger.

[0010] To address the issues of delayed response, inaccurate early warning, and limited control methods in existing photovoltaic station fire prevention and control technologies, this paper proposes a photovoltaic station fire prevention method that incorporates multi-dimensional sensing capabilities, intelligent identification and early warning capabilities, and emergency isolation and control capabilities. By deploying intelligent devices such as wireless temperature sensors and smoke detectors at key nodes, combined with a data platform and intelligent analysis algorithms, this method enables early identification, early warning, and early resolution of faults. Summary of the Invention

[0011] In view of the above-mentioned problems, the present invention is proposed.

[0012] Therefore, the problem to be solved by the present invention is that the existing solutions have problems such as response delay, insufficient monitoring dimensions, strong deployment limitations, and lack of linkage mechanism.

[0013] In order to solve the above technical problems, the present invention provides the following technical solutions: a photovoltaic station fire prevention method, which includes: arranging sensors at the nodes of photovoltaic modules to collect environmental data in real time, and triggering a pre-alarm when the environmental data exceeds a threshold; after the pre-alarm is triggered, accessing the photovoltaic modules through a monitoring system to collect internal real-time data, processing the real-time data based on a feature processing algorithm, and performing feature recognition and labeling on the processed real-time data; inputting the labeled data into a machine learning model of a monitoring platform to obtain a diagnosis result; and triggering an execution action based on the pre-alarm information and the diagnosis result.

[0014] As a preferred solution of the photovoltaic station fire prevention method described in the present invention, the real-time collection of environmental data includes setting a dynamic sensitivity adjustment mechanism to adjust the sensor sampling interval and threshold according to changes in ambient temperature, time period, and equipment load.

[0015] As a preferred solution of the photovoltaic station fire prevention method described in the present invention, when the environmental data exceeds the threshold, the pre-alarm is triggered, including setting an edge computing module to be deployed in the on-site centralized control cabinet to perform local real-time judgment, and triggering a local alarm when the temperature rise curve change rate and the smoke concentration instantaneous jump exceed the threshold.

[0016] As a preferred solution of the photovoltaic station fire prevention method described in the present invention, the method of collecting internal real-time data from photovoltaic modules through a monitoring system includes, while the monitoring system is collecting data, combining infrared thermal imaging inspection data to identify hidden faults in photovoltaic modules, and introducing spectrum analysis and transient current anomaly detection modules to capture short-term interference phenomena.

[0017] As a preferred solution of the photovoltaic station fire prevention method described in the present invention, the feature recognition and labeling of the processed real-time data includes establishing a pattern recognition rule set based on historical fault sample data, comparing and associating the processed real-time data with the pattern recognition rule set, finding a corresponding label in the pattern recognition rule set for each real-time data, and labeling the real-time data.

[0018] As a preferred embodiment of the photovoltaic station fire prevention method described in the present invention, the method further comprises: inputting the labeled data into a machine learning model, constructing a spatial distribution map, and determining whether the anomaly is concentrated in one location or spans multiple components; using a regression model to predict future trends based on the time series data in the data, calculating the slope of the trend curve, fitting the residuals, and outputting the degradation rate of the photovoltaic components; outputting a fault score based on the environmental information and electrical information in the input data; and the structural pattern of the diagnostic result is that when the fault score is higher than a score threshold or the degradation rate of the photovoltaic component is higher than a rate threshold, it is determined to be an anomaly; if the anomaly is concentrated in one location, it is determined to be a single-point anomaly; if the anomaly spans multiple components, it is determined to be a multi-point anomaly.

[0019] As a preferred solution of the photovoltaic station fire prevention method described in the present invention, the triggering of execution actions based on pre-alarm information and diagnosis results includes: when the diagnosis result is a single-point abnormality and the pre-alarm information is that the temperature rise exceeds the threshold or the smoke concentration exceeds the threshold, the risk level is set to level one, and the execution action is a pop-up alarm on the operation and maintenance platform, and a manual re-inspection is performed; when the diagnosis result is a single-point abnormality and the pre-alarm information is that the temperature exceeds 110°C or the smoke concentration exceeds the threshold by 50%, the risk level is set to level two, and the execution action is to start the sound and light alarm and trigger the pre-isolation instruction; when the diagnosis result is a multi-point abnormality and the pre-alarm information is that the temperature rise exceeds the threshold or the smoke concentration exceeds the threshold, the risk level is set to level three, and the execution action is to immediately perform emergency isolation and fire extinguishing.

[0020] In order to solve the above technical problems, the present invention provides the following technical solutions: a photovoltaic station fire prevention system, comprising: an environmental data acquisition module, an edge computing module, an electrical data acquisition module, a transient current anomaly detection module, a feature recognition module, a diagnosis module and an execution module; the environmental data acquisition module collects environmental data in real time based on sensors arranged on photovoltaic modules; the edge computing module performs local real-time judgment, and triggers a local alarm when the temperature rise curve change rate and the smoke concentration instantaneous jump exceed the threshold; the electrical data acquisition module accesses the photovoltaic modules through the monitoring system to collect internal real-time data; the transient current anomaly detection module is used to capture short-term interference phenomena; the feature recognition module processes the real-time data based on the feature processing algorithm, and performs feature recognition and labels the processed real-time data; the diagnosis module inputs the labeled data into the machine learning model of the monitoring platform to obtain the diagnosis result; the execution module triggers the execution action based on the pre-alarm information and the diagnosis result.

[0021] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the photovoltaic station fire prevention method described above when executing the computer program.

[0022] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the photovoltaic station fire prevention method described above.

[0023] The beneficial effects of the present invention are as follows: the present invention forms a closed-loop control mechanism from abnormality detection → cause identification → risk assessment → hierarchical response → fault isolation, realizing early identification of multi-dimensional abnormal factors, improving the timeliness and accuracy of early warning, and the system can issue pre-alarm information when initial signals such as sudden temperature rise and gas concentration changes appear, so as to intervene in potential fire risks in advance.

[0024] The present invention constructs a data-driven intelligent recognition model to enhance the ability to identify hidden dangers and faults. It uses the SCADA system and edge computing module to collect the electrical operation characteristics inside the photovoltaic module. Through the multivariate feature processing algorithm, it extracts the core state variables, improves the recognition accuracy of typical fault modes such as PID effect, loose joints, cable aging, and bypass diode failure, and automatically completes fault labeling to provide standard samples for subsequent modeling.

[0025] Through the classification / scoring model obtained through historical data training, the platform can simultaneously output three key diagnostic indicators: abnormal location judgment (single point / multi-point), component degradation trend, and comprehensive fault score (risk value or risk level). It can continuously track the risk curve of each node, discover the trend of continuous increase in risk, and further improve the foresight of prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 This is a flow chart of a photovoltaic station fire prevention method in Example 1.

[0028] Figure 2 This is a diagram of the sensor layout structure of a photovoltaic station fire prevention method in Example 1. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0031] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a photovoltaic station fire prevention method including: Figure 1 As shown:

[0032] Step 1: Place sensors at the nodes of the photovoltaic modules to collect environmental data in real time. When the environmental data exceeds the threshold, a pre-alarm is triggered.

[0033] Specifically, such as Figure 2 As shown, wireless temperature sensors are installed at key nodes such as junction boxes, combiner boxes, inverter interfaces, and high-voltage side circuit breakers of power station transformers. The sensors are used to accurately capture key dynamic data such as temperature, smoke, and gas concentration.

[0034] Set up a dynamic sensitivity adjustment mechanism to adjust the sensor sampling interval and threshold according to changes in ambient temperature, time period, and equipment load to improve early recognition capabilities.

[0035] An additional edge computing module, deployed in the on-site centralized control cabinet, performs local real-time judgment, reducing the impact of data upload latency on response time. Through edge judgment, when the temperature rise curve change rate or smoke concentration transient jump exceeds the threshold, a local pre-alarm is triggered.

[0036] Step 2: After the pre-alarm is triggered, the monitoring system is connected to the photovoltaic module to collect internal real-time data, the real-time data is processed based on the feature processing algorithm, and the processed real-time data is feature identified and labeled.

[0037] Specifically, while the monitoring system collects data, it combines infrared thermal imaging inspection data to identify hidden faults in photovoltaic modules. At the same time, it introduces spectrum analysis and transient current anomaly detection modules to capture short-term interference phenomena (such as lightning strikes and PID effects) and enhance the system's ability to identify "non-continuous fire source inducements."

[0038] Among them, the feature processing algorithm can use principal component analysis (PCA) to process the data. The specific steps include:

[0039] Raw data construction: The system collects multi-dimensional electrical parameters, including voltage, current, power, insulation resistance, and temperature, from key equipment at PV stations (such as modules, combiner boxes, and inverters). These parameters are aggregated in time series to form a multi-dimensional feature matrix. The samples are monitoring data at different time points or for different modules.

[0040] Standardization: Because electrical parameters have different physical dimensions, such as temperature in degrees Celsius and power in kilowatts, all feature data must be standardized to eliminate the impact of dimensional differences. The goal of standardization is to ensure that the average value of each feature is zero and the fluctuation range is controlled within a certain scale, thereby making data of different dimensions comparable.

[0041] Calculating Correlations Between Features: For the standardized dataset, we calculate the correlations between the electrical features and construct a correlation matrix to identify features that are redundant (for example, current and power may be highly correlated). This step lays the foundation for extracting the most informative features.

[0042] Principal Component Extraction: Through mathematical analysis of the correlation matrix, several principal components are extracted. Each principal component is a linear combination of multiple original features, representing the most important information direction in the data. These principal components are ranked in order of information content. The system selects the top principal components for retention based on their cumulative contribution, typically retaining at least 90% of the total information.

[0043] Data dimensionality reduction: Using the principal components selected above, the original high-dimensional feature data is projected into a low-dimensional space. This preserves the key information while reducing the dimensionality, making subsequent fault identification or cluster analysis more efficient and stable.

[0044] Generate an optimized feature set: After principal component analysis, the system obtains a new, lower-dimensional but higher-information-density dataset. This dataset is used for subsequent fault type identification, risk clustering, labeling, and model training.

[0045] In another possible embodiment, the feature processing algorithm may process data using a K-means clustering method. The specific steps include:

[0046] Collecting and constructing feature sample sets: Electrical operating data such as voltage, current, power, insulation resistance, and temperature are collected from key PV station equipment. The system first summarizes these parameters by time window (e.g., 10 minutes, 1 hour, etc.) and statistically extracts representative characteristic indicators, such as the mean, maximum, and variation of each indicator within the window; the rate of temperature rise, current fluctuation rate, power stability, and the rate of change compared to the previous time window.

[0047] Through these operations, a set of “feature vectors”, i.e. a sample point, is constructed for each monitoring node in each time period, forming a structured data set.

[0048] Preprocessing and standardization: Since the dimensions of each feature are different, in order to avoid a certain feature value dominating the clustering calculation due to its large absolute value, all features need to be standardized (such as normalization or Z-score standardization) to ensure that the contribution of each feature is at the same level.

[0049] Set the number of clusters (K value): Select the number of cluster categories K, which is usually determined by the following methods: engineering experience (for example, if the equipment status is divided into three categories: "normal", "mild abnormality", and "serious abnormality", K = 3 can be set).

[0050] Data-driven methods, such as using the Elbow Method to view the trend of clustering loss function changes corresponding to different K values, and select the K value at the inflection point.

[0051] Perform K-means clustering: The K-means algorithm divides all samples into K clusters, and each sample point is assigned to the nearest cluster center.

[0052] Its core goal is to minimize the sum of the distances between the sample points and the center points to which they belong. The mathematical objective function is:

[0053]

[0054] Among them, minimize means to minimize, x represents the sample feature vector, μ i Denotes the center of the i-th cluster, C i It represents the sample set of the i-th category, and i represents the index.

[0055] The main steps of the algorithm are as follows: randomly initialize K cluster centers; calculate the distance from all sample points to each center and assign each sample to the nearest center; recalculate the center of each cluster (that is, the mean of all samples belonging to this class); until the center point no longer changes significantly or the maximum number of iterations is reached.

[0056] Analyze clustering results: After clustering is complete, all samples are grouped according to their category labels, yielding the following information: The center point of each sample category represents a "typical electrical operating state." If a certain sample category is distributed far from the center, or if characteristics such as temperature and current deviate significantly, it can be preliminarily identified as an abnormal cluster. The "fault risk level" can be further defined by comparing indicators such as the temperature rise and voltage drop across each category.

[0057] The system can also determine which cluster a new sample collected in real time should belong to, to determine whether it belongs to the high-risk anomaly class.

[0058] Output cluster labels and structured storage: Each sample will be labeled with its clustering result (for example, "Cluster_1", "Cluster_3", etc.), and these labels can be converted into meaningful status labels based on engineering definitions, such as: Cluster_0 → normal operation; Cluster_1 → potential degradation; Cluster_2 → high-risk anomaly.

[0059] After associating the labels with metadata such as the original sampling time and device number, the output is a structured result, which provides input data for subsequent fault identification or platform learning models.

[0060] Specifically, feature recognition and labeling of the processed real-time data include establishing a pattern recognition rule set based on historical fault sample data, comparing and associating the processed real-time data with the pattern recognition rule set, finding a corresponding label in the pattern recognition rule set for each real-time data, and labeling the real-time data.

[0061] Step 3: Input the labeled data into the machine learning model of the monitoring platform to obtain the diagnosis results.

[0062] Specifically, the labeled data is fed into a machine learning model to construct a spatial distribution map to determine whether the anomalies are concentrated in one location or across multiple components.

[0063] Based on the time series data in the data, a regression model is used to predict future trends, calculate the slope of the trend curve, fit the residuals, and output the degradation rate of the photovoltaic modules.

[0064] Based on the environmental and electrical information in the input data, a fault score is output.

[0065] The structural pattern of the diagnosis results is that when the fault score is higher than the score threshold or the degradation rate of the PV module is higher than the rate threshold, it is judged as abnormal.

[0066] If the anomaly is concentrated in one location, it is considered a single-point anomaly. If the anomaly spans multiple components, it is considered a multi-point anomaly.

[0067] The fault score can be calculated based on the real-time operating status and historical characteristic data of the PV modules, and a fault risk value between 0 and 1 is used to quantify the degree of danger of the current operating conditions. The closer the value is to 1, the higher the fault risk, and the closer to 0, the safer it is. The specific steps include:

[0068] Feature vector input preparation: From the labeled dataset output in the previous step, select representative feature variables as model input. These typically include the following: environmental and equipment parameters: temperature, smoke concentration, and light intensity; electrical operating characteristics: current, voltage, power, and insulation resistance; derived statistics: current fluctuation rate, voltage drop rate, and temperature rise rate; and cluster labels or principal component values ​​(such as those from a pre-processed K-means or PCA).

[0069] Each sample is a complete feature vector of a certain moment, component or node.

[0070] Build a fault risk scoring model: Select a regression model or probabilistic classification model that can output continuous probabilities, such as logistic regression, random forest with probabilities, gradient boosting tree (such as XGBoost), or lightweight neural network model.

[0071] These models all have the ability to "map input features to a probability value between 0 and 1".

[0072] For example, when using logistic regression, the output value is calculated as the result of sigmoid function processing, which is expressed as:

[0073]

[0074] Here, z is represented as a linear combination of eigenvectors.

[0075] Model training and calibration: The model is trained using a historically labeled dataset (i.e., which time periods actually experienced failures and which were normal). The training goal is to accurately distinguish between "normal" and "faulty" samples and ensure that the model output reflects the actual risk probability as closely as possible.

[0076] The following data can be used as training samples: feature vector (input); label indicating whether a fault has occurred (output); fault type (optional, used for refined analysis); after training, calibration techniques (such as Platt scaling or Isotonic Regression) can be used to probabilistically calibrate the model output to make its risk value in the range of 0 to 1 more realistic and interpretable.

[0077] Real-time Scoring and Output: During the deployment phase, the system receives real-time feature inputs at intervals (e.g., 5 minutes) and uses the trained model to calculate the risk value of the current state. For example, component A's current failure risk score is 0.18 (low risk); component B's is 0.53 (medium risk); and component C's is 0.91 (high risk).

[0078] The system can set segmented reference lines for risk values ​​(e.g. >0.8 for high risk, >0.5 for medium risk), or pass the value directly to the response mechanism to participate in risk level judgment and isolation strategy triggering.

[0079] In another optional embodiment, the fault score can also be a discrete level label (such as low, medium, high). The specific steps include:

[0080] Based on the system's extracted operational characteristics, the system determines the risk level of the current component or subsystem. For example, "Low Risk": normal operation, no intervention required; "Medium Risk": potential hidden dangers, requiring maintenance; "High Risk": obvious fault characteristics, requiring immediate response or isolation. This result serves as direct input to the "Graded Response Mechanism" to trigger the corresponding control strategy.

[0081] Input feature preparation: Select key feature vectors provided by front-end modules (such as "feature extraction" and "fault identification"). These typically include: electrical parameter features such as power drop rate, voltage fluctuation rate, current sag amplitude, and temperature rise rate; derived features or statistics such as short-term mean, peak value, and sliding standard deviation; optional environmental information such as sunlight intensity, temperature, and wind speed; and optional labels or clustering results such as PCA principal component scores and K-means cluster numbers. These features together constitute the discriminant input.

[0082] Sample annotation and label definition: In historical data, the risk level of each time segment is manually or rule-basedly labeled as training labels. Common practices include:

[0083] Based on actual fault records, fault severity is labeled as "high," "medium," or "low." Risk levels are assigned based on certain numerical thresholds (e.g., temperature above 85°C combined with a sudden current drop = high risk). Corresponding cluster center feature analysis: Highly abnormal cluster features are associated with a high-risk label. All samples are assigned clear risk level labels for supervised learning.

[0084] Select a classification model and train it: Use a machine learning model suitable for multi-classification problems, such as decision tree, random forest, gradient boosting tree (such as LightGBM, XGBoost), etc.

[0085] These models take a feature vector as input and output one of three categories: low, medium, or high. The training process is to minimize the error between the input data and the labeled labels.

[0086] In order to enhance the stability and interpretability of the model, the model results can be subjected to cross-validation, feature importance evaluation and other processing.

[0087] Model output and level mapping: After the model is deployed, each time it receives a feature input, it outputs a risk level label. For example, component A currently has a low level; component B currently has a medium level; component C currently has a high level. A threshold can be set to indicate an anomaly if the level is medium or above.

[0088] Step 4: trigger execution actions based on the pre-alarm information and diagnosis results.

[0089] Specifically, when the diagnosis result is a single-point abnormality and the pre-alarm information is that the temperature rise exceeds the threshold or the smoke concentration exceeds the threshold, the risk level is set to level one, and the execution action is to pop up an alarm on the operation and maintenance platform and conduct manual re-inspection.

[0090] When the diagnosis result is a single-point abnormality and the pre-alarm information is that the temperature exceeds 110°C or the smoke concentration exceeds the threshold of 50%, the risk level is set to level two, and the execution action is to activate the sound and light alarm and trigger the pre-isolation instruction.

[0091] When the diagnosis result is multiple-point abnormality and the pre-alarm information is that the temperature rise exceeds the threshold or the smoke concentration exceeds the threshold, the risk level is set to level three, and the execution action is to immediately perform emergency isolation and fire extinguishing.

[0092] Example 2 is the second embodiment of the present invention, which is different from the first embodiment in that: a photovoltaic station fire prevention system includes an environmental data acquisition module, an edge computing module, an electrical data acquisition module, a transient current anomaly detection module, a feature recognition module, a diagnosis module and an execution module; the environmental data acquisition module collects environmental data in real time based on sensors arranged on photovoltaic modules; the edge computing module performs local real-time judgment, and triggers a local alarm when the temperature rise curve change rate and the smoke concentration instantaneous jump exceed the threshold; the electrical data acquisition module accesses the photovoltaic modules through the monitoring system to collect internal real-time data; the transient current anomaly detection module is used to capture short-term interference phenomena; the feature recognition module processes the real-time data based on the feature processing algorithm, and performs feature recognition and labels the processed real-time data; the diagnosis module inputs the labeled data into the machine learning model of the monitoring platform to obtain the diagnosis result; the execution module triggers the execution action based on the pre-alarm information and the diagnosis result.

[0093] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0094] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0095] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0096] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, it can be implemented using a combination of any of the following technologies known in the art: a discrete logic circuit having logic gate circuits for implementing logic functions on data signals, an application-specific integrated circuit having suitable combinational logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0097] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for preventing fires in a photovoltaic station, characterized by: include, Sensors are placed at the nodes of photovoltaic modules to collect environmental data in real time. When the environmental data exceeds the threshold, a pre-alarm is triggered. After the pre-alarm is triggered, the monitoring system accesses the photovoltaic panels to collect internal real-time data, processes the real-time data based on the feature processing algorithm, and performs feature recognition and labeling on the processed real-time data; Input the labeled data into the monitoring platform's machine learning model to obtain diagnostic results; Trigger execution actions based on pre-alarm information and diagnostic results.

2. A photovoltaic station fire prevention method according to claim 1, characterized in that: The real-time collection of environmental data includes setting a dynamic sensitivity adjustment mechanism to adjust the sensor sampling interval and threshold according to changes in ambient temperature, time period, and equipment load.

3. A photovoltaic station fire prevention method according to claim 2, characterized in that: When the environmental data exceeds the threshold, the pre-alarm is triggered, including setting an edge computing module to be deployed in the on-site centralized control cabinet to perform local real-time judgment, and triggering a local alarm when the temperature rise curve change rate and the smoke concentration instantaneous jump exceed the threshold.

4. A photovoltaic station fire prevention method according to claim 3, characterized in that: The collecting of internal real-time data from photovoltaic modules through the monitoring system includes, while the monitoring system is collecting data, combining it with infrared thermal imaging inspection data to identify hidden faults in photovoltaic modules, and introducing spectrum analysis and transient current anomaly detection modules to capture short-term interference phenomena.

5. A photovoltaic station fire prevention method according to claim 4, characterized in that: The feature recognition and labeling of the processed real-time data includes establishing a pattern recognition rule set based on historical fault sample data, comparing and associating the processed real-time data with the pattern recognition rule set, finding a corresponding label in the pattern recognition rule set for each real-time data, and labeling the real-time data.

6. A photovoltaic station fire prevention method according to claim 5, characterized in that: Obtaining the diagnostic results includes inputting the labeled data into a machine learning model to construct a spatial distribution map to determine whether the anomaly is concentrated in one location or across multiple components; Based on the time series data in the data, a regression model is used to predict future trends, calculate the slope of the trend curve, fit the residual, and output the degradation rate of the photovoltaic module; Output a fault score based on the environmental information and electrical information in the input data; The structural pattern of the diagnostic result is that when the fault score is higher than a score threshold or the degradation rate of the photovoltaic module is higher than a rate threshold, it is determined to be abnormal; If the anomaly is concentrated in one location, it is considered a single-point anomaly. If the anomaly spans multiple components, it is considered a multi-point anomaly.

7. A photovoltaic station fire prevention method according to claim 6, characterized in that: The triggering of execution actions based on pre-alarm information and diagnosis results includes: when the diagnosis result is a single point abnormality, and the pre-alarm information is a temperature rise exceeding a threshold or a smoke concentration exceeding a threshold, the risk level is set to level one, and the execution action is a pop-up alarm on the operation and maintenance platform and a manual re-inspection; When the diagnosis result is a single-point abnormality and the pre-alarm information is that the temperature exceeds 110°C or the smoke concentration exceeds the threshold of 50%, the risk level is set to level 2, and the execution action is to activate the sound and light alarm and trigger the pre-isolation instruction; When the diagnosis result is multiple-point abnormality and the pre-alarm information is that the temperature rise exceeds the threshold or the smoke concentration exceeds the threshold, the risk level is set to level three, and the execution action is to immediately perform emergency isolation and fire extinguishing.

8. A photovoltaic station fire prevention system, using the photovoltaic station fire prevention method according to any one of claims 1 to 7, characterized in that: It includes environmental data acquisition module, edge computing module, electrical data acquisition module, transient current anomaly detection module, feature recognition module, diagnosis module and execution module; The environmental data acquisition module collects environmental data in real time based on sensors arranged on the photovoltaic modules; The edge computing module performs local real-time judgment and triggers a local alarm when the temperature rise curve change rate or smoke concentration instantaneous jump exceeds the threshold; The electrical data acquisition module is connected to the photovoltaic module through the monitoring system to collect internal real-time data; The transient current anomaly detection module is used to capture short-term interference phenomena; The feature recognition module processes the real-time data based on the feature processing algorithm, and performs feature recognition and labeling on the processed real-time data; The diagnostic module inputs the labeled data into the machine learning model of the monitoring platform to obtain a diagnostic result; The execution module triggers an execution action based on the pre-alarm information and the diagnosis result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the photovoltaic station fire prevention method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a photovoltaic station fire prevention method according to any one of claims 1 to 7 are implemented.

Images