Multi-source fusion fire sensing and emergency method of photovoltaic power station and middle station equipment

Abstract

The invention discloses a multi-source fusion fire sensing and emergency method of a photovoltaic power station and middle equipment, and relates to the technical field of fire detection. According to the method, middle equipment carries out hot spot scanning on a target photovoltaic power station to obtain a heat map data stream containing temperature values and space coordinates of all pixel points; based on a preset photovoltaic equipment normal working temperature model, calculating a temperature deviation value and a temperature rise rate between the temperature value of each pixel point and the corresponding equipment type and the theoretical working temperature under the current environment condition; marking the pixel points of which the temperature deviation values are greater than a preset deviation threshold value and the temperature rising rates are greater than a first preset rate threshold value as live hot points; calculating the smoke detection confidence of the image data of the live hot spot; calculating the current fluctuation variance of the current data of the circuit where the active hot point is located; when the temperature rise rate, the smoke detection confidence coefficient and the current fluctuation variance of the live hot point meet preset conditions, determining that the live hot point is a real fire point; and dispatching the fire-fighting unmanned aerial vehicle to execute a fire extinguishing task based on the coordinates of the real fire point.

Description

Technical Field

[0001] This application relates to the field of fire detection technology, and in particular to a multi-source integrated fire detection and emergency response method and central platform equipment for photovoltaic power plants. Background Technology

[0002] With the increasing global demand for clean energy, photovoltaic (PV) power generation, as an important form of renewable energy, is expanding at an unprecedented rate. PV power plants typically occupy vast areas, densely packed with thousands of PV modules, combiner boxes, inverters, and related electrical wiring. These devices operate outdoors for extended periods, constantly exposed to complex and variable natural factors such as high temperatures, strong sunlight, wind, sandstorms, rain, and snow. Simultaneously, the equipment itself generates significant heat under high load conditions, and electrical connections are at risk of aging and loosening. These factors combined contribute to a high fire risk in PV power plants.

[0003] Related technologies typically employ fixed sensors for fire detection in photovoltaic power plants. Specifically, fire detectors are deployed in critical areas of the plant, such as inverter rooms, combiner boxes, or surrounding areas. One common approach is to install heat detectors, which contain a thermistor. When the ambient temperature exceeds a preset static threshold (e.g., 85°C), the thermistor's physical properties change, triggering an alarm signal. Another approach is to install smoke detectors. These utilize the photoelectric effect; when smoke particles generated in the early stages of a fire enter the detector's labyrinthine cavity, they scatter or block light beams of specific wavelengths, causing a change in the intensity of light received by the light-receiving element. When this change reaches a preset value, the system detects a fire and issues an alarm.

[0004] However, when photovoltaic equipment is operating under strong sunlight and high load in summer, the normal operating temperature of the equipment surface and surrounding environment may approach or even exceed the static alarm threshold of the detector. Furthermore, photovoltaic power plants are mostly built in open areas or areas with strong winds and sandstorms. Strong winds can quickly disperse the small amount of smoke generated in the early stages of a fire, making its concentration insufficient to trigger the detector threshold, resulting in insufficient accuracy of fire detection in photovoltaic power plants using related technologies. Summary of the Invention

[0005] This application provides a multi-source integrated fire detection and emergency response method and central platform equipment for photovoltaic power plants, which can improve the accuracy of fire detection in photovoltaic power plants.

[0006] Firstly, a multi-source fusion fire detection and emergency response method for photovoltaic power plants is provided, applied to a central platform device. The method includes: the central platform device performing hotspot scanning on the target photovoltaic power plant using a thermal imaging device to obtain a thermal image data stream containing the temperature values ​​and spatial coordinates of each pixel; the central platform device calculating the temperature deviation and temperature rise rate between the temperature value of each pixel in the thermal image data stream and the theoretical operating temperature under the corresponding equipment type, current ambient temperature, and light intensity conditions, based on a preset normal operating temperature model for the photovoltaic equipment; and the central platform device marking pixels with temperature deviation values ​​greater than a preset deviation threshold and temperature rise rates greater than a first preset rate threshold as... The platform identifies a live hotspot. It acquires image data of the live hotspot using a visible light camera and calculates the smoke detection confidence level based on a smoke feature recognition algorithm. It also acquires current data of the circuit containing the live hotspot using a current sensor and calculates the current fluctuation variance. When at least two of the following conditions are met simultaneously: the temperature rise rate of the live hotspot exceeds a second preset rate threshold, the smoke detection confidence level exceeds a preset confidence threshold, and the current fluctuation variance exceeds a preset variance threshold, the platform identifies the live hotspot as a real fire point. Based on the three-dimensional spatial coordinates of the real fire point, the platform dispatches the nearest available firefighting drone to perform firefighting tasks.

[0007] By adopting the above technical solution, the middleware system incorporates a pre-set normal operating temperature model for photovoltaic equipment, making the judgment benchmark for temperature deviation dynamically changeable, thereby reducing false alarms from temperature sensors in hot weather. The middleware system integrates information from three different physical dimensions—thermodynamics, visible light, and electricity—for decision-making, employing a majority voting logic where at least two conditions must be met simultaneously. For example, even if strong winds cause smoke detection to fail, the middleware system can still confirm a fire as long as the temperature rises sharply and is accompanied by significant current fluctuations. Conversely, if only thermal imaging data is abnormal, while smoke and current data are normal, the middleware system can effectively suppress false alarms caused by interference from specular reflection in the thermal imager, thus improving the accuracy of fire detection in photovoltaic power plants.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, before the middleware device performs hot spot scanning on the target photovoltaic power station using a thermal imaging device to obtain a thermal image data stream containing the temperature values ​​and spatial coordinates of each pixel, the method further includes: the middleware device acquiring historical operating data of different types of equipment within the target photovoltaic power station under different environmental conditions; the middleware device performing dimensionality reduction processing on the historical operating data based on principal component analysis to extract the main environmental factors affecting equipment temperature and calculating the weight coefficients of each main environmental factor; the middleware device constructing a mathematical relationship model between equipment temperature and main environmental factors based on a multivariate nonlinear regression algorithm and weight coefficients; the middleware device optimizing the parameters of the mathematical relationship model based on cross-validation; and the middleware device storing the optimized mathematical relationship model according to equipment type to obtain a preset normal operating temperature model for photovoltaic equipment.

[0009] By adopting the above technical solution, the theoretical operating temperature of the middle platform equipment is more accurate based on the preset normal operating temperature model of photovoltaic equipment and the current ambient temperature and light intensity conditions.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the middleware device calculates the temperature deviation and temperature rise rate of the temperature value of each pixel in the heat map data stream based on a preset normal operating temperature model of photovoltaic equipment. Specifically, this includes: the middleware device determining the equipment type corresponding to the pixel based on the spatial coordinates of the pixel and a preset equipment layout database; the middleware device calling a mathematical relationship model from the preset normal operating temperature model of photovoltaic equipment based on the equipment type; the middleware device substituting the current environmental data into the mathematical relationship model to calculate the theoretical operating temperature of the pixel; the middleware device using the difference between the actual temperature value of the pixel and the theoretical operating temperature as the temperature deviation value; and the middleware device using a sliding time window to perform linear regression fitting on the historical temperature data of the pixel, and using the slope of the fitted line as the temperature rise rate.

[0011] By adopting the above technical solution, after determining the equipment type and calling the corresponding model, the platform device substitutes the currently collected environmental data (such as ambient temperature, light intensity, and other major environmental factors) into the model to calculate the theoretical operating temperature of the pixel under the current operating conditions. The resulting temperature deviation value can more realistically reflect whether the equipment has experienced undue overheating, thereby reducing the interference of normal temperature fluctuations caused by environmental changes on fire detection. Furthermore, by performing linear regression analysis on a series of historical temperature data of the pixel within a continuously moving time window, the slope of the fitted line is calculated as the temperature rise rate, thus reducing the impact of random noise on the temperature rise rate.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of the middleware device acquiring image data of active hotspots through a visible light camera and calculating the smoke detection confidence of the image data based on a smoke feature recognition algorithm specifically includes: the middleware device acquiring image data of active hotspots based on a visible light camera; the middleware device identifying dynamic regions with smoke diffusion characteristics based on the time change characteristics of pixel grayscale values ​​in the image data; the middleware device calculating the texture complexity and edge gradient change rate of pixels in the dynamic region; the middleware device marking dynamic regions with texture complexity less than a preset complexity threshold and edge gradient change rate conforming to the smoke diffusion pattern as suspected smoke regions; and the middleware device using the similarity between the color distribution characteristics and transparency change characteristics of the suspected smoke regions and a preset smoke feature template as the smoke detection confidence.

[0013] By adopting the above technical solution, dynamic regions with texture complexity less than a preset complexity threshold and edge gradient change rate conforming to the smoke diffusion pattern are marked as suspected smoke regions. Then, the similarity between the color distribution features and transparency change features of the suspected smoke regions and the preset smoke feature template is used as the smoke detection confidence score, thereby improving the accuracy of smoke detection confidence score.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, when at least two of the following conditions are met simultaneously: the temperature rise rate of the active hotspot is greater than a second preset rate threshold, the smoke detection confidence level is greater than a preset confidence threshold, and the current fluctuation variance is greater than a preset variance threshold, the step of the intermediate platform device confirming the active hotspot as a real fire point specifically includes: the intermediate platform device establishing a multi-dimensional feature vector for the active hotspot, the multi-dimensional feature vector containing at least the values ​​of three dimensions: temperature rise rate, smoke detection confidence level, and current fluctuation variance; the intermediate platform device calculating the number of dimensions in the multi-dimensional feature vector that are greater than a preset dimension threshold; when the number of dimensions is greater than or equal to two, the intermediate platform device calculating the threshold exceedance value of each exceeding threshold dimension, the threshold exceedance value being the value of the exceeding threshold dimension divided by the preset dimension threshold and then minus one; the intermediate platform device weighted summing of the threshold exceedance values ​​of each dimension based on preset weighting coefficients to obtain a comprehensive fire risk score; when the comprehensive fire risk score is greater than a preset risk score threshold, the intermediate platform device determining the active hotspot as a real fire point.

[0015] By adopting the above technical solution, the platform equipment calculates the threshold exceedance value using the formula (value / threshold) - 1. This transforms the original measurement value of each threshold exceedance dimension into a dimensionless index representing the relative severity of its exceedance, providing quantitative input for subsequent comprehensive assessment. The platform equipment then weights and sums the threshold exceedance values ​​for each dimension using preset weighting coefficients. The resulting comprehensive fire risk score is a highly condensed risk index that incorporates multi-dimensional information and has been calibrated by expert knowledge, making the identification of actual fire points based on the comprehensive fire risk score more accurate.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, before the middleware device performs a weighted summation of the threshold exceedance values ​​of each dimension based on preset weight coefficients to obtain a comprehensive fire risk score, the method further includes: the middleware device establishing a circular analysis area with a preset radius centered on the active hotspot, and acquiring the temperature data of all pixels within the analysis area; the middleware device calculating the temperature difference between the active hotspot and other pixels within the circular analysis area, and counting the number of pixels whose temperature difference is greater than a preset difference threshold; the middleware device using the ratio of the number of pixels to the total number of pixels in the analysis area as the thermal anomaly pixel density; the middleware device determining whether the thermal anomaly pixel density is less than a preset density threshold; if so, the middleware device determining the active hotspot as an isolated hotspot, and setting the weight coefficient of the temperature rise rate dimension to a first preset weight value, wherein the first preset weight value is greater than the preset weight coefficient of the temperature rise rate dimension;

[0017] If not, the middleware determines that the active hotspot is a non-isolated hotspot and sets the weight coefficient of the temperature rise rate dimension to the second preset weight value, which is less than the preset weight coefficient of the temperature rise rate dimension.

[0018] By adopting the above technical solution, before calculating the comprehensive risk score, the platform equipment first performs thermal imaging analysis on the microenvironment surrounding the active hotspot. It determines whether the active hotspot is isolated or non-isolated by calculating the pixel density of thermal anomalies. Isolated hotspots typically correspond to a breakdown or poor contact of a component inside the equipment; the fault point is highly concentrated, and its rapid temperature rise is a signal that a fire is about to break out. In this case, the weighting coefficient for the temperature rise rate is set to a higher first preset weighting value, thereby improving the accuracy of identifying such potential risk points. Conversely, non-isolated hotspots mean that the thermal anomaly has already shown a patchy distribution, which may simply be due to large-area sunlight reflection, shading recovery, or a fire that has already begun to spread. In this case, the weighting coefficient for the temperature rise rate is lowered to a second preset weighting value, allowing information from other dimensions such as smoke and current to play a more significant role in the decision-making process. Through this dynamic adjustment of weighting coefficients, the platform equipment can more accurately identify early fires of different causes and forms, thereby improving the accuracy of fire detection in photovoltaic power plants.

[0019] In conjunction with some embodiments of the first aspect, in some embodiments, the step of the central platform device scheduling the nearest available firefighting drone to perform firefighting tasks based on the three-dimensional spatial coordinates of the actual fire point specifically includes: the central platform device determining the nearest available firefighting drone based on the three-dimensional spatial coordinates of the actual fire point; the central platform device calculating the optimal flight waypoint sequence based on the current position of the available firefighting drone and the three-dimensional spatial coordinates of the actual fire point, wherein the waypoint sequence avoids preset no-fly zones; the central platform device determining the type and amount of extinguishing agent to be deployed based on the temperature value and equipment type of the actual fire point, and generating firefighting operation command parameters; and the central platform device controlling the available firefighting drone to perform firefighting tasks based on the optimal flight waypoint sequence and firefighting operation command parameters.

[0020] By adopting the above technical solution, the efficiency of scheduling the nearest available firefighting drone to perform firefighting tasks based on the three-dimensional spatial coordinates of the real fire point has been improved.

[0021] Secondly, embodiments of this application provide a middleware device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the middleware device to perform the method described in the first aspect and any possible implementation thereof.

[0022] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a middleware device, cause the middleware device to execute the method described in the first aspect and any possible implementation thereof.

[0023] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a middleware device, cause the middleware device to perform the method described in the first aspect and any possible implementation thereof.

[0024] It is understood that the platform device provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0025] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0026] 1. Because the middleware system incorporates a pre-defined normal operating temperature model for photovoltaic equipment, the judgment benchmark for temperature deviation values ​​is dynamically changing, thereby reducing false alarms from temperature sensors in hot weather. The middleware system integrates information from three different physical dimensions—thermodynamics, visible light, and electricity—for decision-making, employing a majority voting logic where at least two conditions must be met simultaneously. For example, even if strong winds cause smoke detection to fail, the middleware system can still confirm a fire as long as the temperature rises sharply and is accompanied by significant current fluctuations. Conversely, if only thermal imaging data is abnormal, while smoke and current data are normal, the middleware system can effectively suppress false alarms caused by interference from specular reflections on the thermal imager, thus improving the accuracy of fire detection in photovoltaic power plants.

[0027] 2. Because the platform equipment calculates the threshold exceedance value using the formula (value / threshold) - 1, it transforms the original measurement value of each threshold-exceeding dimension into a dimensionless index representing the relative severity of exceeding the threshold, providing quantitative input for subsequent comprehensive assessment. The platform equipment uses preset weighting coefficients to weight and sum the threshold exceedance values ​​for each dimension, resulting in a comprehensive fire risk score that is a highly condensed multi-dimensional information index calibrated by expert knowledge. This makes the identification of actual fire points based on the comprehensive fire risk score more accurate.

[0028] 3. Before calculating the comprehensive risk score, the platform equipment first performs thermal imaging analysis on the microenvironment surrounding the active hotspot. It determines whether the active hotspot is isolated or non-isolated by calculating the pixel density of thermal anomalies. Isolated hotspots typically correspond to a breakdown or poor contact of a component within the equipment; the fault points are highly concentrated, and a rapid temperature rise is a signal of an impending fire. In this case, the weighting coefficient for the rate of temperature rise is set to a higher first preset weighting value, thereby improving the accuracy of identifying such potential risk points. Conversely, non-isolated hotspots indicate that the thermal anomaly has already shown a patchy distribution, which may simply be due to large-area sunlight reflection, shading recovery, or a fire that has already begun to spread. In this case, the weighting coefficient for the rate of temperature rise is lowered to a second preset weighting value, allowing information from other dimensions such as smoke and current to play a more significant role in the decision-making process. Through this dynamic adjustment of weighting coefficients, the platform equipment can more accurately identify early fires of different causes and forms, thereby improving the accuracy of fire detection in photovoltaic power plants. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a multi-source fusion fire detection and emergency response method for a photovoltaic power station, as described in an embodiment of this application.

[0030] Figure 2 This is another flowchart illustrating a multi-source fusion fire detection and emergency response method for a photovoltaic power station, as described in an embodiment of this application.

[0031] Figure 3 This is a schematic diagram of the physical device structure of the platform equipment in the embodiments of this application. Detailed Implementation

[0032] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0033] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0034] This application provides a multi-source integrated fire detection and emergency response method and central platform equipment for photovoltaic power plants, which can improve the accuracy of fire detection in photovoltaic power plants.

[0035] Please see Figure 1 This is a flowchart illustrating a multi-source fusion fire detection and emergency response method for a photovoltaic power station in an embodiment of this application.

[0036] S101, the middle platform equipment performs hot spot scanning on the target photovoltaic power station through thermal imaging equipment, and obtains a thermal image data stream containing the temperature value and spatial coordinates of each pixel.

[0037] The central platform device is a centralized data processing and control unit, such as a server or cloud computing platform deployed locally at the power plant. Its function is to collect data from front-end devices, run analysis algorithms, and issue control commands. The thermal imaging device is a non-contact temperature measurement device capable of detecting the infrared radiation of an object and converting it into a temperature image. The target photovoltaic power plant is the power plant to which this method is applied, containing photovoltaic modules, combiner boxes, and inverters. Hot spot scanning is a systematic inspection operation in which the central platform device controls the thermal imaging device to detect the infrared temperature of equipment within the power plant. The thermal image data stream is the time-series data generated by the thermal imaging device during scanning. Its core is a series of data frames, each containing a thermal image and its metadata. The pixel temperature value is the temperature reading of the surface of the object being measured corresponding to a single pixel on the thermal image. Spatial coordinates are the coordinate data used to calibrate the position of the object corresponding to that pixel in three-dimensional physical space.

[0038] Specifically, this step is the data acquisition phase of the fire detection process. During execution, the central platform equipment issues scanning commands to one or more thermal imaging devices according to a preset inspection plan (e.g., at fixed time intervals or dynamically adjusted based on the real-time power generation of the photovoltaic power station). These thermal imaging devices can be pan-tilt cameras installed in fixed locations or mounted on mobile platforms such as drones or ground inspection robots. After receiving the command, the thermal imaging devices perform infrared detection on the designated area, generating a thermal map containing the surface temperature distribution of objects. The temperature value of each pixel in the thermal map can be associated with a three-dimensional spatial coordinate. For fixed equipment, this association can be achieved by pre-calibrating the camera position and attitude, establishing a transformation relationship between the pixel coordinate system and the global coordinate system of the power station. For mobile equipment, the central platform equipment needs to integrate the device's own position and attitude data, and use real-time localization and mapping or photogrammetry algorithms to calculate the coordinates of each pixel in the three-dimensional space of the target photovoltaic power station for the thermal map. Finally, these data frames, which integrate timestamps, pixel temperature matrices, and corresponding spatial coordinate matrices, constitute a thermal map data stream that is continuously sent to the central platform equipment for analysis.

[0039] In some embodiments, before the middleware device performs hot spot scanning on the target photovoltaic power station using a thermal imaging device to obtain a thermal image data stream containing the temperature values ​​and spatial coordinates of each pixel, the method further includes a method for constructing a preset normal operating temperature model for the photovoltaic equipment. Specifically, the middleware device first acquires historical operating data of different types of equipment within the target photovoltaic power station under different environmental conditions. This process can be accomplished by controlling the thermal imaging device to perform multiple hot spot scans under different seasons, time periods, and weather conditions, while simultaneously recording environmental factor data such as ambient temperature, light intensity, and wind speed.

[0040] Subsequently, the platform equipment performs dimensionality reduction processing on the collected historical operational data based on principal component analysis. The specific steps are as follows: Due to the different dimensions and orders of magnitude of data such as ambient temperature and light intensity, the platform equipment first performs Z-score standardization on the original environmental factor data to eliminate the influence of dimensions. Then, it calculates the correlation coefficient matrix between the standardized environmental factors to characterize the correlation between the factors. Next, it solves for the eigenvalues ​​and corresponding eigenvectors of the correlation coefficient matrix. The eigenvalues ​​are sorted from largest to smallest, and the cumulative variance contribution rate is calculated. The principal components corresponding to the top k eigenvalues ​​whose cumulative variance contribution rate reaches a preset threshold (e.g., 85%) are selected. At the same time, the absolute values ​​of the loadings of each original environmental factor on these principal components are analyzed, and the original factors whose sum of absolute loadings is greater than the pre-screening threshold are selected as the main environmental factors (denoted as ). ).

[0041] The middleware platform normalizes the weighted average of the loading coefficients of each major environmental factor in each selected principal component (weighted by the eigenvalue contribution rate), and uses this as the weight coefficient of that major environmental factor (denoted as ). This weighting coefficient objectively reflects the importance of the environmental factor on the temperature change of the equipment.

[0042] Next, the platform equipment constructs a mathematical relationship model between equipment temperature and major environmental factors based on a multivariate nonlinear regression algorithm and the weight coefficients calculated above. The platform equipment then introduces the product of the major environmental factors and their weight coefficients as a modified input variable into the regression model to improve the model's ability to fit environmental sensitivity.

[0043] For example, let y be the actual operating temperature of a certain type of photovoltaic equipment (such as photovoltaic modules), and let the main environmental factors be... The weighting coefficient is The mathematical model formula for this photovoltaic device is as follows:

[0044]

[0045] Where Y is the theoretical predicted operating temperature of a specific type of photovoltaic equipment;

[0046] For constant terms, Let be the coefficient of the linear term of the t-th major environmental factor. Let be the coefficient of the interaction term between the t-th and s-th major environmental factors;

[0047] t is the weighted primary environmental factor, reflecting its weight in relation to temperature;

[0048] For the random error term, satisfying , (constant).

[0049] Then, the model parameters are solved using the least squares method. , and This minimizes the sum of squared residuals between the actual temperature y and the predicted temperature y, i.e.:

[0050]

[0051] in, Let i be the actual operating temperature of the device for the i-th sample. This corresponds to the predicted theoretical operating temperature.

[0052] Then, the middleware platform uses cross-validation to optimize the parameters of the constructed mathematical relationship model. For example, by dividing the historical dataset into training and validation sets, the internal parameters of the model are repeatedly adjusted and its prediction accuracy is tested on the validation set until the model's generalization ability reaches the desired level. Finally, the middleware platform stores the optimized mathematical relationship model according to different equipment types such as photovoltaic modules, combiner boxes, and inverters, thereby obtaining a preset normal operating temperature model for photovoltaic equipment used for subsequent real-time comparisons.

[0053] S102. The middleware device calculates the temperature deviation and temperature rise rate between the temperature value of each pixel in the heat map data stream and the theoretical operating temperature under the corresponding equipment type, current ambient temperature and light intensity conditions, based on the preset normal operating temperature model of photovoltaic equipment.

[0054] The preset normal operating temperature model for photovoltaic equipment is a model built in S101. The theoretical operating temperature is the temperature value that the equipment should have under fault-free conditions, calculated using this model. The temperature deviation value is the difference between the actual measured temperature of the same pixel at the same moment and its theoretical operating temperature, used to quantify the degree of temperature anomaly. The temperature rise rate is a physical quantity describing how fast the temperature of a certain pixel changes over time, used to capture the dynamic trend of temperature change.

[0055] Specifically, this step is executed by the middleware device after receiving the heat map data stream generated by S101. First, the middleware device needs to determine the photovoltaic equipment type corresponding to each pixel. This is achieved by matching the pixel's three-dimensional spatial coordinates with a pre-stored local power plant equipment layout database. This database records in detail the three-dimensional position, size, and type information of all equipment within the power plant (such as each photovoltaic module and each combiner box). Through spatial coordinate comparison, the middleware device can assign pixels to specific equipment instances and their types. Second, based on the determined equipment type, the middleware device calls the corresponding mathematical relationship model from a pre-set photovoltaic equipment normal operating temperature model library. Simultaneously, the middleware device acquires current environmental data, such as real-time ambient temperature and light intensity values ​​from a weather station deployed at the power plant site. Then, the middleware device uses this real-time environmental data as input, substituting it into the called mathematical relationship model for calculation, thereby obtaining the theoretical operating temperature of the pixel under the current environmental conditions. Next, the middleware device subtracts the actual measured temperature value of the pixel (from the heat map data stream) from the calculated theoretical operating temperature; the difference is the temperature deviation value of the pixel. A positive deviation value means the actual temperature is higher than the theoretical normal temperature. Finally, to calculate the rate of temperature rise, the platform device needs to utilize the historical temperature data of that pixel. The platform device maintains a fixed-length time window (e.g., all temperature readings within the past 5 minutes) for each pixel or small area. Using a sliding time window approach, a linear regression is performed on the sequence of historical temperature data points within the window to find a straight line that best describes the trend of these data points. The slope of this fitted line represents the amount of temperature change per unit time and is defined as the current rate of temperature rise for that pixel.

[0056] In some embodiments, the method for calculating the temperature deviation and temperature rise rate between the temperature value of each pixel in the heat map data stream and the theoretical operating temperature under the corresponding equipment type, current ambient temperature, and light intensity conditions, based on a preset photovoltaic equipment normal operating temperature model, further includes: the platform device first determines the equipment type corresponding to the pixel based on the spatial coordinates of the pixel and a preset equipment layout database. Specifically, the platform device retrieves the three-dimensional spatial coordinates (X, Y, Z) of each pixel parsed from the heat map data stream from a pre-established three-dimensional digital twin model of the power station. This model contains the geometric boundary information and type identifiers of all equipment. By determining whether the coordinates of a pixel fall within the geometric bounding box of a certain equipment, its equipment type can be determined, such as a monocrystalline silicon photovoltaic module or a string inverter. Secondly, the platform device calls a mathematical relation model from the preset photovoltaic equipment normal operating temperature model based on the equipment type. For example, if it is determined that the pixel belongs to a monocrystalline silicon photovoltaic module, the platform device loads a temperature prediction model specifically for that type of module from the model library. Then, the platform device substitutes the current environmental data into the mathematical relation model to calculate the theoretical operating temperature of the pixel. For example, the current light intensity of 800 W / m² and ambient temperature of 30°C, obtained from a weather station, are used as input variables into the called component temperature model to calculate the theoretical operating temperature, for example, 55°C. Next, the platform device uses the difference between the actual temperature value of a pixel and the theoretical operating temperature as the temperature deviation value. If the heat map shows that the actual temperature of the pixel is 65°C, then its temperature deviation value is 65°C - 55°C = 10°C. Finally, the platform device uses a sliding time window to perform linear regression fitting on the historical temperature data of the pixel, and uses the slope of the fitted line as the temperature rise rate. Specifically, the platform device extracts all temperature records of the pixel over the past 300 seconds (e.g., one data point every 10 seconds, for a total of 30 points), forming a time series. The platform device applies the least squares method to perform linear fitting on this set of data points, obtaining a straight line of the form T = a*t + b, where the coefficient a is the average temperature rise rate during that time period. This value a is used as the temperature rise rate at the current moment.

[0057] S103. The middleware device marks pixels whose temperature deviation value is greater than the preset deviation threshold and whose temperature rise rate is greater than the first preset rate threshold as active hotspots.

[0058] The preset deviation threshold is a pre-defined temperature difference value, such as 15°C, used to define whether the deviation of a pixel's measured temperature from its theoretical normal temperature reaches a level requiring attention. The first preset rate threshold is a pre-defined temperature change rate value, such as 2°C / minute, used to determine whether a pixel's temperature is continuously rising at an abnormal rate. A hot spot is an anomaly point initially screened as having potential fire risk.

[0059] Specifically, in S102, after calculating the temperature deviation value and temperature rise rate for each pixel, the middleware device iterates through all pixels and performs a two-condition logical judgment for each pixel. The first condition is whether the temperature deviation value of the pixel exceeds a preset deviation threshold. This threshold is determined based on expert experience, equipment safety specifications, or statistical analysis of historical data, aiming to filter out minor temperature exceedances caused by model errors or normal operating condition fluctuations. The second condition is whether the temperature rise rate of the pixel exceeds a first preset rate threshold. This threshold is used to distinguish between normal temperature fluctuations of the equipment (such as rapid temperature rise caused by the reappearance of sunlight after cloud cover) and continuous, accelerated temperature rises caused by faults (such as internal short circuits). Only when a pixel meets both conditions simultaneously, i.e., its static temperature anomaly and dynamic temperature rise trend both reach the warning level, will the middleware device mark it as a live hotspot. This marking action can be done by associating a specific status tag for the pixel in the middleware device's memory database and recording the marking time and related data (such as the deviation value and rate value at that time).

[0060] Optionally, the preset deviation threshold, the first preset rate threshold, and all preset thresholds involved in subsequent steps can be determined through statistical analysis of a large amount of historical data on normal and faulty operating conditions. For example, the Receiver Operating Characteristic (ROC) curve analysis method can be used to maximize the detection rate within an acceptable false alarm rate, thereby setting the thresholds more accurately. For instance, by analyzing a large amount of historical datasets containing normal and known early fire conditions, ROC curves can be plotted and analyzed under each threshold, and the threshold point corresponding to the maximum Youden index can be selected to achieve the optimal balance between the false alarm rate and the missed alarm rate.

[0061] S104. The middleware acquires image data of the active hotspot through a visible light camera and calculates the smoke detection confidence of the image data based on a smoke feature recognition algorithm.

[0062] Visible light cameras are conventional cameras capable of capturing light reflected or emitted by objects within the visible spectrum (approximately 400-760 nanometers) and generating color or black-and-white images. They are typically deployed in conjunction with thermal imaging equipment or integrated into the same gimbal / drone / unmanned vehicle. Smoke feature recognition algorithms are computer vision algorithms specifically designed to detect smoke in images or videos. They determine the presence of smoke by analyzing features such as pixel color, texture, shape, and dynamic changes. The smoke detection confidence level is a value between 0 and 1 (or 0% to 100%), quantifying the algorithm's certainty that smoke is indeed present in the image; a higher value indicates a greater probability that smoke is present.

[0063] Specifically, once the central platform device marks one or more active hotspots in S103, it immediately initiates visual information acquisition and analysis for those hotspots. First, the central platform device acquires image data of the active hotspots using a visible light camera. Once a hotspot is marked, the central platform device immediately queries its three-dimensional coordinates and instructs the nearest high-definition visible light camera to aim its lens at that coordinate point and perform optical zoom to obtain a clear video stream. Second, the central platform device identifies dynamic regions with smoke diffusion characteristics based on the temporal variation characteristics of pixel grayscale values ​​in the image data. Specifically, the central platform device performs inter-frame differencing or background subtraction on consecutive video frames to separate the moving foreground in the scene. For these foreground regions, the central platform device further analyzes the variation pattern of their pixel grayscale values ​​over time, searching for regions that conform to the gradual grayscale change pattern of smoke from dense to light and then diffuses, initially marking them as dynamic regions. Then, the central platform device calculates the texture complexity and edge gradient change rate of the pixels in the dynamic regions. For identified dynamic regions, the platform uses algorithms such as Local Binary Pattern (LBP) or Gray-Level Co-occurrence Matrix (GLCM) to quantify the complexity of their texture. Smoke textures are typically smooth and have low complexity. Simultaneously, the platform calculates the gradient changes of the region's edges across consecutive frames. Real smoke edges gradually blur and the gradient decreases, while the motion of rigid objects usually maintains sharp edges. Next, the platform marks dynamic regions with texture complexity below a preset complexity threshold and edge gradient change rates conforming to smoke diffusion patterns as suspected smoke regions. Only when a dynamic region simultaneously satisfies both the conditions of smooth internal textures and blurred external edges is it considered highly suspected smoke. Finally, the platform uses the similarity between the color distribution features and transparency change features of suspected smoke regions and a preset smoke feature template as the smoke detection confidence score. The platform extracts the color histogram of suspected smoke regions and compares it with the typical color distribution (usually biased towards achromatic gray-white tones) of a large number of real smoke samples stored in the database. Simultaneously, by analyzing the change in the degree of occlusion of the area on its background over time, the transparency change is evaluated to determine whether it conforms to the semi-transparent characteristics of smoke. The platform device then fuses the evaluation results of multiple features, such as color similarity and transparency change conformity, using a weighted function or a pre-trained classifier (such as a support vector machine, SVM), to obtain a final smoke detection confidence score.

[0064] S105. The middleware obtains the current data of the circuit where the hot spot is located through the current sensor and calculates the current fluctuation variance of the current data.

[0065] A current sensor is an electronic component installed in a circuit to measure real-time current values. Examples include Hall effect sensors or shunts. Its measurement data is typically integrated into the monitoring and data acquisition system of a photovoltaic power plant. Current fluctuation variance is a statistical indicator used to quantify the degree to which current data deviates from its average value within a certain time window. It reflects the stability of the current. Abnormal arcing or poor contact usually causes severe and irregular current fluctuations, thus increasing the variance.

[0066] Specifically, after marking the active hotspot in S103, the central platform device first maps the physical location of the active hotspot onto the electrical topology. The central platform device internally stores a detailed digital twin database of the power plant. This database not only contains the three-dimensional spatial coordinates of all equipment but also records their electrical connections (e.g., which component belongs to which string, and which string connects to which branch of which combiner box). The central platform device uses the three-dimensional coordinates of the active hotspot to query this database, locate its corresponding circuit, and identify the unique identifier of the current sensor responsible for monitoring that circuit. Subsequently, the central platform device sends a command to the SCADA system, requesting to obtain the high-frequency current data of that specific sensor over a recent period (e.g., the past minute). After obtaining this time-series current data, the central platform device calculates the variance of the dataset. The calculation process can be as follows: first, calculate the arithmetic mean of all current readings within the time window; then, calculate the square of the difference between each current reading and its mean; finally, calculate the average of these squared differences. This result is the current fluctuation variance. This variance value provides a key quantitative indicator from an electrical perspective for determining whether a device has malfunctioned.

[0067] S106. When at least two of the following conditions are met simultaneously: the temperature rise rate of the active hot spot is greater than the second preset rate threshold, the smoke detection confidence level is greater than the preset confidence threshold, and the current fluctuation variance is greater than the preset variance threshold, the central platform device will confirm the active hot spot as a real fire point.

[0068] The second preset rate threshold is a more stringent temperature rise rate threshold than the first preset rate threshold in S103, for example, 5℃ / minute. This indicates that the temperature has entered a runaway rise stage, and the fire risk is extremely high. The preset confidence threshold is a high confidence benchmark for smoke detection, for example, 0.9 (or 90%). The preset variance threshold is an upper limit set according to the current fluctuation range under normal operating conditions. Exceeding this value indicates that there is a significant electrical abnormality in the circuit (such as arc discharge).

[0069] Specifically, this step is the final decision-making stage of the fire detection process. The central platform device gathers judgment criteria from three independent information sources: temperature dynamics (temperature rise rate) obtained from thermal imaging analysis, visual representation (smoke detection confidence level) obtained from visible light analysis, and electrical status (current fluctuation variance) obtained from electrical sensor analysis. The central platform device continuously monitors the values ​​of these three indicators for each active hot spot. Then, the central platform device applies a two-out-of-three majority voting logic for final confirmation. That is, when at least two of the following conditions are met simultaneously: the temperature rise rate of the active hot spot is greater than a second preset rate threshold, the smoke detection confidence level is greater than a preset confidence threshold, and the current fluctuation variance is greater than a preset variance threshold, the central platform device confirms the active hot spot as a real fire point.

[0070] S107. The central platform equipment schedules the nearest available firefighting drone to perform firefighting tasks based on the three-dimensional spatial coordinates of the real fire point.

[0071] The actual fire point is the fire location ultimately confirmed in S106. "Available status" means the firefighting drone is in a standby state, i.e., its battery is fully charged, its extinguishing agent is sufficient, its communication is normal, and there are no system malfunctions.

[0072] Specifically, once S106 confirms the actual fire location, the central platform immediately triggers an automatic fire suppression response. First, the central platform accesses its managed fire resource database, which is updated in real-time with the status of all fire suppression drones, including their landing pad locations, battery levels, extinguishing agent types, and remaining quantities. The central platform then filters out all available drones. Next, using the 3D spatial coordinates of the actual fire location and the landing pad coordinates of each available drone, it calculates the straight-line distance or optimal flight path sequence for each drone to the fire location. This flight path sequence avoids pre-defined no-fly zones and can be adjusted based on the approach direction (e.g., considering wind direction, approaching from the upwind direction). The central platform dispatches the nearest drone as the primary execution unit based on the 3D spatial coordinates of the actual fire location to minimize response time. Then, the central platform generates a fire suppression mission instruction package, which includes the 3D spatial coordinates of the actual fire location, the flight path sequence, and the extinguishing agent deployment mode. Upon receiving the instruction, the drone automatically unlocks and takes off, flying towards the target location using its built-in navigation system and flight control algorithm. Upon reaching the airspace above the fire, the drone uses its onboard sensors (such as a high-definition camera or thermal imager) to perform final positioning and targeting of the fire, and then executes the fire extinguishing agent deployment. After completing the mission, the drone automatically returns to base and transmits the mission results back to the central control system. Upon receiving the mission completion report, the central control system continues to monitor the fire area using thermal imaging equipment to confirm whether the fire has been successfully extinguished.

[0073] In the above embodiments, the middleware system incorporates a preset normal operating temperature model for photovoltaic equipment, making the judgment benchmark for temperature deviation dynamically changeable, thereby reducing false alarms from temperature sensors in hot weather. The middleware system integrates information from three different physical dimensions—thermodynamics, visible light, and electricity—for decision-making, employing a majority voting logic where at least two conditions must be met simultaneously. For example, even if strong winds cause smoke detection to fail, the middleware system can still confirm a fire as long as the temperature rises sharply and is accompanied by significant current fluctuations. Conversely, if only thermal imaging data is abnormal, while smoke and current data are normal, the middleware system can effectively suppress false alarms caused by interference from specular reflection in the thermal imager, thereby improving the accuracy of fire detection in photovoltaic power plants.

[0074] However, while the above embodiments can determine the existence of anomalies in multiple dimensions, they struggle to quantify and differentiate the severity of these anomalies. For example, consider two scenarios: in scenario A, both the rate of temperature rise and the variance of current fluctuations just exceed their respective preset thresholds. In scenario B, the rate of temperature rise far exceeds the thresholds, accompanied by high-confidence smoke and severe current fluctuations. Based on the logic of the above embodiments, these two scenarios with drastically different risk levels could both be identified as real fire points, which to some extent limits the precision of fire situation awareness and the accuracy of decision-making. To more accurately quantify fire risk and differentiate combinations of anomalies of different severity, this application proposes the following method.

[0075] Please see Figure 2 This is another flowchart illustrating a multi-source fusion fire detection and emergency response method for a photovoltaic power station in an embodiment of this application.

[0076] S201. The central platform equipment performs hot spot scanning on the target photovoltaic power station using thermal imaging equipment, and obtains a thermal image data stream containing the temperature value and spatial coordinates of each pixel.

[0077] S202. The middleware device calculates the temperature deviation and temperature rise rate between the temperature value of each pixel in the heat map data stream and the theoretical operating temperature under the corresponding equipment type, current ambient temperature and light intensity conditions, based on the preset normal operating temperature model of photovoltaic equipment.

[0078] S203. The middleware device marks pixels whose temperature deviation value is greater than the preset deviation threshold and whose temperature rise rate is greater than the first preset rate threshold as active hotspots.

[0079] S204. The middleware acquires image data of live hotspots through visible light camera equipment and calculates the smoke detection confidence of the image data based on the smoke feature recognition algorithm.

[0080] S205. The middleware obtains the current data of the circuit where the hot spot is located through the current sensor and calculates the current fluctuation variance of the current data.

[0081] Step S201 is similar to step S101, step S202 is similar to step S102, step S203 is similar to step S103, step S204 is similar to step S104, and step S205 is similar to step S105, so they will not be repeated here.

[0082] S206, The middleware establishes multi-dimensional feature vectors for active hotspots.

[0083] The multidimensional feature vector contains at least three dimensions: temperature rise rate, smoke detection confidence, and current fluctuation variance.

[0084] Specifically, after the middleware executes S205 (calculate current variance), for a specific active hotspot, the middleware has acquired values ​​for three dimensions: temperature rise rate, smoke detection confidence level, and current fluctuation variance. The core task of this step is to integrate these three scalar data points, which come from different sources and have different physical meanings, into a unified mathematical representation. The middleware creates a data object, or multidimensional feature vector, for the active hotspot in memory. It uses the calculated temperature rise rate, smoke detection confidence level, and current fluctuation variance as the three components of this vector, filling them in a predetermined order (e.g., (temperature rise rate, smoke detection confidence level, current fluctuation variance)). In this way, each active hotspot is transformed into a point in a three-dimensional feature space, whose coordinates directly reflect the degree of anomaly at that point in different dimensions.

[0085] S207. The middleware device calculates the number of dimensions in the multidimensional feature vector that are greater than the preset dimension threshold.

[0086] Among them, the preset dimension threshold refers to a set of thresholds corresponding to each dimension of the multidimensional feature vector. Each threshold in the set is a pre-set critical value used to determine whether the value of the corresponding dimension has entered an abnormal state. For example, the set specifically includes a second preset rate threshold, a preset reliability threshold, and a preset variance threshold.

[0087] Specifically, after the middleware constructs a multi-dimensional feature vector for the active hotspot, it extracts the feature vector and loads the corresponding dimension threshold vector from the system configuration. Then, the middleware compares each dimension of the vector one by one: First, it checks if the temperature rise rate is greater than a second preset rate threshold. Second, it checks if the smoke detection confidence level is greater than a preset confidence threshold. Finally, it checks if the current fluctuation variance is greater than a preset variance threshold. The middleware counts the number of true results in these three comparison operations. For example, if the temperature of an active hotspot rises extremely rapidly and the circuit current is extremely unstable, but no obvious smoke has been detected, then both the temperature rise rate and current fluctuation variance dimensions exceed the threshold, while the smoke detection confidence level dimension does not. In this case, the calculated number of dimensions is 2.

[0088] S208. When the number of dimensions is greater than or equal to two, the middleware device calculates the threshold exceedance value for each threshold dimension.

[0089] In this context, the "threshold exceeding dimension" refers to the dimension whose value is determined to be greater than its corresponding preset dimension threshold in S207. The threshold exceeding degree value is a quantitative indicator used to measure the relative magnitude or severity of how much the actual measured value of a dimension exceeds its preset threshold. In an exemplary embodiment, the threshold exceeding degree value can be the value of the threshold exceeding dimension divided by the preset dimension threshold and then minus one.

[0090] Specifically, when the calculation results of S207 indicate that a live hotspot exhibits anomalies in at least two dimensions (e.g., both the temperature rise rate and smoke confidence exceed the limit), the platform device iterates through all dimensions marked as exceeding the threshold in the multidimensional feature vector of that live hotspot. For each such dimension, the platform device performs a calculation to obtain its threshold exceedance value. For example, assuming the second preset rate threshold is 5℃ / minute, and the actual measured temperature rise rate of the current live hotspot is 7℃ / minute, then the threshold exceedance value for its temperature rise rate dimension is (7 / 5)-1=0.4. Similarly, if the preset confidence threshold is 0.9, and the actual measured smoke detection confidence is 0.99, then the threshold exceedance value for its smoke dimension is (0.99 / 0.9)-1=0.1.

[0091] S209. The central platform equipment calculates a weighted sum of the threshold exceedance values ​​of each dimension based on preset weight coefficients to obtain a comprehensive fire risk score.

[0092] The preset weight coefficients refer to a predefined set of values. Each coefficient in the set corresponds to a dimension of the multidimensional feature vector (such as the rate of temperature rise, the confidence level of smoke detection, and the variance of current fluctuation), which is used to represent the relative importance of that dimension in the comprehensive assessment of fire risk. The sum of all weight coefficients is usually 1.

[0093] Specifically, after calculating the exceedance values ​​for all dimensions exceeding the threshold, the platform device loads a preset weight coefficient vector from the configuration library. These weight coefficients are pre-set by fire safety experts and data scientists based on extensive historical fire case analysis and physical model simulations. For example, in photovoltaic power plant fires, electrical arcing caused by electrical faults is a common cause, so the weight of current fluctuation variance can be set relatively high. Subsequently, the platform device performs a dot product operation between the threshold exceedance values ​​for each dimension calculated in S208 and the weight coefficient vector. It is worth noting that this summation only applies to those dimensions judged to exceed the threshold in S207; dimensions that do not exceed the threshold have exceedance values ​​of 0 and are not included in the calculation.

[0094] In some embodiments, the calculation of the comprehensive fire risk score can be achieved in various ways. Optionally, a dynamic weight adjustment method based on the isolation of hotspots can be adopted. Specifically, firstly, before performing weighted summation, the platform device establishes a circular analysis area with a preset radius (e.g., 5 meters) centered on the active hotspot and acquires the temperature data of all pixels within this area. Secondly, the platform device calculates the temperature difference between the temperature of other pixels within this area, excluding the central active hotspot, and the temperature of the active hotspot, and counts the number of pixels with a temperature difference less than a certain threshold (e.g., 5°C, indicating similar temperatures). Then, it calculates the proportion of this number to the total number of pixels in the analysis area, i.e., the thermal anomaly pixel density. Next, the platform device determines whether the thermal anomaly pixel density is less than a preset density threshold (e.g., 1%). If so, it indicates that there are no other high-temperature points around the active hotspot, making it an isolated hotspot. This usually means that it may be caused by a concentrated point-like fault inside the equipment (such as diode breakdown), and its rapid temperature rise is a strong indicator. Therefore, the platform device temporarily increases the weight coefficient of the temperature rise rate dimension to a higher first preset weight value. If not, it indicates that a high-temperature area has formed around the active hotspot, possibly caused by external environmental factors (such as sunlight reflection) or a fire that has already begun to spread. In this case, the indicative significance of the temperature rise rate at a single point is relatively weakened. The platform device then adjusts the weight coefficient of the temperature rise rate dimension to a lower second preset weight value, which is less than the preset weight coefficient of the temperature rise rate dimension. Finally, the platform device uses this dynamically adjusted weight coefficient vector to perform a weighted summation of the threshold exceedance values ​​for each dimension to obtain a comprehensive fire risk score. Optionally, an adaptive weight allocation method based on equipment type can also be used. Specifically, before performing the weighted summation, the platform device queries a digital twin database based on the spatial coordinates of the active hotspot to determine its equipment type (such as photovoltaic modules, combiner boxes, inverters). Second, the platform device searches for a specific weight coefficient vector corresponding to the equipment type from a preset equipment type-weight strategy mapping table. For example, fires involving photovoltaic modules are typically accompanied by noticeable smoke (from burning encapsulation materials), so the weight of smoke detection confidence level may be set higher. Conversely, for combiner boxes or inverters, electrical faults are the primary risk source, so the weight of current fluctuation variance will be assigned the highest value. Finally, the platform equipment uses this set of more specialized weighting coefficients, tailored to specific equipment types, to calculate the comprehensive fire risk score.

[0095] S210. When the comprehensive fire risk score is greater than the preset risk score threshold, the central platform equipment determines the active hot spot as the actual fire point.

[0096] S211: The central platform equipment schedules the nearest available firefighting drone to perform firefighting tasks based on the three-dimensional spatial coordinates of the real fire point.

[0097] Step S211 is similar to step S107, and will not be described again here.

[0098] In the above embodiments, before calculating the comprehensive risk score, the platform device first performs thermal imaging analysis on the microenvironment surrounding the active hotspot. It determines whether the active hotspot is isolated or non-isolated by calculating the density of thermal anomalies. Isolated hotspots typically correspond to a breakdown or poor contact of a component inside the equipment; the fault points are highly concentrated, and a rapid temperature rise is a signal that a fire is about to break out. In this case, the weighting coefficient for the rate of temperature rise is set to a higher first preset weighting value, thereby improving the accuracy of identifying such potential risk points. Conversely, non-isolated hotspots mean that the thermal anomaly has already shown a patchy distribution, which may simply be due to large-area sunlight reflection, shading recovery, or a fire that has already begun to spread. In this case, the weighting coefficient for the rate of temperature rise is lowered to a second preset weighting value, allowing information from other dimensions such as smoke and current to play a more significant role in the decision-making process. Through this dynamic adjustment of weighting coefficients, the platform device can more accurately identify early fires of different causes and forms, thereby improving the accuracy of fire detection in photovoltaic power plants.

[0099] The above describes a multi-source fusion fire detection and emergency response method for a photovoltaic power station in an embodiment of this application. The following describes an exemplary middleware device 300 provided in an embodiment of this application.

[0100] Figure 3 This is a schematic diagram of an exemplary hardware structure of the middleware device 300 provided in an embodiment of this application. In some embodiments, the middleware device 300 is a computer device. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with other external terminals or servers via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements a multi-source fusion fire detection and emergency response method for a photovoltaic power plant according to an embodiment of this application.

[0101] Those skilled in the art will understand that Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0102] In some embodiments of this application, a computer-readable storage medium is also provided, including instructions that, when executed on the middleware device 300, cause the middleware device 300 to execute a multi-source fusion fire detection and emergency response method for a photovoltaic power station according to an embodiment of this application.

[0103] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0104] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0105] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0106] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A multi-source fusion fire detection and emergency response method for photovoltaic power plants, characterized in that, Applied to middleware devices, the method includes: The central platform device scans the target photovoltaic power station with thermal imaging equipment to obtain a thermal image data stream containing the temperature value and spatial coordinates of each pixel. The middle platform device calculates the temperature deviation and temperature rise rate of each pixel in the heat map data stream from the theoretical working temperature under the corresponding device type, current ambient temperature and light intensity conditions, based on a preset normal working temperature model of photovoltaic equipment. The middleware device marks pixels whose temperature deviation value is greater than a preset deviation threshold and whose temperature rise rate is greater than a first preset rate threshold as active hotspots. The central platform device acquires image data of the active hotspot through a visible light camera and calculates the smoke detection confidence level of the image data based on a smoke feature recognition algorithm; The middleware device acquires the current data of the circuit where the active hotspot is located through a current sensor, and calculates the current fluctuation variance of the current data. When at least two of the following conditions are met simultaneously: the temperature rise rate of the active hot spot is greater than the second preset rate threshold, the smoke detection confidence is greater than the preset confidence threshold, and the current fluctuation variance is greater than the preset variance threshold, the central platform device will confirm the active hot spot as a real fire point. The central platform device schedules the nearest available firefighting drone to perform firefighting tasks based on the three-dimensional spatial coordinates of the actual fire point.

2. The method according to claim 1, characterized in that, Before the step of the intermediate platform device performing hot spot scanning on the target photovoltaic power station using a thermal imaging device to obtain a thermal image data stream containing the temperature value and spatial coordinates of each pixel, the method further includes: The central platform equipment acquires historical operating data of different types of equipment within the target photovoltaic power station under different environmental conditions; The middleware platform uses principal component analysis to perform dimensionality reduction on the historical operating data, extracts the main environmental factors affecting the equipment temperature, and calculates the weight coefficients of each of the main environmental factors. The middle platform device constructs a mathematical relationship model between the device temperature and the main environmental factors based on a multivariate nonlinear regression algorithm and the weighting coefficients. The middleware device optimizes the parameters of the mathematical relationship model based on cross-validation. The middleware platform stores the optimized mathematical relationship model according to equipment type to obtain the preset normal operating temperature model of photovoltaic equipment.

3. The method according to claim 2, characterized in that, The middleware platform, based on a preset normal operating temperature model for photovoltaic equipment, calculates the temperature deviation and temperature rise rate between the temperature value of each pixel in the thermal image data stream and the theoretical operating temperature under the corresponding equipment type, current ambient temperature, and light intensity conditions. Specifically, this includes: The middleware device determines the device type corresponding to the pixel based on the spatial coordinates of the pixel and a preset device layout database; The middleware device calls the mathematical relationship model from the preset normal operating temperature model of photovoltaic equipment based on the equipment type; The middleware device substitutes the current environmental data into the mathematical relationship model to calculate the theoretical operating temperature of the pixel. The middleware uses the difference between the actual temperature value of the pixel and the theoretical working temperature as the temperature deviation value. The platform device uses a sliding time window to perform linear regression fitting on the historical temperature data of the pixels, and uses the slope of the fitted line as the rate of temperature rise.

4. The method according to claim 1, characterized in that, The steps of acquiring image data of the active hotspot through a visible light camera and calculating the smoke detection confidence of the image data based on a smoke feature recognition algorithm specifically include: The middleware device acquires image data of the active hotspot based on a visible light camera device; The middleware device identifies dynamic regions with smoke diffusion characteristics based on the time-varying characteristics of pixel grayscale values ​​in the image data; The platform device calculates the texture complexity and edge gradient change rate of pixels in the dynamic region; The middleware device marks the dynamic region whose texture complexity is less than a preset complexity threshold and whose edge gradient change rate conforms to the smoke diffusion pattern as a suspected smoke region; The middleware uses the similarity between the color distribution features and transparency change features of the suspected smoke area and the preset smoke feature template as the confidence level for smoke detection.

5. The method according to claim 1, characterized in that, The step of the middleware device confirming the active hot spot as a real fire point when at least two of the following conditions are met simultaneously: the temperature rise rate of the active hot spot is greater than a second preset rate threshold, the smoke detection confidence level is greater than a preset confidence threshold, and the current fluctuation variance is greater than a preset variance threshold: specifically includes: The middleware establishes a multi-dimensional feature vector for the active hotspot, and the multi-dimensional feature vector includes at least the values ​​of three dimensions: temperature rise rate, smoke detection confidence, and current fluctuation variance. The middleware device calculates the number of dimensions in the multidimensional feature vector that are greater than a preset dimension threshold; When the number of dimensions is greater than or equal to two, the middleware device calculates the threshold exceedance value of each threshold exceedance dimension. The threshold exceedance value is the value of the threshold exceedance dimension divided by the preset dimension threshold and then minus one. The central platform device calculates a weighted sum of the threshold exceedance values ​​of each dimension based on preset weighting coefficients to obtain a comprehensive fire risk score. When the overall fire risk score is greater than a preset risk score threshold, the central platform device determines the active hot spot as a real fire point.

6. The method according to claim 5, characterized in that, Before the step where the intermediate platform device weights and sums the threshold exceedance values ​​of each dimension based on preset weighting coefficients to obtain a comprehensive fire risk score, the method further includes: The central platform device establishes a circular analysis area with a preset radius centered on the active hot spot, and acquires the temperature data of all pixels within the analysis area; The platform device calculates the temperature difference between the active hot spot and other pixels within the circular analysis area, and counts the number of pixels whose temperature difference is greater than a preset difference threshold. The middleware uses the ratio of the number of pixels to the total number of pixels in the analysis area as the thermal anomaly pixel density. The middleware determines whether the density of the thermally abnormal pixels is less than a preset density threshold. If so, the middleware determines that the active hotspot is an isolated hotspot and sets the weight coefficient of the temperature rise rate dimension to a first preset weight value, which is greater than the preset weight coefficient of the temperature rise rate dimension. If not, the middleware determines that the active hotspot is a non-isolated hotspot and sets the weight coefficient of the temperature rise rate dimension to a second preset weight value, which is less than the preset weight coefficient of the temperature rise rate dimension.

7. The method according to claim 1, characterized in that, The steps of the central platform device scheduling the nearest available firefighting drone to perform firefighting tasks based on the three-dimensional spatial coordinates of the actual fire point specifically include: The central platform device determines the nearest available firefighting drone based on the three-dimensional spatial coordinates of the actual fire point; The central platform device calculates the optimal flight waypoint sequence based on the current location of the available firefighting drone and the three-dimensional spatial coordinates of the actual fire point, and the waypoint sequence avoids the preset no-fly zone; The central platform device determines the type and dosage of extinguishing agent based on the temperature value of the actual fire point and the equipment type, and generates fire extinguishing operation command parameters. The central platform device controls the available firefighting drones to perform firefighting tasks based on the optimal flight waypoint sequence and the firefighting operation command parameters.

8. A middleware platform, characterized in that, The middleware device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the middleware device to perform the method as described in any one of claims 1-7.

9. A computer program product containing instructions, characterized in that, When the computer program product is run on the middleware, the middleware performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the middleware device, the middleware device performs the method as described in any one of claims 1-7.

Images