A power transmission line fire prediction method, system, device and storage medium

Abstract

The present disclosure relates to the technical field of power transmission lines, and particularly relates to a power transmission line fire prediction method, system, device and storage medium, wherein the method comprises: acquiring a multispectral image of a target area through a first observation satellite, and locating a fire point in the multispectral image; according to the spatial relationship between the fire point and the power transmission line, and selecting a fire point greater than a preset alarm level to be monitored through a second observation satellite to obtain a monitoring image; performing downscaling mapping on the multispectral image according to the pixel scale of the monitoring image to obtain theoretical radiance; performing ground object end member decomposition on the monitoring image pixels to obtain mixed pixel radiance; according to the position of the power transmission line, and based on the time sequence of the second observation satellite, the change in brightness temperature and the time interval of the monitoring, the spread speed of the fire point greater than the preset alarm level is calculated to obtain a fire condition prediction result; the time resolution and the spatial resolution are simultaneously considered, and the dynamic change of the fire condition is accurately reflected.

Description

Technical Field

[0001] This disclosure relates to the field of power transmission line technology, and in particular to a method, system, device and storage medium for predicting fires on power transmission lines. Background Technology

[0002] The threat to power grid security from wildfires is becoming increasingly serious. Existing wildfire monitoring systems typically suffer from the following problems: polar-orbiting satellites offer high spatial resolution but low temporal resolution, making rapid response difficult; geostationary satellites offer high temporal resolution but low spatial resolution, hindering the monitoring of small-scale wildfires; simultaneously, traditional methods lack precision in data processing and analysis, making accurate dynamic tracking of fire development challenging. Therefore, current volcano monitoring technologies struggle to balance temporal and spatial resolution and cannot accurately reflect the dynamic changes in fire conditions. Summary of the Invention

[0003] To address the aforementioned technical problems, this disclosure provides a method, system, device, and storage medium for predicting fires in power transmission lines.

[0004] This disclosure provides a method for predicting fires on power transmission lines, including: The first observation satellite acquires multispectral images of the target area and locates fire points in the multispectral images; Based on the spatial relationship between the fire point and the transmission line, the fire point is divided into multiple alarm levels according to a preset distance threshold, and fire points with alarm levels greater than the preset alarm level are selected for monitoring by a second observation satellite to obtain monitoring images. The multispectral image is downscaled according to the pixel scale of the monitored image to obtain the theoretical radiance; Based on the monitored image pixels, the composition of different land features is estimated according to the terrain, landform, geographical location and seasonal conditions, and the end-members of different land features are mixed in proportion to obtain the radiance of the mixed pixels. The deviation between the theoretical radiance and the mixed pixel radiance is corrected to obtain the second observation satellite time series; Based on the location of the transmission line and the spread capability of different ground features, and by calculating the spread rate of fire points exceeding the preset alarm level based on the brightness temperature changes of the second observation satellite time series and the monitoring time interval, fire prediction results are obtained.

[0005] Further, the step of downscaling the multispectral image according to the pixel scale of the monitored image to obtain the theoretical radiance includes: Obtain the radiance of the multispectral image pixels in the current band; Calculate the area ratio of the multispectral image pixels in the monitored image pixels; The theoretical radiance is calculated based on the radiance of the multispectral image pixels in the current band and the area ratio.

[0006] Furthermore, based on the monitored image pixels, the calculation of different land cover components according to vegetation index, topography, landform, geographical location, and seasonal conditions, and the proportional mixing of different land cover endmembers to obtain mixed pixel radiance, includes: Based on the monitoring images, different land features are estimated according to the terrain, landform, geographical location and seasonal conditions, and a vegetation index is calculated according to the vegetation conditions. If the vegetation index is within a preset range, the mixed pixel radiance is calculated by obtaining the area ratio of each land feature in the monitoring image pixels and the radiance of each land feature in the current band.

[0007] Further, the step of correcting the deviation between the theoretical radiance and the mixed pixel radiance to obtain the second observation satellite time series includes: Calculate the deviation between the theoretical radiance and the mixed pixel radiance; Observational parameters are acquired synchronously for each deviation sample; A feature matrix is ​​constructed based on the observed parameters, and the regression coefficients are obtained by solving the feature matrix using the least squares method. The prediction bias is calculated based on the regression coefficients, and the mixed pixel radiance is corrected.

[0008] Furthermore, the step of calculating the spread rate of fire points exceeding a preset alarm level based on the location of the transmission line and the spread capability of different ground features, and based on the brightness temperature changes of the second observation satellite time series and the monitoring time interval, to obtain fire prediction results includes: Based on the corrected brightness temperature time series of the second observation satellite, the brightness temperature difference between adjacent observation times is calculated to obtain the brightness temperature vector; Based on the brightness temperature vector and the spread capability of each ground feature, the spread rate of fire points exceeding the preset alarm level is calculated by detecting the time interval. Based on the spread speed, spread direction, and spatial relationship with the power transmission line, the fire spread risk and alarm level are obtained.

[0009] Furthermore, after acquiring the multispectral images and the monitoring images, geometric correction, radiometric correction, atmospheric correction, fire point identification, and fire point location are performed.

[0010] This disclosure also provides a system for predicting fires on transmission lines, including: The positioning module is used to acquire multispectral images of the target area through the first observation satellite and locate fire points in the multispectral images; The monitoring module is used to classify the fire point into multiple alarm levels according to the spatial relationship between the fire point and the transmission line and a preset distance threshold, and select fire points with alarm levels greater than the preset alarm level for monitoring by a second observation satellite to obtain monitoring images; The mapping module is used to downscale the multispectral image according to the pixel scale of the monitoring image to obtain the theoretical radiance; The hybrid simulation module is used to calculate the composition of different land features based on the monitored image pixels, according to the terrain, landform, geographical location and seasonal conditions, and to mix the end-members of different land features in proportion to obtain the radiance of the hybrid pixels. The correction module is used to correct the deviation between the theoretical radiance and the mixed pixel radiance to obtain the second observation satellite time series; The prediction module is used to calculate the spread rate of fire points exceeding the preset alarm level based on the location of the transmission line and the spread capacity of different ground features, and based on the brightness temperature change of the second observation satellite time series and the monitoring time interval, so as to obtain the fire prediction result.

[0011] Furthermore, the mapping module is specifically used for: Obtain the radiance of the multispectral image pixels in the current band; Calculate the area ratio of the multispectral image pixels in the monitored image pixels; The theoretical radiance is calculated based on the radiance of the multispectral image pixels in the current band and the area ratio.

[0012] This disclosure also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the method for predicting fires on power transmission lines.

[0013] This disclosure also provides a computer-readable storage medium storing computer-readable instructions that, when executed by a processor, implement the steps of the method for predicting fires on power transmission lines.

[0014] The technical solution provided in this disclosure has the following advantages compared with the prior art: Multispectral images of the target area are acquired by a first observation satellite, and fire points in the multispectral images are located. Based on the spatial relationship between the fire points and the transmission lines, the fire points are divided into multiple alarm levels according to a preset distance threshold. Fire points with alarm levels higher than the preset threshold are selected for monitoring by a second observation satellite to obtain monitoring images. The multispectral images are downscaled according to the pixel scale of the monitoring images to obtain theoretical radiance. Ground feature endmember decomposition is performed on the pixels of the monitoring images to obtain mixed pixel radiance. The deviation between the theoretical radiance and the mixed pixel radiance is corrected to obtain the second observation satellite time series. Based on the location of the transmission lines and the brightness temperature changes of the second observation satellite time series and the monitoring time interval, the spread rate of fire points with alarm levels higher than the preset threshold is calculated to obtain fire prediction results. This achieves simultaneous consideration of temporal and spatial resolution and accurately reflects the dynamic changes of the fire situation. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0016] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of a method for predicting fires on power transmission lines provided in an embodiment of this disclosure; Figure 2 A schematic diagram of a method for calculating theoretical radiance provided in an embodiment of this disclosure; Figure 3 This is a schematic diagram of a power transmission line fire prediction system provided in an embodiment of this disclosure. Detailed Implementation

[0018] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0020] Figure 1 This is a schematic diagram of a method for predicting fires on transmission lines provided in an embodiment of this disclosure; as shown.Figure 1 As shown, a method for predicting fires on power transmission lines includes: Step S1: Acquire multispectral images of the target area using the first observation satellite and locate the fire points in the multispectral images; In this embodiment, the first observation satellite is a high spatial resolution polar-orbiting or low-orbiting remote sensing platform. Its mission is to perform predetermined push-broom or array observations of the target area surrounding the power transmission line, acquiring multispectral images including short-wave visible, near-infrared, and mid-to-long-wave infrared channels. The acquisition process includes mission planning (determining the observation window, scanning path, and time slots), attitude and orbit control, and transmitting image data back to the ground station via laser communication or a traditional downlink. The acquired raw images must be accompanied by complete metadata—observation time, solar zenith / azimuth, satellite observation zenith / azimuth, sensor gain / bias, and calibration parameters, etc.—for subsequent processing. After receiving the image, preliminary preprocessing is performed on the ground, including bad pixel repair, radiometric response linearization or quadratic calibration, and coarse georegistration. Then, cloud and flare detection algorithms (such as cloud masks based on shortwave band thresholds or multispectral ratios) are used to remove human-caused and atmospheric interference pixels. Then, thermally anomalous pixels are identified in the mid- and long-wave infrared bands using brightness temperature thresholding or dynamic local thresholding. Combined with spatial connectivity analysis, area and shape filtering (to remove noisy hotspots smaller than a set number of pixels) and time series verification (if historical images exist, they are compared to remove transient radiometric interference), the candidate fire points are finally located and their confidence is evaluated. To improve positioning accuracy, rigorous geometric corrections are typically applied to satellite imagery. Ground control points or precise orbital data are used, and strict geometric models are applied to control the georeferenced error of the imagery to the pixel level or smaller. This enables high-precision detection and location of fire points, providing reliable input for subsequent graded alarms and synchronous satellite tracking. Simultaneously, complete metadata and preprocessing ensure traceable radiometric and geometric baselines during subsequent cross-sensor fusion, reducing uncertainties caused by sensor differences at the source. Furthermore, cloud, flare, and noise removal significantly reduce false alarm rates, thereby minimizing mis-triggered ground maintenance and satellite mission scheduling, saving manpower and response costs for power grid operators.

[0021] Step S2: Based on the spatial relationship between the fire point and the transmission line, the fire point is divided into multiple alarm levels according to a preset distance threshold, and fire points with alarm levels greater than the preset alarm level are selected for monitoring by the second observation satellite to obtain monitoring images. In this embodiment, the spatial relationship between identified fire points and transmission lines (including key assets such as towers, conductor corridors, and converter stations) is measured and risk-classified within a unified geographic reference system. This supports resource-constrained second observation satellites in conducting high-frequency, targeted monitoring of key targets. First, transmission line data is imported in vector form (tower coordinates, corridor buffer zones, historical tripping nodes, line load and importance weights, etc.). Then, in a GIS (Geographic Information System) environment, the shortest planar distance from each fire point to the line or the point-to-line distance to the nearest surrounding tower is calculated. Based on engineering and safety management requirements, several critical distance thresholds (e.g., 500 meters, 1500 meters, 3000 meters, 5000 meters) are preset to correspond to different alarm levels. Different threshold adjustment coefficients can be assigned to different line types or load levels (e.g., a more conservative distance threshold can be used for ultra-high voltage lines). After classifying fire points by level, the set of fire points exceeding the preset trigger level is prioritized. The prioritization criteria can comprehensively consider: fire point distance, fire point thermal intensity (brightness temperature), surrounding vegetation cover (NDVI (Normalized Difference Vegetation Index) or fuel load index), recent wind speed and direction forecasts, historical wildfire power outage records, and line importance weights. The prioritization results are used to generate observation task lists for the second observation satellite, including the observation center coordinates, required bands, observation window, repetition frequency, and observation duration. The second observation satellite is typically in a geostationary or inclined geosynchronous orbit, possessing high temporal resolution or flexible pointing capabilities. The ground task scheduling system will prioritize these tasks and, within available resources, strive to ensure continuous coverage of high-priority fire points. The acquired monitoring images have clear timestamps and metadata compared to multispectral images. Upon reception, radiometric and geometric preprocessing is required, and parameters such as observation angle, satellite zenith angle, and solar position must be recorded for subsequent cross-scale comparisons. By spatial hierarchical classification and priority scheduling, the monitoring frequency of fires around critical lines is significantly improved, enabling early detection of rapidly evolving fires. Furthermore, by combining multi-factor ranking with historical fault records, priority monitoring of fires posing the greatest threat to the power grid is achieved, thereby maximizing risk management benefits with limited satellite observation resources. At the same time, clear time reference samples are generated for subsequent downscaling mapping and time series analysis, ensuring that cross-sensor data fusion and time series inference have operable inputs and time references.

[0022] Step S3: Downscale the multispectral image according to the pixel scale of the monitoring image to obtain the theoretical radiance; In this embodiment, the high spatial resolution radiance information from the first observation satellite is converted into theoretical radiance that matches the observation scale of the second observation satellite, so that it can be directly compared and corrected with the actual observation values ​​of the second observation satellite. First, the pixel geometric correspondence between the two types of sensors needs to be determined: based on the spatial resolution and observation grid of both, the set of multiple first observation satellite pixels covered by each second observation satellite pixel on the ground is calculated, and the area ratio coefficient is calculated. The influence of observation angle, Earth curvature, and terrain needs to be considered. If necessary, a DEM (Digital Elevation Model) is introduced for projection correction to ensure the accuracy of pixel correspondence. Then, radiometric homogenization processing is performed on the high-resolution image: first, the DN (Digital Number) values ​​of the high-resolution image are converted into physical radiance or brightness temperature, using the previously performed secondary calibration relationship or sensor calibration coefficients, and then a weighted sum is calculated on each low-resolution pixel to be downscaled. To improve the reliability of theoretical radiance, high-resolution pixels can be filtered or noise suppressed (e.g., using median filtering or wavelet-based denoising) before weighted summation, and pixels obscured by clouds or strong reflective objects can be removed and interpolated. Furthermore, to adapt to temporal variations and sensor characteristics, it is recommended to construct a dynamic mapping parameter library (e.g., recording correction coefficients under different incident angles, seasons, and atmospheric conditions), and automatically call the corresponding parameter set in the production process based on current observation conditions. Mapping detailed information from high-resolution observations to a low-resolution observation scale in a physically consistent manner provides a radiance baseline that should theoretically be observed by the second satellite; it provides a clear reference for subsequent mixed pixel decomposition and radiation bias correction, making cross-sensor comparisons no longer simple pixel value comparisons but verification based on physical quantities; and uncertainty estimation provides a quantitative basis for error propagation and risk assessment, thereby supporting more robust fire situation assessment and alarm decisions.

[0023] Step S4: Based on the monitored image pixels, calculate the composition of different land features according to the terrain, landform, geographical location and seasonal conditions, and mix the end-members of different land features in proportion to obtain the radiance of the mixed pixels. In this embodiment, the mixed ground feature signals in each low spatial resolution pixel of the second observation satellite are separated into a combination of contributions from several endmembers (e.g., vegetation, bare land, buildings, fire spots / combustion zones, etc.). This allows the pixel-level radiance to be resolved into the abundance of various ground features and the radiance of endmembers, providing a clear compositional structure for bias analysis and physical interpretation. First, an endmember spectral library needs to be constructed or selected. Endmembers can be derived from the following approaches: one, extracting pure endmember samples from the study area using high-resolution first observation satellite imagery and clustering them for averaging; two, using ground-measured spectra or empirical spectral libraries; and three, employing a segmentation-based adaptive endmember extraction method. For areas with relatively simple surface cover or sparse vegetation (determined by NDVI; 0.2–0.5 can be classified as sparse vegetation), a linear mixture model is used, and the abundance is solved using non-negative constrained least squares. The solution results are then constrained and validated (e.g., threshold correction, spatial smoothing, and connectivity consistency checks) to remove outlier estimates. When dealing with complex land cover types or significant nonlinear interactions (e.g., building shadows, steep terrain, and areas of mixed smoke), a nonlinear mixture model is adopted, introducing second-order interaction terms. Parameters are solved iteratively using gradient descent or other optimization methods, and overfitting can be suppressed through regularization. To improve decomposition robustness, it is recommended to introduce spatial priors (e.g., constraining the smoothness of abundance within a local window) and temporal priors (utilizing the asymptotic nature of abundance in time series) during the solution process, and to use cross-validation or hold-out methods to evaluate the generalization performance of the endmember library. The final output of the mixed pixel radiance includes not only the total radiance but also the abundance distribution of each endmember and its estimation uncertainty. This information can be used to identify the proportion of fire points in low-resolution pixels, estimate the sub-pixel distribution of combustion intensity, and provide land cover structure constraints for subsequent bias correction. This transforms the observations from the second satellite from indivisible black boxes into quantities with definite physical composition, thus providing a basis for cross-scale alignment and energy conservation verification. Secondly, endmember abundance information can significantly improve the sensitivity of fire point sub-pixel detection, allowing the identification of small-scale but thermally intense combustion areas in low-resolution data. Finally, the decomposition results can be used as input for difference analysis, radiation correction, and subsequent fire spread modeling, enhancing the interpretability and operability of the overall monitoring system.

[0024] Step S5: Correct the deviation between the theoretical radiance and the mixed pixel radiance to obtain the second observation satellite time series; In this embodiment, the theoretical radiance obtained is aligned with the mixed pixel radiance obtained from the decomposition of the second observation satellite in terms of dimensions and physical meaning, eliminating systematic deviations introduced by observation time difference, observation angle difference, sensor response difference and atmospheric condition changes, thereby generating a unified, comparable and time-consistent second observation satellite brightness temperature / radiance time series. First, sample bias calculation: bias samples are calculated pixel-by-pixel and band-by-band to form a bias sample set, while recording the accompanying characteristics of each sample. Second, feature acquisition: multi-source parameters that may affect the bias are simultaneously recorded as feature vectors, including observation time difference, satellite observation zenith and azimuth angles, solar zenith angle, aerosol optical thickness, atmospheric water vapor content, land cover type coding, fire point sub-pixel area, and average brightness temperature. Third, regression model establishment and solution: linear or extended polynomial regression is used, and coefficients are solved using least squares or regularized least squares. Cross-validation is used to select the model order and regularization parameters to prevent overfitting. Fourth, bias prediction and correction: real-time second-observation satellite observation samples are substituted into the regression coefficients obtained from training to calculate the predicted bias, resulting in corrected radiance or brightness temperature. Fifth, uncertainty assessment and adaptive updating: the uncertainty of each correction is estimated based on the training residual distribution, and the regression coefficients are periodically updated using incremental learning or sliding window retraining strategies when new observation data arrives to adapt to changes in statistical characteristics under seasonality, sensor aging, or extreme weather conditions. First, it achieves consistency across sensor dimensions, making theoretical expectations and actual observations comparable at the physical level. Second, it concretizes abstract multi-factor correction into engineeringable steps through explicit feature acquisition and regression solutions, making it easy to implement and verify. Third, it improves the reliability and long-term stability of time series products through uncertainty output and model adaptation mechanisms, thereby providing a reliable data foundation for subsequent dynamic trend analysis and fire prediction.

[0025] Step S6: Based on the location of the transmission line and the spread capability of different ground features, and based on the brightness temperature change of the second observation satellite time series and the monitoring time interval, calculate the spread rate of fire points greater than the preset alarm level, and obtain the fire prediction result.

[0026] In this embodiment, the obtained corrected brightness temperature time series of the second observation satellite is converted into a dynamic spread index and used for fire prediction and line risk assessment. The first step is time series registration and difference calculation: Under the premise that geometric correction and time synchronization are performed for each observation, the brightness temperature difference vector D (which can be the brightness temperature difference of a single pixel or a spatial brightness temperature gradient field with the fire center as the core) is calculated for the same spatial location or through spatial weighted aggregation (considering the change in fire abundance within a pixel) at adjacent observation times. Then, the initial spread rate is calculated based on the time interval between two observations. If a spatial gradient is used... The system can combine motion estimation methods for brightness temperature contour lines (such as optical flow or isosurface tracing methods) to infer the spread direction and rate. Then, it couples the spread rate with the geometric relationship of the transmission line: intersecting the spread direction vector with the line corridor, it calculates the estimated arrival time to the line or the nearest tower, and adjusts the confidence interval of the arrival time based on fire intensity, wind speed / direction forecasts, surface fuel load, and suppression conditions (such as water bodies and road zones). If the arrival time is less than a preset response threshold or the expected heat radiation / flame length exceeds the line's tolerance limit, it triggers an upgrade of the alarm level or generates emergency inspection / power outage recommendations. This system can identify spread events that pose a substantial threat to transmission lines in advance, providing maintenance units with a time window for response. Furthermore, by outputting confidence and uncertainty levels, it supports decision-makers in prioritizing resources (e.g., prioritizing drone inspections, ground teams, or initiating line protection measures), thereby achieving a balance between ensuring grid safety and reducing unnecessary power outages, and significantly improving the foresight and accuracy of fire response.

[0027] In some possible ways of implementation, Figure 2 This is a schematic diagram of the method for calculating theoretical radiance provided in an embodiment of this disclosure; as shown. Figure 2 As shown, step S3, which involves downscaling the multispectral image according to the pixel scale of the monitoring image to obtain the theoretical radiance, includes: step S31, obtaining the radiance of the multispectral image pixel in the current band; step S32, calculating the area ratio of the multispectral image pixel in the monitoring image pixel; and step S33, calculating the theoretical radiance based on the radiance and area ratio of the multispectral image pixel in the current band.

[0028] In this embodiment, downscaling first obtains the radiance values ​​of each high-resolution pixel in the multispectral image acquired by the first observation satellite at the current analysis band, and performs unit unification and radiation normalization on the radiance to ensure the comparability of physical quantities between different sensors. Then, based on the pixel grid structure of the second observation satellite's monitoring image, a spatial overlay relationship is established, and the area ratio of each high-resolution multispectral pixel within the corresponding monitoring image pixel range is calculated. Area weighting coefficients are obtained through vector boundary clipping or resampling raster overlay. Based on this, according to the principle of energy conservation, the radiance values ​​of multiple multispectral pixels falling within the same monitoring image pixel range are weighted and summed according to area ratios to calculate the theoretical radiance value of that monitoring image pixel in the current band. Through these steps, scale unification from high spatial resolution data to low spatial resolution data can be achieved without changing the physical energy distribution characteristics, thereby constructing a theoretical radiation benchmark consistent with the observation scale of the second observation satellite. On the one hand, it avoids the radiation distortion problem caused by simple interpolation or averaging, ensuring that the fire point energy information remains physically consistent during the scale conversion process; on the other hand, it provides comparable theoretical reference values ​​for subsequent radiation deviation analysis, improves the accuracy and stability of cross-sensor data fusion, and enhances the ability to quantitatively analyze changes in fire point intensity.

[0029] In some possible implementations, step S4, based on the monitored image pixels, calculates the composition of different land features according to vegetation index, topography, landform, geographical location and seasonal conditions, and mixes the end-members of different land features in proportion to obtain the radiance of the mixed pixels, including: based on the monitored image, calculating different land features according to topography, landform, geographical location and seasonal conditions, and calculating the vegetation index according to the vegetation conditions; if the vegetation index is within a preset range, calculating the radiance of the mixed pixels by obtaining the area ratio of each land feature in the monitored image pixels and the radiance of each land feature in the current band.

[0030] In this embodiment, the land cover endmember decomposition first calculates the vegetation index based on the monitoring image of the second observation satellite. For example, the normalized vegetation index is calculated using the red and near-infrared bands, and the calculation result is compared with a preset vegetation index range to determine the complexity of the land cover of the current pixel. When the vegetation index is within the preset range, it indicates that there are mixed vegetation and non-vegetation features in the area. At this time, the area proportion information of various land cover types within the monitoring image pixel is obtained. This area proportion can be obtained through historical land cover data or inversion from high-resolution imagery of the first observation satellite. At the same time, the standard radiance or endmember radiance value of each land cover in the current band is obtained, and the radiance of each land cover is weighted and superimposed according to the area proportion based on the linear mixing model to calculate the mixed pixel radiance of the monitoring image pixel. Through this processing, the mixed pixel radiative response that was originally caused by insufficient spatial resolution can be decomposed and reconstructed, making it closer to the real surface composition structure. It effectively reduces the interference of complex surface background on the radiation characteristics of fire points and improves the ability to distinguish small-scale fire source signals; at the same time, it provides a more accurate observation basis for subsequent analysis of the difference between theoretical radiance and actual observed radiance, thereby enhancing the sensitivity and reliability of fire monitoring.

[0031] In some possible implementations, step S5, correcting the deviation between the theoretical radiance and the mixed pixel radiance to obtain the second observation satellite time series, includes: calculating the deviation between the theoretical radiance and the mixed pixel radiance; synchronously acquiring observation parameters for each deviation sample; constructing a feature matrix based on the observation parameters and solving the feature matrix using the least squares method to obtain regression coefficients; calculating the prediction deviation based on the regression coefficients and correcting the mixed pixel radiance.

[0032] In this embodiment, the difference between the theoretical radiance and the mixed pixel radiance is calculated to obtain radiation deviation samples at each observation time and in each band. Subsequently, corresponding observation parameters are synchronously acquired for each deviation sample, including observation time difference, observation zenith angle, atmospheric aerosol optical thickness, water vapor content, land cover type code, fire point area, and fire point brightness temperature. These parameters are then combined to form a feature vector to construct a feature matrix. Based on this, the regression relationship between the feature matrix and the deviation samples is solved using the least squares method to obtain the regression coefficients of each influencing factor. Furthermore, the regression coefficients are used to calculate the prediction deviation of the mixed pixel radiance of the current monitoring image and correct it, forming a second observation satellite brightness temperature time series at a unified physical scale. This allows for quantitative compensation of systematic errors caused by differences in observation angles, changes in atmospheric conditions, and time misalignment between different sensors, thereby constructing continuous, stable, and comparable time series data. This improves the consistency and comparability of cross-temporal data, reduces spurious brightness temperature fluctuations, improves the accuracy of fire evolution trend analysis, and provides a reliable data foundation for subsequent spread rate calculations.

[0033] In some possible implementations, step S6, based on the location of the transmission line and the spread capability of different ground features, and based on the brightness temperature change of the second observation satellite time series and the monitoring time interval, calculates the spread rate of fire points exceeding the preset alarm level to obtain fire prediction results, including: calculating the brightness temperature difference between adjacent observation times based on the corrected brightness temperature time series of the second observation satellite to obtain a brightness temperature vector; calculating the spread rate of fire points exceeding the preset alarm level based on the brightness temperature vector and the spread capability of each ground feature through the detection time interval; and obtaining the fire spread risk and alarm level based on the spread rate, spread direction, and spatial relationship with the transmission line.

[0034] In this embodiment, based on the corrected brightness temperature time series data from the second observation satellite, the brightness temperature difference between two adjacent observation times is calculated to form a brightness temperature change vector. This vector contains both the amplitude of the brightness temperature change and the spatial distribution direction information. Subsequently, combined with the corresponding monitoring time interval, the brightness temperature change amplitude is time-normalized to obtain the brightness temperature growth rate per unit time. This growth rate is then overlaid with the spatial location of the fire point to calculate the spread speed and direction of the fire point. Furthermore, considering the spatial distance relationship between the fire point and the transmission line, it is determined whether the spread direction is towards the transmission line. The fire spread risk level and corresponding alarm level are determined by comprehensively considering the spread speed and spatial proximity. This transforms simple brightness temperature change data into dynamic fire parameters with spatial physical significance, enabling predictive analysis of fire development trends. It shifts from static fire point identification to dynamic risk assessment, improving the early warning lead time. Simultaneously, through coupling analysis with the spatial relationship of the transmission line, the risk assessment becomes more targeted and has practical engineering value, thereby enhancing the disaster prevention capabilities of the transmission channel.

[0035] In some possible implementations, geometric correction, radiometric correction, atmospheric correction, fire identification, and fire location are performed after acquiring multispectral and monitoring images.

[0036] In this embodiment, after acquiring data from the first and second observation satellites, geometric correction is first performed on both types of images. Ground control point matching or orbital parameter calculation eliminates geometric distortions caused by sensor attitude and terrain undulations, achieving precise spatial registration. Subsequently, radiometric correction is performed, converting digital quantization values ​​into physical radiance values ​​and performing sensor response consistency processing. Further atmospheric correction is performed based on an atmospheric radiative transfer model to eliminate atmospheric scattering and absorption effects, ensuring the image data reflects the true surface radiation characteristics. On this basis, fire point identification is performed using mid-infrared or thermal infrared band thresholding methods, and high-reflectivity buildings or solar reflection interference is eliminated through multi-band discrimination rules, ultimately determining the spatial coordinates of the fire point. This ensures consistency and accuracy of the input data in terms of spatial location, radiation scale, and atmospheric conditions. It improves the accuracy and reliability of fire point identification, reduces false alarm and false negative rates, and provides high-quality basic data for subsequent downscaling mapping, hybrid pixel decomposition, and time series analysis, thereby enhancing the overall stability and engineering application value of the fire monitoring and risk prediction system.

[0037] In some possible implementations, polar-orbiting satellites perform high-resolution push-broom scans of the area near the power grid according to the mission plan, acquire multispectral images, and transmit the images to the ground via a laser communication link.

[0038] The ground receiving and telemetry integrated system processes polar-orbiting satellite data, employing rigorous geometric models and GCPs (Ground Control Points) for coarse and fine geometric corrections; it uses a quadratic calibration formula for radiometric correction; it utilizes the Modtran model (Moderate Resolution Atmospheric Transmission) for atmospheric correction; and it uses the medium- and long-wave threshold method (with short waves processed for flare and cloud masking) to identify and locate fire points.

[0039] L=a DN2+b DN+c; The formula is a quadratic calibration formula, where L is the radiance and DN is the infrared image output. Since the spaceborne infrared imaging system has linear and nonlinear effects within the dynamic response range, the quadratic model can better restore the infrared image output value to its energy form.

[0040] Simultaneously with issuing the alarm, select the fire area near the important line and issue synchronous satellite observation instructions. The alarm levels are divided into four levels: Level 1 (fire point and line distance 500m), Level 2 (fire point and line distance 1500m), Level 3 (fire point and line distance 3000m), and Level 4 (fire point and line distance 5000m). In the polar-orbiting satellite infrared image, each level is distributed as follows: N Level 1 alarms X{X1, X2, X3, X4, ..., Xn}; I Level 2 alarms X{X1, X2, X3, X4, ..., Xi}; J Level 3 alarms X{X1, X2, X3, X4, ..., Xj}; K Level 4 alarms X{X1, X2, X3, X4, ..., Xk}. Based on data from dense power grid channels, historical wildfire tripping lines, and level 1 and 2 alarm fire points, it is known that P important fires are selected for continuous dynamic observation according to priority level, such as PX{PX1, PX2, PX3, PX4, ..., Xp}, where P≤N+I. Each fire point PX1 includes the fire point's latitude and longitude, the latitude and longitude of surrounding towers within 1 km, the fire point's reflectivity in shortwave, and the brightness temperature value in medium and longwave. The boundary of the fire point's latitude and longitude is extended by 2° to obtain the study area containing all PX fire points. The mission instructions for observation location and duration are sent to geostationary satellites (the duration is manually confirmed based on the fire intensity). The reception and preprocessing of geostationary satellite data includes geometric correction, radiometric correction, and atmospheric correction, and the processing method is the same as that for polar-orbiting satellites.

[0041] Polar-orbiting satellite infrared images are downscaled using a formula to reduce the spatial resolution (60m-375m) to that of geostationary satellites (1-2km). Simultaneously, the radiance before downscaling is also weighted by abundance to obtain the downscaled radiance, which serves as the theoretical geostationary satellite data. Then, the geostationary satellite imagery is decomposed into mixed pixels, such as trees, landslides, buildings, and wildfires, which serves as the test satellite data. Finally, based on the actual test results and theoretical data, polynomials are used to perform radiometric bias correction. Through this series of steps, the radiance variation of fire points in polar-orbiting satellite infrared images can be extrapolated to geostationary satellite data, thereby enabling dynamic monitoring of fire points from geostationary satellite infrared data.

[0042] The downscaling formula for converting polar-orbiting satellite imagery to geostationary satellite imagery (multiple pixels merged) is as follows: ; R(k, λ): Radiance of the k-th geostationary satellite in band λ after downscaling (W / m² / sr / μm); fi: Area proportion (abundance) of the i-th polar-orbiting satellite pixel in the geostationary satellite pixel; R(i, λ): Radiance of the i-th polar-orbiting satellite pixel in band λ (W / m² / sr / μm). Model error.

[0043] The mixed pixels from geostationary satellites are decomposed according to fire and different land cover types. The geostationary satellite mixed pixel decomposition model includes either a linear or nonlinear mixed model. For sparse vegetation cover areas (identified by the vegetation index NDVI, 0.2-05 is considered sparse, NDVI = (near-infrared reflectance - infrared reflectance) / (near-infrared reflectance + infrared reflectance)), a linear mixed model can be selected, and then a non-negative constrained least squares method is used (with the added constraint of non-negativity of abundance (Fi≥0)). The linear model is as follows: R(λ)=F(Fi LF(i,λ))+ε(λ); For other complex regions, a nonlinear hybrid model can be selected, and then the gradient descent method can be used for iterative solution. The nonlinear model is as follows: R(λ)=F(Fi LF(i, λ))+F(B(i, j) LF(i,λ)·LF(j,λ))+ε(λ); R(λ): Radiance of the mixed pixel observed by the geostationary satellite in band λ (W / m² / sr / μm); Fi: Area proportion (abundance) of the i-th land feature, representing the proportion of the land feature in the pixel (0≤Fi≤1, and ΣFi=1); LF(i,λ): Radiance of the i-th land feature (endmember) in band λ (W / m² / sr / μm); ε(λ): Residual term, representing model error, such as error caused by noise, atmospheric influence, etc.; LF(j,λ): Radiance of the j-th land feature (endmember) in band λ (W / m² / sr / μm); B(i,j): Nonlinear parameter, representing the interaction strength between the i-th and j-th land features.

[0044] Due to factors such as downscaling, time differences, sensor variations, and atmospheric effects, there is a discrepancy between the theoretical radiance and the radiance actually observed by geostationary satellites. This discrepancy needs to be corrected to improve monitoring accuracy. The formula is as follows: ΔR(λ)=R(observed,λ)-R(theoretical,λ); ΔR(λ): Radiance deviation; R(observed, λ): Radiance actually observed by the geostationary satellite; R(theoretical, λ): Theoretical Radiance calculated using the hybrid pixel model; A large amount of historical data, including time difference, observation angle difference, atmospheric parameters (e.g., aerosol optical thickness, water vapor content), land cover type, fire point size, and temperature, was used to establish a regression model (polynomial) for correction. The regression model is as follows: ΔR(λ) = f(time difference, observation angle difference, atmospheric parameters, land cover type, fire point size, fire point temperature) + ε; Based on the location of the transmission lines, long-term sequence analysis using geostationary satellites was used to monitor changes in the brightness temperature (direction and magnitude) of the fire points and assess the risk of fire spread and alarms in important transmission line corridors. The fire was at its peak on the 16th, and Level 1 alarms were issued for surrounding lines. From the 17th to the 18th, the fire fluctuated in size but did not spread further, and the alarms reached Level 2-3. On the 19th, the fire weakened and cloud cover increased, with the possibility of rainfall, and the alarms reached Level 3-4.

[0045] Figure 3 This is a schematic diagram of a transmission line fire prediction system provided in an embodiment of this disclosure; as shown. Figure 3 As shown, this disclosure also provides a system for predicting fires on transmission lines, comprising: The system comprises the following modules: a positioning module 401, which acquires multispectral images of the target area via a first observation satellite and locates fire points within the multispectral images; a monitoring module 402, which classifies fire points into multiple alarm levels based on the spatial relationship between the fire points and the transmission lines, according to a preset distance threshold, and selects fire points with alarm levels higher than the preset threshold for monitoring via a second observation satellite to obtain monitoring images; a mapping module 403, which downscales the multispectral images according to the pixel scale of the monitoring images to obtain theoretical radiance; a hybrid simulation module 404, which calculates the composition of different land features based on the monitoring image pixels, considering topography, landforms, geographical location, and seasonal conditions, and mixes different land feature endmembers proportionally to obtain mixed pixel radiance; a correction module 405, which corrects the deviation between the theoretical radiance and the mixed pixel radiance to obtain a second observation satellite time series; and a prediction module 406, which calculates the spread rate of fire points with alarm levels higher than the preset threshold based on the location of the transmission lines and the spread capacity of different land features, and on the brightness temperature change and monitoring time interval of the second observation satellite time series, to obtain fire prediction results.

[0046] In some possible implementations, the mapping module is specifically used to: obtain the radiance of multispectral image pixels in the current band; calculate the area ratio of multispectral image pixels in the monitored image pixels; and calculate the theoretical radiance based on the radiance and area ratio of multispectral image pixels in the current band.

[0047] This disclosure also provides a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement steps for a method for predicting fires on power transmission lines.

[0048] This disclosure also provides a computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement steps for a method for predicting fires on power transmission lines.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0050] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting fires in power transmission lines, characterized in that, include: The first observation satellite acquires multispectral images of the target area and locates fire points in the multispectral images; Based on the spatial relationship between the fire point and the transmission line, the fire point is divided into multiple alarm levels according to a preset distance threshold, and fire points with alarm levels greater than the preset alarm level are selected for monitoring by a second observation satellite to obtain monitoring images. The multispectral image is downscaled according to the pixel scale of the monitored image to obtain the theoretical radiance; Based on the monitored image pixels, the composition of different land features is estimated according to the terrain, landform, geographical location and seasonal conditions, and the end-members of different land features are mixed in proportion to obtain the radiance of the mixed pixels. The deviation between the theoretical radiance and the mixed pixel radiance is corrected to obtain the second observation satellite time series; Based on the location of the transmission line and the spread capability of different ground features, and by calculating the spread rate of fire points exceeding the preset alarm level based on the brightness temperature changes of the second observation satellite time series and the monitoring time interval, fire prediction results are obtained.

2. The method for predicting fires on transmission lines according to claim 1, characterized in that, The step of downscaling the multispectral image according to the pixel scale of the monitored image to obtain the theoretical radiance includes: Obtain the radiance of the multispectral image pixels in the current band; Calculate the area ratio of the multispectral image pixels in the monitored image pixels; The theoretical radiance is calculated based on the radiance of the multispectral image pixels in the current band and the area ratio.

3. The method for predicting fires on transmission lines according to claim 1, characterized in that, Based on the monitored image pixels, and according to vegetation index, topography, landform, geographical location, and seasonal conditions, the composition of different land features is calculated, and the endmembers of different land features are mixed in proportion to obtain the radiance of the mixed pixels, including: Based on the monitoring images, different land features are estimated according to the terrain, landform, geographical location and seasonal conditions, and a vegetation index is calculated according to the vegetation conditions. If the vegetation index is within a preset range, the mixed pixel radiance is calculated by obtaining the area ratio of each land feature in the monitoring image pixels and the radiance of each land feature in the current band.

4. The method for predicting fires on transmission lines according to claim 1, characterized in that, The step of correcting the deviation between the theoretical radiance and the mixed pixel radiance to obtain the second observation satellite time series includes: Calculate the deviation between the theoretical radiance and the mixed pixel radiance; Observational parameters are acquired synchronously for each deviation sample; A feature matrix is ​​constructed based on the observed parameters, and the regression coefficients are obtained by solving the feature matrix using the least squares method. The prediction bias is calculated based on the regression coefficients, and the mixed pixel radiance is corrected.

5. The method for predicting fires on transmission lines according to claim 1, characterized in that, The method of calculating the spread rate of fire points exceeding a preset alarm level based on the location of the transmission line and the spread capacity of different ground features, and based on the brightness temperature changes of the second observation satellite time series and the monitoring time interval, to obtain fire prediction results includes: Based on the corrected brightness temperature time series of the second observation satellite, the brightness temperature difference between adjacent observation times is calculated to obtain the brightness temperature vector; Based on the brightness temperature vector and the spread capability of each ground feature, the spread rate of fire points exceeding the preset alarm level is calculated by detecting the time interval. Based on the spread speed, spread direction, and spatial relationship with the power transmission line, the fire spread risk and alarm level are obtained.

6. The method for predicting fires on transmission lines according to any one of claims 1 to 5, characterized in that, After acquiring the multispectral images and the monitoring images, geometric correction, radiometric correction, atmospheric correction, fire point identification, and fire point location are performed.

7. A fire prediction system for power transmission lines, characterized in that, include: The positioning module is used to acquire multispectral images of the target area through the first observation satellite and locate fire points in the multispectral images; The monitoring module is used to classify the fire point into multiple alarm levels according to the spatial relationship between the fire point and the transmission line and a preset distance threshold, and select fire points with alarm levels greater than the preset alarm level for monitoring by a second observation satellite to obtain monitoring images; The mapping module is used to downscale the multispectral image according to the pixel scale of the monitoring image to obtain the theoretical radiance; The hybrid simulation module is used to calculate the composition of different land features based on the monitored image pixels, according to the terrain, landform, geographical location and seasonal conditions, and to mix the end-members of different land features in proportion to obtain the radiance of the hybrid pixels. The correction module is used to correct the deviation between the theoretical radiance and the mixed pixel radiance to obtain the second observation satellite time series; The prediction module is used to calculate the spread rate of fire points exceeding the preset alarm level based on the location of the transmission line and the spread capacity of different ground features, and based on the brightness temperature change of the second observation satellite time series and the monitoring time interval, to obtain fire prediction results.

8. The fire prediction system for power transmission lines according to claim 7, characterized in that, The mapping module is specifically used for: Obtain the radiance of the multispectral image pixels in the current band; Calculate the area ratio of the multispectral image pixels in the monitored image pixels; The theoretical radiance is calculated based on the radiance of the multispectral image pixels in the current band and the area ratio.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the method for predicting fires on transmission lines as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the method for predicting fires on transmission lines as described in any one of claims 1 to 6.

Images