Electrical equipment intelligent monitoring method and system based on intelligent sensor

Abstract

The invention relates to the field of electrical equipment safety monitoring, and discloses an electrical equipment intelligent monitoring method and system based on an intelligent sensor. The method comprises the following steps: acquiring temperature data of an equipment shell through a thermal image sensor, calculating a temperature gradient, and generating an initial thermal image containing coordinates of a temperature rising region; smoke sensor data is mapped to a corresponding pixel grid through image fusion, and an overlapping region boundary is determined; the gas sensor collects boundary harmful gas spectral line data and compares the data with a threshold value to generate a concentration index. And updating a fusion image layer based on the index, detecting temperature and smoke diffusion association frame by frame to obtain a diffusion path vector, extracting spatial-temporal characteristics, integrating boundaries, and calculating a fire alarm precursor probability value. If the probability exceeds a threshold value, polling data recollection is triggered, a temperature abnormal area is positioned, the resolution of the thermal image sensor is adjusted, and a monitoring circulation image is formed. According to the method, accurate identification and continuous monitoring of the fire premonition are realized, the monitoring reliability is improved, and support is provided for equipment operation and maintenance protection.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment safety monitoring technology, and in particular to an intelligent monitoring method and system for electrical equipment based on intelligent sensors. Background Technology

[0002] Electrical equipment safety monitoring is a core guarantee for industrial production and public facility operation and maintenance. The accuracy of monitoring directly affects equipment stability and personnel safety. Statistics show that from January to October 2023, electrical fires accounted for 29.1% of all fires in my country, ranking first among fire types, and more than 80% of these fires originated from hidden hazards such as aging equipment and poor connections. With the development of intelligent sensor technology, multi-source data fusion monitoring has become an important direction for preventing electrical fires. By integrating information from multiple sensors to capture equipment status, it provides support for preventative maintenance and is crucial for reducing fire risks. Developing efficient multi-sensor monitoring technology has become an urgent need in the safety field. Current monitoring methods have significant limitations. Traditional systems often rely on a single sensor, and temperature or smoke data alone cannot fully reflect fault characteristics—temperature anomalies may originate from multiple causes such as overload or short circuits, which cannot be distinguished by a single data point, leading to a high false alarm rate. In some scenarios, even adjusting to high-level values ​​increases the risk of missed alarms. Even when using multiple sensors, there is a lack of effective fusion mechanisms, failing to explore the correlation between temperature and smoke data, and relying mostly on static threshold judgments, making it difficult to adapt to dynamic operating environments. The shortcomings are even more pronounced in complex industrial scenarios. When equipment such as high-voltage switchgear operates under high load, heat accumulation and smoke diffusion exhibit complex spatiotemporal correlations, but traditional systems cannot collaboratively analyze multi-dimensional data. For example, when a large distribution cabinet experiences localized temperature rise accompanied by slight smoke, ventilation may prevent a single indicator from reaching a threshold, making the system unable to correlate and identify it, leading to the neglect of early fire warning signs. Furthermore, traditional systems have low intelligence, lack dynamic verification capabilities, and are prone to misinterpreting normal fluctuations or missing abnormal signals, limiting the value of data application. Therefore, more efficient multi-sensor fusion monitoring technology is urgently needed to solve these problems. Summary of the Invention

[0003] This invention provides a method and system for intelligent monitoring of electrical equipment based on intelligent sensors, so as to achieve accurate identification and continuous monitoring of fire alarm precursors, improve monitoring reliability, and provide support for equipment operation and maintenance protection.

[0004] In a first aspect, to address the aforementioned technical problems, the present invention provides an intelligent monitoring method for electrical equipment based on intelligent sensors, comprising: The device's outer casing surface temperature distribution data is collected by a thermal imaging sensor, and the temperature gradient value is calculated to generate an initial thermal image containing the coordinates of the temperature rise area. The particle concentration change sequence collected by the smoke sensor is obtained, and the particle concentration change sequence is mapped to the pixel grid corresponding to the coordinates of the temperature rise area using an image fusion algorithm to determine the boundary of the overlapping area. Harmful gas spectral data at the boundary of the overlapping region are collected by a gas sensor. Harmful gas concentration index is generated by comparing the matching degree between the intensity of the harmful gas spectral data and a preset spectral threshold. The fused image layer in the pixel grid is updated based on the harmful gas concentration index. The dynamic correlation between the temperature rise trend and the smoke diffusion path in the fused image layer is detected by frame-by-frame comparison to obtain the diffusion path vector. Spatiotemporal feature sequences are extracted from the diffusion path vector, and an image fusion algorithm is used to integrate the spatiotemporal feature sequences with the boundary of the overlapping area to calculate the probability value of fire alarm precursors. If the probability value of the fire alarm precursor is higher than the preset alarm threshold, a polling mechanism is triggered to re-collect data by controlling the thermal imaging sensor and the smoke sensor to generate a fusion dataset containing updated temperature gradient and particle concentration. Based on the updated fusion dataset, key areas of temperature anomalies are located and the resolution parameters of the thermal imaging sensor are adjusted to form a monitoring loop image.

[0005] In one optional implementation, the step of acquiring temperature distribution data of the device casing surface using a thermal imaging sensor and calculating temperature gradient values ​​to generate an initial thermal image containing coordinates of temperature rise areas includes: Temperature data is collected from the surface of the device casing using thermal imaging sensors, and the spatial location information of each collection point is recorded simultaneously to form raw temperature location correlation data containing temperature distribution data. The original temperature location correlation data is analyzed using a temperature gradient calculation method to determine the temperature change amplitude between adjacent acquisition points, thereby obtaining the temperature gradient distribution results on the surface of the equipment casing. Based on a preset temperature gradient threshold, regions where the temperature gradient exceeds the temperature threshold are selected from the temperature gradient distribution results, and temperature rise regions are marked by combining the recorded spatial location information. The spatial location information of the temperature rise area is extracted and converted into corresponding coordinates, and then integrated with the temperature distribution data in the original temperature location association data to generate an initial thermal image.

[0006] In one optional implementation, the step of acquiring the particle concentration change sequence collected by the smoke sensor and using an image fusion algorithm to map the particle concentration change sequence to the pixel grid corresponding to the coordinates of the temperature rise area, and determining the boundary of the overlapping region, includes: The sequence of particle concentration changes collected by the smoke sensor is obtained, and the collection time and spatial location information corresponding to each concentration data are recorded simultaneously to form spatiotemporal correlation data of particle concentration. An image fusion algorithm is invoked, using the pixel grid corresponding to the coordinates of the temperature rise area as a reference frame, to map the concentration change information in the spatiotemporal correlation data of particle concentration point by point to the corresponding position of the pixel grid; By overlaying and analyzing the temperature rise zone data and particle concentration distribution data within the mapped pixel grid, the overlapping areas simultaneously covered by the two types of data in the pixel grid are identified. Based on the overlay analysis results, the edge contours of the overlapping regions are extracted and determined to obtain the boundaries of the overlapping regions.

[0007] In one optional implementation, the step of collecting harmful gas spectral line data at the boundary of the overlapping region using a gas sensor, and generating a harmful gas concentration index by comparing the intensity of the harmful gas spectral line data with a preset spectral threshold, includes: Based on the spatial range of the boundary of the overlapping region, the gas sensor is controlled to detect harmful gases and collect the harmful gas spectral data. The collected spectral data of harmful gases are preprocessed to remove spectral clutter caused by environmental interference and retain effective spectral data that reflect the true characteristics of harmful gases. The intensity of the effective spectral line data is compared with a preset spectral line threshold, and the degree of deviation of the effective spectral line intensity relative to the preset spectral line threshold is calculated to form a matching degree analysis result between the two. Based on the matching degree analysis results, and combined with the correspondence between the effective spectral line intensity and the concentration of harmful gases, the degree of deviation is converted into a quantified indicator of the concentration of harmful gases.

[0008] In one optional implementation, updating the fused image layer in the pixel grid based on the harmful gas concentration index, and obtaining the diffusion path vector by frame-by-frame comparison and detection of the dynamic correlation between the temperature increase trend and the smoke diffusion path in the fused image layer, includes: Based on the harmful gas concentration index, differentiated feature identifiers are set according to concentration levels and embedded in the pixel grid fusion image layer to replace pixel data without concentration information, so that the fusion image layer synchronously contains temperature, smoke and harmful gas concentration information to complete the update. The updated fused image layer generates a continuous frame sequence at preset time intervals, and extracts the location coordinates and direction of change of the temperature rise area, as well as the coverage range and expansion direction of the smoke diffusion from each frame. By employing a frame-by-frame comparison method, and through spatial overlap calculation and rate of change correlation analysis, the dynamic correlation between temperature rise trend and smoke diffusion change in adjacent frames is detected. Based on the dynamic correlation, smoke diffusion trajectories that are highly correlated with the temperature rise trend are selected, and the direction and trajectory characteristics of the smoke diffusion trajectories are sorted out and transformed into diffusion path vectors containing directional attributes and path nodes.

[0009] In one optional implementation, the step of extracting the spatiotemporal feature sequence from the diffusion path vector, integrating the spatiotemporal feature sequence with the boundary of the overlapping region using an image fusion algorithm, and calculating the probability value of the fire precursor includes: From the diffusion path vector, the temporal variation pattern and spatial extension characteristics of the smoke diffusion process are extracted and organized into a spatiotemporal feature sequence reflecting the dynamic process of diffusion according to the monitoring time sequence; An image fusion algorithm is invoked, using the spatial range of the overlapping region boundary as a reference framework, to embed the dynamic feature information in the spatiotemporal feature sequence into the pixel coordinates of the overlapping region boundary, thereby completing data integration. The integrated data is subjected to feature filtering to extract key features related to fire alarm precursors; among which, the key features include at least changes in diffusion rate and increase in regional coverage rate. Based on the preset fire alarm precursor determination model, the extracted key features are substituted into the fire alarm precursor determination model for calculation to obtain a quantified fire alarm precursor probability value.

[0010] In one optional implementation, the step of re-acquiring data by controlling the thermal imaging sensor and the smoke sensor to generate a fused dataset containing updated temperature gradients and particle concentrations includes: The thermal imaging sensor is controlled to re-acquire temperature data of the device casing surface in the original monitoring area, and the spatial location information of the acquisition point is recorded synchronously. The updated temperature gradient is calculated based on the temperature data; the temperature gradient includes the spatial location identifier of the corresponding acquisition point. Using the thermal imaging sensor's acquisition area as a reference, the smoke sensor is controlled to re-acquire particle concentration data in the same area, synchronously recording the acquisition time and spatial location information. This spatial location information is consistent with the spatial location identifier of the temperature gradient, forming an updated particle concentration sequence. The updated temperature gradient data and particle concentration sequence were calibrated in terms of spatial location coordinates and acquisition time dimension to ensure a perfect match between the two and eliminate data bias. The calibrated and matched temperature gradient data is associated and integrated with the particle concentration sequence, and bound according to the preset format to form a fusion dataset containing updated temperature gradients and particle concentrations.

[0011] In one optional implementation, the step of locating key areas of temperature anomalies and adjusting the resolution parameters of the thermal imaging sensor based on the updated fused dataset to form a monitoring loop image includes: The updated fused dataset is retrieved, and the updated temperature gradient data is extracted. By analyzing the abnormal fluctuation characteristics of the temperature gradient, key areas of temperature anomalies where the temperature changes exceed the normal range are located. Based on the spatial range and monitoring requirements of the key temperature anomaly area, the resolution parameters of the thermal imaging sensor are adjusted so that the temperature acquisition accuracy of the sensor in that area matches the monitoring requirements. After adjusting the parameters, the thermal imaging sensor continuously collects temperature data on the critical areas of temperature anomalies and their surroundings, and synchronously generates continuous temperature monitoring image frames. The continuous temperature monitoring image frames are organized according to the acquisition time sequence, and supplemented with particle concentration information from the fusion dataset to form a monitoring cycle image reflecting the state of abnormal areas.

[0012] The present invention also provides an intelligent monitoring system for electrical equipment based on intelligent sensors, comprising: Thermal image generation module: Collects temperature distribution data of the device casing surface through thermal imaging sensors and calculates temperature gradient values ​​to generate an initial thermal image containing the coordinates of the temperature rise area; Overlapping boundary determination module: acquires the particle concentration change sequence collected by the smoke sensor, and uses an image fusion algorithm to map the particle concentration change sequence to the pixel grid corresponding to the coordinates of the temperature rise area to determine the boundary of the overlapping area; Gas concentration generation module: Collects harmful gas spectral line data at the boundary of the overlapping region through a gas sensor, and generates a harmful gas concentration index by comparing the matching degree of the intensity of the harmful gas spectral line data with a preset spectral line threshold. Diffusion vector acquisition module: Based on the harmful gas concentration index, update the fused image layer in the pixel grid, and use frame-by-frame comparison to detect the dynamic correlation between the temperature rise trend and the smoke diffusion path in the fused image layer to obtain the diffusion path vector; Fire alarm probability calculation module: Extracts spatiotemporal feature sequence from the diffusion path vector, integrates the spatiotemporal feature sequence with the boundary of the overlapping area using an image fusion algorithm, and calculates the probability value of fire alarm precursor; Fusion Dataset Update Module: If the probability value of the fire alarm precursor is higher than the preset alarm threshold, a polling mechanism is triggered to re-collect data by controlling the thermal imaging sensor and the smoke sensor to generate a fusion dataset containing updated temperature gradient and particle concentration. Monitoring cycle image generation module: Based on the updated fusion dataset, it locates key areas of temperature anomalies and adjusts the resolution parameters of the thermal imaging sensor to generate monitoring cycle images.

[0013] Compared with the prior art, the present invention has the following beneficial effects: (1) By collecting temperature data and calculating temperature gradients through thermal imaging sensors, an initial thermal image containing the coordinates of the heating zone is generated, which can accurately locate the abnormal temperature area on the equipment shell and fully capture the spatial characteristics of temperature distribution. This temperature gradient analysis avoids the limitations of single temperature data, lays the foundation for the correlation of multi-sensor data, improves the early detection capability of thermal anomalies in equipment, and solves the problem that traditional single sensors cannot accurately locate anomalies. (2) The particle concentration change sequence of the smoke sensor is mapped to the corresponding pixel grid of the temperature rise area through image fusion to determine the boundary of the overlapping area and realize the spatial correlation of temperature and smoke data. This fusion method explores the spatial correlation between the two types of data, accurately locates the high-risk area of ​​"temperature anomaly + smoke", avoids the risk misjudgment caused by the isolation of traditional monitoring data, and provides a clear range for subsequent targeted monitoring. (3) Harmful gas spectral data in overlapping areas are collected by gas sensors and compared with preset thresholds to generate concentration indicators, supplementing the chemical characteristic dimension of equipment faults. This multi-dimensional acquisition breaks through the limitations of traditional methods that rely solely on physical parameters. It can use the concentration of harmful gases to assist in judging the fault type, reduce the false alarm rate of single-parameter monitoring, and improve the accuracy of fault judgment. (4) The fused image layer is updated based on the concentration index of harmful gases, and the dynamic correlation between temperature rise and smoke diffusion is detected frame by frame to obtain the diffusion path vector, so as to realize multi-parameter spatiotemporal collaborative analysis. This method captures the temporal coupling relationship between temperature and smoke changes, and can identify the hidden fire alarm precursor of "weak temperature rise + slow smoke", which solves the defect of traditional static threshold monitoring that cannot identify weak correlation signals and improves the sensitivity of early warning. (5) Extract the spatiotemporal feature sequence of the diffusion path vector, integrate it with the boundary of the overlapping area to calculate the probability value of the fire alarm precursor, and transform the multi-dimensional data into a quantitative risk indicator. This quantitative assessment avoids the subjectivity of traditional qualitative judgment, makes the risk assessment more objective and operable, provides a precise basis for triggering the polling mechanism, and reduces the underreporting or over-alarming caused by experience judgment. (6) When the probability of a fire alarm precursor exceeds the threshold, a polling mechanism is triggered to recollect data, locate the temperature anomaly area, and adjust the resolution of the thermal imaging sensor to form a monitoring loop image, thereby achieving dynamic monitoring adaptation. This dynamic adjustment improves the monitoring accuracy for high-risk areas. By tracking risk changes through loop monitoring, it solves the problem of insufficient accuracy of traditional fixed parameter monitoring in risk escalation scenarios, ensuring accurate and continuous control of abnormal areas and providing a guarantee for the safe operation and maintenance of equipment. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating an intelligent monitoring method for electrical equipment based on intelligent sensors, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an intelligent monitoring system for electrical equipment based on intelligent sensors, provided in an embodiment of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Reference Figure 1 This invention provides a method for intelligent monitoring of electrical equipment based on smart sensors, comprising the following steps: S11, Collect temperature distribution data on the surface of the device casing using a thermal imaging sensor and calculate the temperature gradient value to generate an initial thermal image containing the coordinates of the temperature rise area; S12, acquire the particle concentration change sequence collected by the smoke sensor, and use an image fusion algorithm to map the particle concentration change sequence to the pixel grid corresponding to the temperature rise area coordinates to determine the boundary of the overlapping area. S13, collect harmful gas spectral line data of the boundary of the overlapping region through a gas sensor, and generate a harmful gas concentration index by comparing the matching degree of the intensity of the harmful gas spectral line data with a preset spectral line threshold. S14, update the fused image layer in the pixel grid based on the harmful gas concentration index, and use frame-by-frame comparison to detect the dynamic correlation between the temperature rise trend and the smoke diffusion path in the fused image layer to obtain the diffusion path vector; S15, extract the spatiotemporal feature sequence from the diffusion path vector, and use an image fusion algorithm to integrate the spatiotemporal feature sequence with the boundary of the overlapping area to calculate the probability value of the fire alarm precursor. S16, if the probability value of the fire alarm precursor is higher than the preset alarm threshold, the polling mechanism is triggered, and the thermal imaging sensor and smoke sensor are controlled to re-collect data to generate a fusion dataset containing updated temperature gradient and particle concentration. S17, based on the updated fusion dataset, locates key areas of temperature anomalies and adjusts the resolution parameters of the thermal imaging sensor to form a monitoring loop image.

[0017] In step S11, the temperature distribution data of the device housing surface is acquired by a thermal imaging sensor and the temperature gradient value is calculated to generate an initial thermal image containing the coordinates of the temperature rise area.

[0018] In one embodiment, this example uses a 10kV high-voltage distribution cabinet as the monitoring object. An FLIRA655sc infrared thermal imaging sensor (resolution 1024×768 pixels, temperature range -20℃ to 150℃, accuracy ±2%, concentration detection range 0 to 1000mg / m³) is selected and fixed 1.5m directly in front of the distribution cabinet. Temperature data of the cabinet's outer surface is collected at a sampling frequency of 30Hz. Simultaneously, a two-dimensional plane coordinate system is established with the lower left corner of the distribution cabinet as the origin, recording the spatial coordinates (x, y) of each collection point, forming raw data with a one-to-one correspondence between "temperature value and spatial coordinates." For example, coordinates (320, 480) correspond to a temperature of 42.3℃, and coordinates (640, 320) correspond to a temperature of 31.5℃. Before data collection, calibration is performed using a blackbody radiation source at three standard temperature points: 30℃, 50℃, and 80℃. Deviations are adjusted according to JJF1371-2012 "Infrared Thermometer Calibration Specification" to ensure that the temperature measurement error is ≤ ±0.5℃.

[0019] The original data is arranged into a 1024×768 two-dimensional temperature matrix T(x,y) (T(x,y) represents the temperature value at coordinate (x,y)). The Sobel operator algorithm is used to calculate the temperature gradient: horizontal gradient formula. This reflects the temperature difference between the sampling point (x, y) and its left and right adjacent points, representing the horizontal temperature variation; the vertical gradient formula... This reflects the temperature difference between the sampling point (x, y) and its adjacent points, representing the vertical temperature variation. For points on the matrix edges (e.g., x=0 or x=1023, y=0 or y=767), a one-sided difference method is used for correction. For example, the horizontal gradient formula for the edge point (0, y) is adjusted as follows: The formula for the vertical gradient of the edge point (x,0) is adjusted to: This ensures accurate gradient calculation in edge regions.

[0020] Then, using the formula for the total temperature gradient magnitude (where the radical sign represents the square root operation, G) x (x,y) 2 G represents the square of the horizontal gradient. y (x,y) 2The total temperature gradient value at each sampling point is calculated by representing the square of the vertical gradient, and the results are arranged according to spatial coordinates to form the temperature gradient distribution on the surface of the equipment casing. Referring to the safety requirements for temperature gradients of electrical equipment casings in GB7251.1-2022 "Low-voltage switchgear and controlgear assemblies - Part 1: General", the preset temperature gradient threshold is 5℃ / cm. In this embodiment, the temperature gradient values ​​in the lower left corner of the cabinet are mostly 1.2℃ / cm to 2.5℃ / cm, while the temperature gradient values ​​in the upper middle area of ​​the cabinet (coordinates x=512~600, y=300~380) reach 8.3℃ / cm to 12.1℃ / cm, significantly higher than the surrounding areas.

[0021] From the temperature gradient distribution results, regions with a total temperature gradient value |G(x,y)| exceeding 5℃ / cm were selected. Combining the spatial location information of the previously recorded collection points, an eight-neighbor connected region labeling algorithm was used to group adjacent gradient exceeding the threshold points into the same region, eliminating single isolated exceeding the threshold points (judged as noise interference). In addition, based on the physical structure of the cabinet (such as cabinet doors, wiring terminals, and the location of heat dissipation holes), false gradient increase regions caused by the airflow of heat dissipation holes were excluded. Finally, three temperature increase regions were marked: region A (coordinate range x=512~600, y=300~380), region B (coordinate range x=720~780, y=420~460), and region C (coordinate range x=280~320, y=240~280).

[0022] Extract the spatial location information of the three temperature rise zones, and map them using the pixel coordinate mapping formula (pixel coordinates). Pixel coordinates The spatial coordinates (x, y) are converted into pixel coordinates (u, v) in the thermal image sensor image, where 1024 and 768 are the horizontal and vertical resolutions of the thermal image sensor, respectively. The converted pixel coordinates are then integrated with the temperature distribution data (i.e., the two-dimensional temperature matrix T(x, y)) in the original temperature location association data. A pseudo-color mapping algorithm is used to visualize the temperature values ​​(areas with temperatures below 30℃ are displayed in blue, areas between 30℃ and 45℃ are displayed in green, areas between 45℃ and 60℃ are displayed in yellow, and areas above 60℃ are displayed in red). The pixel coordinate ranges of the three temperature rise zones are marked with red dashed boxes in the visualized image, and the average temperature value of each temperature rise zone is also marked (e.g., the average temperature of area A is 52.7℃). Finally, an initial thermal image containing the location coordinates of the temperature rise zones is generated.

[0023] In step S12, the particle concentration change sequence collected by the smoke sensor is obtained, and the particle concentration change sequence is mapped to the pixel grid corresponding to the temperature rise area coordinates using an image fusion algorithm to determine the boundary of the overlapping area.

[0024] In one embodiment, this example continues the monitoring scenario of a 10kV high-voltage distribution cabinet. Based on the initial thermal image and the determination of temperature rise zones (Area A: x=512~600, y=300~380; Area B: x=720~780, y=420~460; Area C: x=280~320, y=240~280), further operations are carried out on smoke sensor data acquisition and overlapping area boundary determination. An MQ-135 smoke sensor is selected, with a concentration detection range of 0~1000mg / m³ and an accuracy of ±5%. It is deployed in an array around the distribution cabinet—one sensor is placed at the center of the top of the cabinet and one at the midpoint of each of the four sides. The sensor sampling frequency is set to 5Hz. Before collection, the sample is calibrated using a smoke calibration source with standard concentrations (50 mg / m³, 200 mg / m³, 500 mg / m³). The deviation is adjusted according to JJG653-2019 "Verification Procedure for Light Scattering Smoke Meters" to ensure that the concentration detection error is ≤ ±3 mg / m³.

[0025] During formal data collection, the sensor captures particle concentration data around the power distribution cabinet in real time, and synchronously records the collection time (accurate to the second) and spatial location (two-dimensional coordinates with the lower left corner of the power distribution cabinet as the origin, such as the coordinates (600, 1000) corresponding to the top center sensor) for each set of concentration data, forming spatiotemporal correlation data of particle concentration. For example, the particle concentration corresponding to the collection time 10:05:01 and coordinates (600, 1000) is 18 mg / m³, and the particle concentration corresponding to the collection time 10:05:02 and coordinates (300, 600) (midpoint of the side) is 15 mg / m³.

[0026] Subsequently, an image fusion algorithm based on wavelet transform was invoked (the algorithm type was specified, and the algorithm parameters were supplemented: the wavelet basis function was db4, and the decomposition level was 3). Using the pixel grid corresponding to the temperature rise zone coordinates (consistent with the thermal imaging sensor's resolution of 1024×768 pixels, with each pixel corresponding to a 0.5mm×0.5mm physical area on the cabinet surface) as the baseline framework, the spatiotemporal correlation data of particle concentration was first preprocessed—concentration data between adjacent collection points was supplemented using a linear interpolation algorithm to match the spatial distribution density of the concentration data with the pixel grid; then, based on the spatial coordinates corresponding to the concentration data, the coordinate mapping formula (…) was used… , Each concentration value is mapped point by point to the corresponding position of the pixel grid, forming a particle concentration pixel distribution matrix C(u,v) (where C(u,v) represents the particle concentration value at pixel (u,v)).

[0027] Referring to the background concentration test data during fault-free operation of the distribution cabinet (72 hours of continuous monitoring, with a mean background concentration of 8 mg / m³ and a standard deviation of 2 mg / m³), a particle concentration threshold of 12 mg / m³ was set (mean + 2 times the standard deviation, ensuring a scientifically sound threshold). Overlay analysis was performed on the mapped pixel grid data: the ranges of the three previously determined temperature rise zones within the pixel grid were extracted (e.g., pixel coordinates u=512~600, v=300~380 for zone A). This range was spatially overlaid with the particle concentration pixel distribution matrix C(u,v), and pixels with particle concentrations ≥12 mg / m³ within the temperature rise zones were selected. The areas where these pixels are located represent the overlapping regions simultaneously covered by the temperature rise zones and the particle concentration exceeding the standard zones. Through overlay analysis, in region A, the particle concentration in the pixel coordinate range of u=520~590 and v=310~370 is ≥12mg / m³ (up to 25mg / m³), forming a significant overlapping area; in region B, only a small number of pixels at the edge (u=770~780, v=430~440) meet the concentration standard, and the overlapping range is extremely small; in region C, there are no pixels with the concentration standard, and there is no overlapping area.

[0028] The overlapping region boundary is determined based on the overlay analysis results: For the pixel set of the overlapping region in area A, the Canny edge detection algorithm is used to extract the edge contour (supplementary algorithm parameters: Gaussian filter kernel size 3×3, standard deviation σ=1.2, high threshold 70, low threshold 30). First, the particle concentration distribution image is smoothed by Gaussian filtering, and then the image gradient is calculated (using the Sobel operator) to determine the edge intensity. Edges are connected by high and low thresholds to finally obtain the pixel coordinate sequence of the edge contour of the overlapping region in area A, such as (u=520, v=310), (u=590, v=310), (u=590, v=370), (u=520, v=370). These coordinates are connected in sequence to form a closed boundary, which is the final determined overlapping region boundary; Due to the extremely small overlap range in area B and the lack of overlap in area C, no effective overlapping region boundary is generated for the time being.

[0029] In step S13, harmful gas spectral line data of the boundary of the overlapping region are collected by a gas sensor. By comparing the matching degree of the intensity of the harmful gas spectral line data with the preset spectral line threshold, a harmful gas concentration index is generated.

[0030] In one embodiment, this embodiment continues the monitoring scenario of a 10kV high-voltage distribution cabinet. After determining the boundary of the overlapping area of ​​area A (pixel coordinates u=520~590, v=310~370, corresponding cabinet physical coordinates x=520×(1.2m / 1024)~590×(1.2m / 1024), y=310×(2.0m / 768)~370×(2.0m / 768)), the operation of collecting harmful gas spectral data and generating concentration indicators is carried out. Considering the limited space around the 10kV high-voltage distribution cabinet, a miniaturized Fourier transform infrared (FTIR) gas sensor (model: Micro-FTIR-100, size 150mm×100mm×80mm, suitable for small spaces around the equipment) was selected. This sensor can accurately detect common harmful gases in electrical equipment faults such as carbon monoxide (CO, characteristic spectral wavelength 4.6μm) and sulfur dioxide (SO2, characteristic spectral wavelength 7.3μm), with a spectral resolution of 1cm⁻¹, a detection range of 0~500ppm, and an accuracy of ±2ppm.

[0031] The sensor probe was fixed 0.3m directly in front of the boundary of the overlapping area in Zone A using an adjustable bracket, ensuring that the detection range completely covered the cabinet surface within this boundary. Simultaneously, environmental conditions were recorded, including ambient temperature (25.3℃), humidity (45%RH), and atmospheric pressure (101.2kPa), to provide a basis for subsequent data preprocessing to eliminate environmental interference. During the actual data acquisition, the sensor scanned the gas composition within the boundary of the overlapping area in Zone A at a frequency of 10 times per second to obtain raw spectral data of harmful gases (including mixed spectral lines of CO, SO2, and ambient background gases).

[0032] The raw spectral data were preprocessed as follows: First, a baseline correction algorithm was used, taking the spectral lines of the ambient background gas (acquired before the power distribution cabinet was in operation, and the average value was taken after 10 minutes of continuous collection) as the benchmark, and subtracting the background baseline from the raw spectral lines to eliminate the interference of atmospheric water vapor, carbon dioxide, etc. on the spectral lines. Then, the corrected spectral lines were processed by Savitzky-Golay smoothing filter (supplementary algorithm parameters: window size of 5 points, polynomial order of 2) to remove spectral clutter caused by high-frequency noise, and finally obtained effective spectral line data that only reflects the characteristics of CO and SO2. The peak intensity of the effective CO spectral line at a wavelength of 4.6 μm is 0.92 μW, and the peak intensity of the effective SO2 spectral line at a wavelength of 7.3 μm is 0.35 μW.

[0033] Referring to the hazardous gas limit requirements for electrical equipment faults in GB / T50493-2019 "Design Standard for Detection and Alarm of Combustible and Toxic Gases in Petrochemical Industry", and combined with the background concentration test data when there is no fault (average background concentration of CO is 0.17ppm, and average background concentration of SO2 is 0.08ppm), the preset hazardous gas spectral thresholds are: CO characteristic spectral threshold is set to 0.5μW (corresponding to 3 times the background concentration), and SO2 characteristic spectral threshold is set to 0.2μW (corresponding to 2.5 times the background concentration). The effective spectral line intensity is compared with a preset threshold to calculate the matching degree. For CO, the deviation between the effective spectral line intensity (0.92 μW) and the threshold (0.5 μW) is (0.92-0.5) / 0.5×100%=84%, and the matching degree is judged as "high matching". For SO2, the deviation between the effective spectral line intensity (0.35 μW) and the threshold (0.2 μW) is (0.35-0.2) / 0.2×100%=75%, and the matching degree is judged as "medium matching".

[0034] Based on the correlation between effective spectral line intensity and harmful gas concentration (established through sensor calibration experiments, which followed JJG695-2019 "Verification Procedure for Infrared Gas Analyzers of Carbon Monoxide and Carbon Dioxide", using three standard concentration points (CO: 0.2ppm, 0.5ppm, 1.0ppm; SO2: 0.1ppm, 0.3ppm, 0.6ppm) for linear regression to obtain the calibration formula): The calibration formula for CO is as follows. , where c CO CO concentration (ppm), I CO The effective spectral intensity of CO is given in μW; the calibration formula for SO2 is... , where c SO2 SO2 concentration (ppm), I SO2 Let be the effective spectral line intensity of SO2 (μW). Substituting the effective spectral line intensity into the formula, we obtain the CO concentration c. CO =0.8×0.92-0.1=0.636ppm, SO2 concentration c SO2 =1.2×0.35-0.05=0.37ppm, and the final generated harmful gas concentration index is "CO: 0.636ppm, SO2: 0.37ppm", which provides a basis for gas parameters for subsequent updates and fusion of image layers.

[0035] In step S14, the fused image layer in the pixel grid is updated based on the harmful gas concentration index, and the dynamic correlation between the temperature rise trend and the smoke diffusion path in the fused image layer is detected by frame-by-frame comparison to obtain the diffusion path vector.

[0036] In one embodiment, this example continues the monitoring scenario of a 10kV high-voltage distribution cabinet. After obtaining the concentration indicators of harmful gases (CO: 0.636ppm, SO2: 0.37ppm), the image layer update and diffusion path vector acquisition operations are performed. Referring to the classification standard of harmful gas concentration in electrical equipment in GB50169-2016 "Code for Construction and Acceptance of Grounding Devices", and combined with the monitoring requirements of this invention, the harmful gas concentration levels and corresponding feature identifiers are set as follows: CO concentration 0~0.3ppm and SO2 concentration 0~0.2ppm are set as "low concentration", and the corresponding feature identifier is a blue semi-transparent layer (80% transparency); CO concentration 0.3~0.8ppm and SO2 concentration 0.2~0.5ppm are set as "medium concentration", and the corresponding feature identifier is a yellow semi-transparent layer (60% transparency); CO concentration >0.8ppm and SO2 concentration >0.5ppm are set as "high concentration", and the corresponding feature identifier is a red semi-transparent layer (40% transparency).

[0037] Since both CO and SO2 concentrations are currently in the "medium concentration" range, a yellow semi-transparent layer is embedded into the pixel grid (1024×768 pixels) mapped from the previous temperature rise area and smoke concentration in the fused image layer. This replaces the original pixel data that lacks concentration information. Specifically, this layer is overlaid in the overlapping area of ​​region A (pixel coordinates u=520~590, v=310~370) and within a 50-pixel radius around it. This allows the fused image layer to simultaneously contain three types of information: temperature distribution (pseudo-color mapping), smoke concentration (grayscale mapping: the higher the concentration, the larger the grayscale value), and harmful gas concentration (semi-transparent color mark). This completes the update of the fused image layer.

[0038] The updated fused image layer was collected at a preset time interval of 0.5 seconds / frame (determined based on sensor sampling frequency and data processing efficiency to ensure complete and non-redundant data in each frame), generating a continuous frame sequence, for a total of 60 frames (covering a 30-second monitoring period). Key information was extracted from each frame: information on the temperature rise zone was obtained by identifying the yellow and red areas in the pseudo-color mapping, and the center coordinates of the temperature rise zone were calculated using the centroid method (e.g., the center coordinates of area A in frame 1 are (u=555, v=340)). The direction of change was determined by the offset of the center coordinates of adjacent frames (the center coordinates from frame 1 to frame 30 change from (u=555, v=340) to (u=562, v=345), with the direction of change biased towards the upper right side of the cabinet); smoke diffusion information was extracted using the particle concentration pixel distribution matrix C(u,v), and a concentration ≥15mg / m³ was set as the effective diffusion range (above the threshold of 12mg / m³ to reduce noise interference). An edge detection algorithm was used to determine the boundary of the diffusion range, and the expansion direction was determined by the boundary offset (consistent with the direction of change of the temperature rise zone).

[0039] Dynamic correlation is detected by frame-by-frame comparison: Spatial overlap is calculated by determining the proportion of pixels intersecting the temperature rise area and the smoke diffusion range in each frame to the total number of pixels in the temperature rise area. For example, the overlap is 82% in frame 1, 85% in frame 10, and 88% in frame 30, all maintaining a high overlap (overlap ≥ 80% is considered "strong spatial correlation"). The rate of change correlation analysis is performed by calculating the displacement rate of the center coordinates of the temperature rise area within 10 adjacent frames (average rate of 0.14 pixels / frame from frame 1 to 10, corresponding to a physical rate of 0.14 × (1.2m / 1024) / 0.5s ≈ 3.28 × 10⁻⁶). -4 The displacement rate of the smoke diffusion range boundary (m / s) and the average rate of displacement of the smoke diffusion range boundary (simultaneous average rate 0.15 pixels / frame, corresponding to a physical rate of 0.15×(1.2m / 1024) / 0.5s≈3.52×10) is approximately 3.52×10⁻⁶ m / s. -4 (m / s), the speed difference between the two is ≤5×10 -5 m / s, indicating a high correlation in the rate of change. Combining the results of the two types of analysis, the dynamic correlation between the temperature increase trend and the smoke diffusion change in adjacent frames is "strong correlation".

[0040] Based on the "strong correlation" relationship, smoke diffusion trajectories highly correlated with the temperature rise trend within the overlapping area of ​​region A were selected. Using 10-second intervals as time nodes, the center coordinates of the smoke diffusion range (frame 1 (u=555, v=340), frame 10 (u=558, v=342), frame 20 (u=560, v=343), frame 30 (u=562, v=345)) were extracted as path nodes. The trajectory direction (5° angle with the cabinet surface, biased towards the vertical) was determined through vector calculation, transforming it into a diffusion path vector containing direction attributes and path nodes: the direction attribute is "5° angle with the cabinet surface (biased towards the vertical)," the path node sequence is [(555, 340), (558, 342), (560, 343), (562, 345)], and the vector magnitude (total displacement from frame 1 to frame 30) is... Each pixel corresponds to a physical displacement of approximately 8.6 × (1.2 m / 1024) ≈ 0.010 m in the cabinet, ultimately forming a complete diffusion path vector.

[0041] In step S15, a spatiotemporal feature sequence is extracted from the diffusion path vector, and an image fusion algorithm is used to integrate the spatiotemporal feature sequence with the boundary of the overlapping area to calculate the probability value of the fire alarm precursor.

[0042] In one embodiment, this example continues the monitoring scenario of a 10kV high-voltage distribution cabinet. After obtaining the diffusion path vector (direction attribute is "5° angle with the horizontal of the cabinet surface (biased to the vertical upward)", path node sequence [(555,340),(558,342),(560,343),(562,345)], physical displacement within 30 seconds is about 0.010m), spatiotemporal feature sequence extraction, data integration and fire alarm precursor probability value calculation are carried out.

[0043] Spatiotemporal features were extracted from the diffusion path vector: the temporal variation pattern was obtained by calculating the time interval between adjacent path nodes (one node every 10 seconds, time interval Δt = 10s) and the corresponding displacement (0.003m for the first 10 seconds, 0.002m for the first 10 seconds, and 0.005m for the first 20 seconds), thus obtaining the diffusion rate for each time period (rate for the first 10 seconds: 0.003m / 10s = 3 × 10). -4 m / s, the velocity from the 10th to the 20th second is 0.002m / 10s = 2 × 10 -4 m / s, the speed in the 20th-30th second is 0.005m / 10s = 5×10 -4 The temporal variation pattern is "slow at first, then fast"; the spatial extension feature is determined by analyzing the coordinate offset of the path nodes, which shows that the spatial extension direction is stably pointing to the upper right side of the cabinet (positive u-axis direction, positive v-axis direction), and the extension range expands from the initial overlapping area of ​​area A (u=520~590, v=310~370) to the surrounding area. The expansion area is about 200 pixels within 30 seconds. According to the correspondence between pixels and physical area (1 pixel corresponds to 0.5mm×0.5mm=0.25mm²), the actual expanded physical area is about 50mm². The temporal variation pattern and spatial extension feature are organized into a spatiotemporal feature sequence according to the monitoring time sequence (0s, 10s, 20s, 30s): [(t=0s, v=0m / s, S=0mm²), (t=10s, v=3×10 -4 m / s, S=15mm²), (t=20s, v=2×10 -4 m / s, S=30mm²), (t=30s, v=5×10 -4 m / s, S=50mm²).

[0044] A weighted average-based image fusion algorithm is invoked (the algorithm's weight setting logic is clearly defined: temperature data weight 0.4, smoke data weight 0.3, and spatiotemporal feature data weight 0.3, determined based on the degree of influence of the three types of data on fire alarm judgment). Using the spatial range of the overlapping area boundary in region A (pixel coordinates u=520~590, v=310~370) as the baseline framework, the dynamic feature information in the spatiotemporal feature sequence is embedded into the corresponding pixel coordinates within the boundary. For the diffusion rate v at each time node, the rule is "the higher the rate, the higher the corresponding pixel grayscale value" (v=5×10). -4 m / s corresponds to a grayscale value of 200, v = 2 × 10 -4 m / s corresponds to a grayscale value of 100), and the rate information is mapped to the pixel area within the boundary corresponding to the time period; for the extended area S, according to the rule that "the larger the area, the lower the transparency of the corresponding pixel" (S=50mm² corresponds to 30% transparency, S=15mm² corresponds to 70% transparency), the area information is superimposed on the pixel area in the form of a semi-transparent layer, so that the pixel coordinates of the overlapping area boundary simultaneously carry the original spatial location information and the dynamic information of the spatiotemporal feature sequence, thus completing the data integration.

[0045] Feature filtering was performed on the integrated data to extract key features related to fire alarm precursors: changes in diffusion rate were calculated by determining the maximum increase in rate within 30 seconds (the rate increase from 20-30 seconds to 10-20 seconds was 5 × 10⁻⁶). -4 -2×10 -4 ) / 2×10 -4 ×100%=150%) and rate fluctuation coefficient (standard deviation / mean = 1.58×10) -4 / 3.33×10 -4 The diffusion rate was determined to be "rapidly increasing and fluctuating significantly" by calculating the increase in area per unit time (15mm² / 10s = 1.5mm² / s in the 10th second, 15mm² / 10s = 1.5mm² / s in the 20th second, and 20mm² / 10s = 2mm² / s in the 30th second). The regional coverage growth rate was determined to be "accelerated in the later stage". Both characteristics were used as key features.

[0046] Based on the pre-set fire alarm precursor judgment model (this model is trained on 500 sets of historical electrical equipment fire alarm data, constructed using a logistic regression algorithm, with inputs being the spread rate increase, rate fluctuation coefficient, and area coverage increase, and outputting a probability value of 0-1), the model formula is as follows: The weights (0.3, 0.4, 0.3) were optimized using the gradient descent algorithm. 'a' represents the diffusion rate increase (normalized using min-max, 150% corresponds to a=0.75), 'b' is the rate fluctuation coefficient (normalized using Z-score, 0.47 corresponds to b=0.8), and 'c' is the area coverage increase rate (normalized using min-max, 2mm² / s corresponds to c=0.6). Substituting the extracted key features into the model calculation: P=0.3×0.75+0.4×0.8+0.3×0.6=0.225+0.32+0.18=0.725, which yields a quantified fire alarm precursor probability value of 0.725 (72.5%).

[0047] In step S16, if the fire alarm precursor probability value is higher than the preset alarm threshold, a polling mechanism is triggered to re-collect data by controlling the thermal imaging sensor and smoke sensor to generate a fusion dataset containing updated temperature gradients and particle concentrations.

[0048] In one implementation, this embodiment continues the monitoring scenario of a 10kV high-voltage distribution cabinet. After obtaining a fire alarm precursor probability value of 0.725 (72.5%), referring to the fire alarm warning threshold requirements for electrical equipment in GB50493-2019 "Design Standard for Detection and Alarm of Combustible and Toxic Gases in Petrochemical Industry", the thermal imaging sensor and smoke sensor are controlled step by step to re-collect data and generate a fused dataset. First, the previously deployed FLIRA655sc infrared thermal imaging sensor is controlled to re-collect temperature data for the original monitoring area (the overlapping area of ​​the upper A zone in the distribution cabinet and the surrounding extended range of 50 pixels, corresponding to the cabinet's physical coordinates x=0.6m~0.7m, y=0.8m~1.0m). The sampling frequency is increased to 60Hz (higher than the initial 30Hz, increasing the data density to capture subtle temperature changes), and the acquisition time is set to 10 seconds. The spatial location information of each acquisition point is recorded simultaneously (still the two-dimensional coordinates with the lower left corner of the distribution cabinet as the origin). The updated temperature gradient is calculated based on the collected temperature data: First, a 1024×768 two-dimensional temperature matrix T'(x,y) is constructed for each frame within 10 seconds (T'(x,y) represents the temperature value at the newly collected coordinates (x,y)). Then, the Sobel operator algorithm, consistent with the previous one, is used to calculate the horizontal gradient using the formula... Vertical gradient formula (One-sided difference correction for edge points) and total gradient formula The updated temperature gradient data is obtained, and each gradient value is bound to the spatial location identifier of the corresponding acquisition point (e.g., the gradient value corresponding to coordinates (550, 350) is 9.8℃ / mm).

[0049] Next, using the re-acquisition area of ​​the thermal imaging sensor as a reference, the MQ-135 smoke sensor array is controlled to synchronously re-acquire particle concentration data in the same area: the sensor sampling frequency is adjusted to 10Hz (matching the acquisition frequency of the thermal imaging sensor to ensure synchronization of time dimension data), and the acquisition duration is also 10 seconds, ensuring that there is corresponding concentration data at each acquisition time point (accurate to 0.1 seconds), and the recorded spatial location information is completely consistent with the spatial location identifier of the temperature gradient—for example, when the thermal imaging sensor acquires temperature data at coordinates (550, 350), the smoke sensor at the corresponding location synchronously acquires the particle concentration at that coordinate, resulting in an updated particle concentration sequence, such as a concentration of 19mg / m³ at time t=0.1 seconds and coordinates (550, 350), and a concentration of 20mg / m³ at the same coordinate at t=0.2 seconds, forming a sequence of data associated with "time-spatial coordinates-concentration value".

[0050] Subsequently, the updated temperature gradient data and particle concentration sequence were dimensionally calibrated: spatial location coordinate calibration was achieved through coordinate mapping verification, and the least squares method was used to correct the deviation, ensuring that the error of the spatial location markers of the two was ≤1 pixel (corresponding to a physical error of ≤0.5mm for the cabinet). For a few data with excessive deviations, linear interpolation was used for correction. The acquisition time dimension calibration was based on the timestamp of the thermal imaging sensor (the thermal imaging sensor has a built-in GPS module with a time accuracy of ±1ms). The timestamp of the smoke sensor data was adjusted through the NTP network time protocol to ensure that the time difference between the two at the same monitoring moment was ≤0.01 seconds, eliminating the data time deviation caused by sensor response delay, and finally ensuring that the two types of data were completely matched in the spatial and temporal dimensions.

[0051] Finally, the calibrated and matched temperature gradient data and particle concentration sequence were correlated and integrated: they were bound according to the preset format of "spatial coordinates-acquisition time-temperature gradient value-particle concentration value". For example, coordinates (550,350) and time t=0.1 seconds corresponded to "(550,350)-0.1s-9.8℃ / mm-19mg / m³", and coordinates (550,350) and time t=0.2 seconds corresponded to "(550,350)-0.2s-10.1℃ / mm-20mg / m³". All bound data were grouped by spatial coordinates and sorted in chronological order to clarify the data organization logic and organize them into a fusion dataset containing updated temperature gradients and particle concentrations. This dataset can be directly used for subsequent adjustment of thermal imaging resolution parameters and verification of the spatiotemporal correlation stability of smoke.

[0052] In step S17, based on the updated fusion dataset, key areas of temperature anomalies are located and the resolution parameters of the thermal imaging sensor are adjusted to form a monitoring loop image.

[0053] In one embodiment, this embodiment continues the monitoring scenario of a 10kV high-voltage switchgear. After acquiring a fused dataset containing updated temperature gradients and particle concentrations (e.g., coordinates (550, 350), time t=0.1 seconds corresponds to "9.8℃ / cm-19mg / m³", t=0.2 seconds corresponds to "10.1℃ / cm-20mg / m³"), the following operations are performed: locating key areas of temperature anomalies, adjusting thermal imaging sensor parameters, and forming monitoring cycle images.

[0054] The fused dataset was retrieved, and the updated temperature gradient data was extracted as the main focus. The abnormal fluctuation characteristics of the temperature gradient were analyzed using a sliding window algorithm (with supplementary algorithm parameters: window size set to 5 time points, step size 1 time point). Based on the gradient statistics of the distribution cabinet during normal operation (72 hours of continuous fault-free monitoring, mean temperature gradient 3.1℃ / cm, standard deviation 0.8℃ / cm), the normal temperature gradient range was set to 1.2℃ / cm~5℃ / cm (mean ± 2 times standard deviation, with a clear basis for threshold setting). When the temperature gradient at a certain spatial coordinate exceeds this range for 3 consecutive time points and the fluctuation amplitude is >2℃ / cm, it is determined to be a temperature anomaly. Analysis revealed that in the upper part of the distribution cabinet, at coordinates (x=550~570, y=340~360) (corresponding to pixel coordinates u=550~570, v=340~360), the temperature gradient increased from 9.8℃ / cm to 12.5℃ / cm within 10 seconds, with a fluctuation range of 2.7℃ / cm, which continuously exceeded the normal range. Therefore, this area was identified as a critical area of ​​temperature anomaly, with a spatial range of approximately 20×20 pixels (corresponding to the physical dimensions of the cabinet, 10mm×10mm).

[0055] Based on the spatial range of the critical temperature anomaly area (requiring high-precision monitoring in a small area) and the required monitoring accuracy (temperature difference capture accuracy needs to reach 0.1℃ to meet the need for early fault identification of subtle temperature differences), the resolution parameters of the FLIRA655sc infrared thermal imaging sensor were adjusted: the original sensor's default resolution was 1024×768 pixels. For the critical temperature anomaly area, a "region enhancement mode" was adopted. Through the sensor's built-in pixel interpolation algorithm (bilinear interpolation method, clearly defining the algorithm type), the local resolution of this area was increased to 2048×1536 pixels (the original resolution was maintained in non-local areas to balance data volume and processing efficiency). At the same time, the sensor's temperature measurement sensitivity parameters were adjusted, increasing the quantization bits of temperature acquisition from 14 bits to 16 bits. Through laboratory calibration and verification, the improved resolution can stably capture a temperature difference of 0.1℃ (supplementary verification of the effect after parameter adjustment), improving the temperature acquisition accuracy of this area from ±0.5℃ to ±0.1℃, fully adapting to the high-precision monitoring needs of small-scale anomaly areas and avoiding the loss of temperature difference details due to insufficient resolution.

[0056] After adjusting the parameters, the thermal imaging sensor continuously collects temperature data on the critical temperature anomaly area and its surrounding 50-pixel range (ensuring coverage of the possible expansion of the anomaly area). The acquisition frequency is maintained at 60Hz, and the acquisition duration is set to 60 seconds (covering a 1-minute monitoring cycle, which can capture dynamic changes while avoiding data redundancy, and clearly defines the duration setting logic). Simultaneously, 3600 consecutive temperature monitoring image frames are generated (each frame contains the temperature distribution of the local high-resolution anomaly area and the surrounding normal-resolution area, using pseudo-color mapping: <30℃ blue, 30℃~45℃ green, 45℃~60℃ yellow, >60℃ red, and the critical anomaly area is marked with a white dashed box, with the real-time temperature value marked inside the box (supplementing image annotation details)).

[0057] 3600 consecutive temperature monitoring images were sequentially sorted according to their acquisition time (from 0 seconds to 60 seconds). Simultaneously, particle concentration information at corresponding time points was extracted from the fused dataset (e.g., t=30 seconds, particle concentration of 25 mg / m³ at coordinates (x=560, 350) around the critical abnormal area). The particle concentration information around the critical abnormal area at that time point was supplemented by a "value + grayscale bar" format in the upper right corner of each temperature monitoring image (grayscale bar range 0~50 mg / m³, darker bars represent higher concentrations, clearly defining the grayscale bar mapping rules), forming a continuous dynamic image sequence. This image sequence was packaged in a "loop playback" format (60 seconds / loop), with the frame rate set to 25 frames / second (conforming to human visual comfort, clearly defining the frame rate setting basis). This generated a monitoring loop image reflecting the temperature change trend of the critical temperature abnormal area and the surrounding particle concentration status, which could be displayed in real time on the monitoring terminal. Maintenance personnel could intuitively grasp the dynamic changes in the abnormal area through the images (e.g., whether the temperature continues to rise, whether smoke is spreading to the surrounding area), and promptly judge the risk development trend.

[0058] refer to Figure 2 The second embodiment of the invention provides an intelligent monitoring system for electrical equipment based on intelligent sensors, comprising: Thermal image generation module: Collects temperature distribution data of the device casing surface through thermal imaging sensors and calculates temperature gradient values ​​to generate an initial thermal image containing the coordinates of the temperature rise area; Overlapping boundary determination module: acquires the particle concentration change sequence collected by the smoke sensor, and uses an image fusion algorithm to map the particle concentration change sequence to the pixel grid corresponding to the coordinates of the temperature rise area to determine the boundary of the overlapping area; Gas concentration generation module: Collects harmful gas spectral line data at the boundary of the overlapping region through a gas sensor, and generates a harmful gas concentration index by comparing the matching degree of the intensity of the harmful gas spectral line data with a preset spectral line threshold. Diffusion vector acquisition module: Based on the harmful gas concentration index, update the fused image layer in the pixel grid, and use frame-by-frame comparison to detect the dynamic correlation between the temperature rise trend and the smoke diffusion path in the fused image layer to obtain the diffusion path vector; Fire alarm probability calculation module: Extracts spatiotemporal feature sequence from the diffusion path vector, integrates the spatiotemporal feature sequence with the boundary of the overlapping area using an image fusion algorithm, and calculates the probability value of fire alarm precursor; Fusion Dataset Update Module: If the probability value of the fire alarm precursor is higher than the preset alarm threshold, a polling mechanism is triggered to re-collect data by controlling the thermal imaging sensor and the smoke sensor to generate a fusion dataset containing updated temperature gradient and particle concentration. Monitoring cycle image generation module: Based on the updated fusion dataset, it locates key areas of temperature anomalies and adjusts the resolution parameters of the thermal imaging sensor to generate monitoring cycle images.

[0059] It should be noted that the intelligent monitoring system for electrical equipment based on intelligent sensors provided in this embodiment of the invention is used to execute all the process steps of the intelligent monitoring method for electrical equipment based on intelligent sensors in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0060] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A smart sensor-based intelligent monitoring method of an electrical device, characterized by, The method comprises the following steps: acquire the temperature distribution data of the equipment shell surface by the thermal imaging sensor and calculate the temperature gradient value, generate the initial thermal image containing the temperature rising area coordinate; acquire the particle concentration change sequence collected by the smoke sensor, map the particle concentration change sequence to the pixel grid corresponding to the temperature rising area coordinate by using the image fusion algorithm, and determine the boundary of the overlapping area; acquire the harmful gas spectrum line data of the boundary of the overlapping area by the gas sensor, generate the harmful gas concentration index by comparing the matching degree of the harmful gas spectrum line data intensity and the preset spectrum line threshold value; update the fusion image layer in the pixel grid based on the harmful gas concentration index, and detect the dynamic correlation between the temperature rising trend and the smoke diffusion path in the fusion image layer by frame-by-frame comparison to obtain the diffusion path vector; extract the space-time feature sequence from the diffusion path vector, integrate the space-time feature sequence and the boundary of the overlapping area by using the image fusion algorithm, and calculate the fire alarm precursor probability value; if the fire alarm precursor probability value is higher than the preset alarm threshold value, trigger the polling mechanism, reacquire data by controlling the thermal imaging sensor and the smoke sensor, and generate the fusion data set containing the updated temperature gradient and particle concentration; based on the updated fusion data set, locate the temperature anomaly key area and adjust the resolution parameter of the thermal imaging sensor to form a monitoring cycle image.

2. The intelligent monitoring method of electrical equipment based on intelligent sensors according to claim 1, characterized in that, The method comprises the following steps: acquire the temperature distribution data of the equipment shell surface by the thermal imaging sensor and calculate the temperature gradient value, generate the initial thermal image containing the temperature rising area coordinate; acquire the temperature data of the equipment shell surface by the thermal imaging sensor, synchronously record the spatial position information of each acquisition point, and form the original temperature position correlation data containing the temperature distribution data; analyze the temperature change amplitude between adjacent acquisition points by using the temperature gradient calculation method, obtain the temperature gradient distribution result of the equipment shell surface, and mark the temperature rising area according to the preset temperature gradient threshold value and the recorded spatial position information; extract the spatial position information of the temperature rising area and convert it into corresponding coordinates, integrate it with the temperature distribution data in the original temperature position correlation data, and generate the initial thermal image.

3. The intelligent monitoring method of electrical equipment based on intelligent sensors according to claim 1, characterized in that, The method comprises the following steps: acquire the particle concentration change sequence collected by the smoke sensor, synchronously record the acquisition time and spatial position information corresponding to each concentration data, and form the particle concentration space-time correlation data; call the image fusion algorithm, take the pixel grid corresponding to the temperature rising area coordinate as the reference frame, and map the concentration change information in the particle concentration space-time correlation data to the corresponding position of the pixel grid point by point; superimpose and analyze the temperature rising area range data and the particle concentration distribution data in the mapped pixel grid, and identify the overlapping area covered by the two types of data in the pixel grid; According to the superposition analysis result, an edge contour of the overlapping area is extracted and determined to obtain a boundary of the overlapping area.

4. The intelligent monitoring method of electrical equipment based on intelligent sensors according to claim 1, characterized in that, The harmful gas spectral line data of the boundary of the overlapping area is collected by the gas sensor, and a harmful gas concentration index is generated by comparing the matching degree of the harmful gas spectral line data intensity with a preset spectral line threshold, including: According to the spatial range of the boundary of the overlapping area, the gas sensor is controlled to detect the harmful gas, and the harmful gas spectral line data is collected; The collected harmful gas spectral line data is preprocessed to remove spectral line clutter caused by environmental interference and retain effective spectral line data reflecting the true characteristics of the harmful gas; The intensity of the effective spectral line data is compared with the preset spectral line threshold, the deviation degree of the effective spectral line intensity relative to the preset spectral line threshold is calculated, and the matching degree analysis result of the two is formed; According to the matching degree analysis result, the deviation degree is converted into a quantitative harmful gas concentration index in combination with the corresponding relationship between the effective spectral line intensity and the harmful gas concentration.

5. The intelligent monitoring method of electrical equipment based on intelligent sensors as claimed in claim 1, wherein, The fusion image layer in the pixel grid is updated based on the harmful gas concentration index, and the dynamic correlation between the temperature rise trend and the smoke diffusion path in the fusion image layer is detected by frame-by-frame comparison to obtain a diffusion path vector, including: According to the harmful gas concentration index, a differentiated feature mark is set according to the concentration level, and is embedded in the fusion image layer of the pixel grid to replace the pixel data without concentration information, so that the fusion image layer synchronously contains temperature, smoke and harmful gas concentration information, and the update is completed. A continuous frame sequence is generated from the updated fusion image layer at a preset time interval, and the position coordinates of the temperature rise area, the change direction, the coverage range of smoke diffusion and the expansion direction are extracted from each frame. The dynamic correlation between the temperature rise trend and the smoke diffusion change in adjacent frames is detected by frame-by-frame comparison through spatial position overlap degree calculation and change rate correlation analysis. According to the dynamic correlation, the smoke diffusion trajectory highly correlated with the temperature rise trend is screened out, the direction and trajectory characteristics of the smoke diffusion trajectory are sorted out, and the diffusion path vector containing direction attributes and path nodes is converted.

6. The intelligent sensor-based electrical device monitoring method of claim 1, wherein, The space-time feature sequence is extracted from the diffusion path vector, and the space-time feature sequence is integrated with the boundary of the overlapping area by using an image fusion algorithm to calculate a fire alarm precursor probability value, including: The time variation law and the space extension characteristics in the smoke diffusion process are extracted from the diffusion path vector, and are sorted into a space-time feature sequence reflecting the diffusion dynamic process according to the monitoring time sequence. The dynamic feature information in the space-time feature sequence is embedded in the pixel coordinates of the boundary of the overlapping area by calling the image fusion algorithm and taking the spatial range of the boundary of the overlapping area as the reference framework, and the data integration is completed. The integrated data is subjected to feature screening to extract key features related to fire alarm precursors; wherein the key features at least include diffusion rate change and area coverage increase rate. The extracted key features are substituted into the fire alarm precursor judgment model for calculation according to the preset fire alarm precursor judgment model to obtain a quantitative fire alarm precursor probability value.

7. The intelligent monitoring method of electrical equipment based on intelligent sensors according to claim 1, characterized in that, The method comprises the following steps: The temperature data of the original monitoring area equipment shell surface is reacquired by controlling the thermal image sensor, and the spatial position information of the collection point is recorded synchronously. The updated temperature gradient is calculated based on the temperature data. The temperature gradient contains the spatial position identifier of the corresponding collection point. The particle concentration data of the same area is reacquired by controlling the smoke sensor based on the thermal image sensor collection area, and the collection time and spatial position information are recorded synchronously. The spatial position information is consistent with the spatial position identifier of the temperature gradient, and the updated particle concentration sequence is formed. The spatial position coordinates and collection time dimension of the updated temperature gradient data and the particle concentration sequence are calibrated to ensure complete matching and eliminate data deviation. The calibrated and matched temperature gradient data and the particle concentration sequence are associated and integrated, and are correspondingly bound in a preset format to form a fusion data set containing updated temperature gradient and particle concentration.

8. The intelligent monitoring method of electrical equipment based on intelligent sensors according to claim 1, characterized in that, Based on the updated fusion data set, the temperature abnormal key area is located and the thermal image sensor resolution parameter is adjusted to form a monitoring cycle image, which comprises the following steps: The updated fusion data set is called, and the updated temperature gradient data is extracted. The temperature abnormal key area whose temperature change exceeds the conventional range is located by analyzing the abnormal fluctuation characteristics of the temperature gradient. According to the spatial range of the temperature abnormal key area and the monitoring demand, the resolution parameter of the thermal image sensor is adjusted to adapt the temperature collection accuracy of the sensor to the monitoring demand. The thermal image sensor with the adjusted parameter continuously collects temperature data of the temperature abnormal key area and the surrounding area, and synchronously generates continuous temperature monitoring image frames. The continuous temperature monitoring image frames are arranged according to the collection time sequence, and the image is supplemented by combining the particle concentration information in the fusion data set to form a monitoring cycle image reflecting the state of the abnormal area.

9. An intelligent sensor based electrical equipment intelligent monitoring system, characterized in that, It comprises: The thermal image generation module generates an initial thermal image containing temperature rise coordinates by collecting the temperature distribution data of the equipment shell surface through the thermal image sensor and calculating the temperature gradient value. The overlapping boundary determination module obtains the particle concentration change sequence collected by the smoke sensor, maps the particle concentration change sequence to the pixel grid corresponding to the temperature rise coordinates by using the image fusion algorithm, and determines the overlapping area boundary. The gas concentration generation module collects the harmful gas spectrum line data of the overlapping area boundary through the gas sensor, compares the matching degree of the harmful gas spectrum line data intensity and the preset spectrum line threshold, and generates a harmful gas concentration index. The diffusion vector acquisition module updates the fusion image layer in the pixel grid based on the harmful gas concentration index, detects the dynamic correlation between the temperature rise trend and the smoke diffusion path in the fusion image layer by frame comparison, and obtains a diffusion path vector. The fire alarm probability calculation module extracts the space-time feature sequence from the diffusion path vector, integrates the space-time feature sequence and the overlapping area boundary by using the image fusion algorithm, and calculates a fire alarm precursor probability value. The fusion data set updating module: if the fire precursor probability value is higher than a preset alarm threshold, a polling mechanism is triggered, data is reacquired by controlling the thermal image sensor and the smoke sensor, and a fusion data set containing an updated temperature gradient and particle concentration is generated; The monitoring cycle image forming module: based on the updated fusion data set, a temperature anomaly key area is located and a thermal image sensor resolution parameter is adjusted, and a monitoring cycle image is formed.

Images