Power transmission line forest fire prediction method and device based on multi-modal data fusion

Abstract

The invention relates to the technical field of power transmission line forest fire monitoring and early warning, and discloses a power transmission line forest fire prediction method and device based on multi-modal data fusion. The method comprises the following steps: acquiring fixed images based on a plurality of fixed cameras, splicing the fixed images into a full image, identifying global forest fire characteristics from the full image, calculating a gathering degree value of a global forest fire position and a generation speed value, a disappearance speed value and a development degree value in a preset first time period, and further obtaining a forest fire prediction value. If the predicted value is greater than the reference predicted value, calculating the central position of a gathering area, calling a mobile camera to move to the position, shooting a plurality of mobile area images during movement, identifying local mountain fire features from the mobile area images, calculating a local gathering area, and calculating a gathering area movement vector according to the plurality of local gathering areas; and when the local aggregation area corresponds to the central position and the central position is on the aggregation area movement vector, giving out an early warning. According to the method, through multi-modal data fusion and dynamic monitoring, the monitoring precision is improved, and the false alarm rate is reduced.

Description

Technical Field

[0001] The present application relates to the technical field of transmission line wildfire monitoring and early warning, and in particular to a transmission line wildfire prediction method and device using multimodal data fusion. Background Art

[0002] Wildfires on power transmission lines pose a significant threat to the safe operation of power systems. When a wildfire occurs, the intense flames and dense smoke can damage the insulation of power lines, causing conductors to burn, tripping circuits, and even widespread power outages. Furthermore, the spread of wildfires endangers the lives and property of surrounding residents. Fire control is difficult, and restoring power is time-consuming, resulting in significant economic losses. Currently, there are two main methods for predicting wildfires along power transmission lines. The first is based on meteorological data. This involves collecting weather parameters such as temperature, humidity, and wind speed, and combining them with historical wildfire data to build models to predict the probability of wildfire occurrence. The second is based on video surveillance. Tower-mounted cameras capture real-time footage of the power transmission line area, allowing for manual or intelligent fire detection. The former allows for early risk assessment of a large area, while the latter allows for direct detection of on-site fire signs. However, these two conventional approaches share significant shortcomings: Forecasts based on meteorological data are susceptible to factors such as local terrain and vegetation, making them inaccurately reflecting specific risks around the lines. Monitoring based on video surveillance is limited by weather (such as fog and heavy rain) and lighting conditions, making it prone to recognition delays or misjudgments. Both rely on a single data source, and the accuracy of the resulting wildfire data needs to be improved. Summary of the Invention

[0003] In order to improve the accuracy of wildfire data, the present application provides a transmission line wildfire prediction method and device based on multimodal data fusion.

[0004] In the first aspect, the present application provides a method for predicting wildfires on power transmission lines using multimodal data fusion, which adopts the following technical solutions: A multimodal data fusion method for predicting wildfires on power transmission lines comprises the following steps: Acquire multiple fixed images based on multiple fixed cameras, where the fixed cameras are located at fixed positions in different locations; stitch and fuse the multiple fixed images into a fixed full-frame image; Identifying global wildfire features from the fixed full-frame image, outputting the identified global wildfire features and corresponding global wildfire locations, and calculating a concentration level value for the global wildfire locations; calculating a generation rate value and a disappearance rate value of the global wildfire features within a preset first time period, and calculating a development level value based on the generation rate value and the disappearance rate value; and calculating a wildfire prediction value based on the concentration level value and the development level value. If the wildfire prediction value is greater than a preset reference prediction value, the center position of the global wildfire location gathering area is calculated, and a mobile camera is called to move to the center position; The mobile camera captures a plurality of moving area images at different positions when moving, and the positions of the moving area images are located in the area corresponding to the fixed full-frame image; Identifying local wildfire features based on the moving area image, outputting the identified local wildfire features and corresponding local wildfire locations, and calculating a localized concentration area of the local wildfire features based on the local wildfire locations; Calculating a plurality of local gathering areas based on the plurality of moving area images, and calculating a gathering area movement vector according to positions of the plurality of local gathering areas; wherein the direction of the gathering area movement vector is opposite to the direction of the movement vector of the moving camera; If the local gathering area corresponds to the center position and the center position is on the moving vector of the gathering area, a wildfire warning is issued.

[0005] By adopting the above technical solutions, through multimodal data fusion and dynamic monitoring mechanisms, a dual verification system is constructed from global early warning to local precision judgment, improving monitoring accuracy, reducing false alarm rates, and optimizing resource allocation. Specifically, fixed cameras are used to stitch full-frame images to achieve large-scale monitoring, and quantitative calculations are combined with wildfire characteristics to predict risks. When the risk is high, mobile cameras are called to focus on the core area, and accurate data is obtained through local feature recognition and cluster area analysis. The movement vector of the cluster area is constrained to be opposite to the movement vector of the mobile camera, and dual verification is combined with the position overlap condition, greatly reducing the false alarm rate. At the same time, fixed and mobile devices work together to activate resources on demand, effectively saving costs.

[0006] Optionally, before comparing the wildfire prediction values, the following sub-steps are further included: Searching for an optical fiber temperature measuring device within a preset distance range from the center position, and obtaining a detected temperature value generated by the optical fiber temperature measuring device; Calculating a core temperature based on the detected temperature value, and calculating a temperature ratio between the core temperature and a preset reference temperature value; If the temperature ratio is greater than a preset reference ratio, the wildfire prediction value is adjusted according to the positive correlation of the temperature ratio.

[0007] By implementing this technical solution, through data fusion from fiber-optic temperature measurement devices, multi-source data complementation, and a dynamic adjustment mechanism, the accuracy of wildfire predictions is significantly improved. Fiber-optic temperature measurement can detect temperature anomalies in blind spots, enhancing the universality of judgments through quantified temperature ratios. When the temperature ratio exceeds a threshold, the wildfire prediction value is adjusted in a positive correlation, enabling dynamic correction of risk assessments. In addition to visual data, temperature data is supplemented for dual verification, reducing false alarm rates and accurately locating risk areas.

[0008] Optionally, the method further comprises the following steps: Calculating the shortest distance between the optical fiber temperature measuring device and the center position; The reference ratio is adjusted in a positive correlation according to the shortest distance value; the smaller the shortest distance, the smaller the reference ratio; the larger the shortest distance, the larger the reference ratio; The distance range is adjusted in a positive correlation with the development level value; the smaller the development level value, the smaller the distance range; and the larger the development level value, the larger the distance range.

[0009] By adopting this technical solution, when the fiber optic temperature measurement device is close to the central location, the reference ratio is reduced, which means that the sensitivity to temperature anomalies is increased, allowing for the timely detection of potential fires in the vicinity. At longer distances, the reference ratio is increased to avoid misjudgments caused by errors in long-distance temperature measurement, making temperature verification more accurate to the actual risk. If the wildfire severity value is low and the fire spreads slowly, the distance range is narrowed to focus on the high-risk core area and reduce invalid data processing. If the severity value is high and the fire spreads rapidly, the distance range is expanded to monitor surrounding potentially affected areas in advance, buying more time for emergency decision-making.

[0010] Optionally, the step of calculating the generation speed value, the disappearance speed value, and the development degree value includes the following sub-steps: The number of new global wildfire features in the first preset time period T1 is counted as Q1, and the speed value V1 is calculated: V1=Q1 / T1; The number of global wildfire characteristics reduced in the first preset time period T1 is Q2, and the disappearance speed value V2 is calculated: V2=Q2 / T1; Development level value = V1 / V2; The step of calculating the aggregation degree value includes the following sub-steps: Taking the center of the clustered area of the global wildfire location as the center of the circle, count the number of data points Q3 within the set neighborhood radius R. If the number of data points Q3 exceeds the preset number threshold, calculate the clustering degree value: Clustering degree value = Q3 / (π×R 2 )).

[0011] By employing this technical solution, the fire generation / disappearance rate is calculated to reflect the fire's characteristic increase / decrease rate in real time. The fire development rate, through the ratio of the two, predicts the direction of fire evolution and can sensitively capture fire transitions, such as the transition from isolated fire points to widespread spread. A density formula is used to calculate the concentration rate, converting the spatial clustering of fire locations into point density per unit area, directly reflecting the concentration of risk.

[0012] Optionally, the method further includes the following sub-steps: adjusting the duration of the first time period in anti-correlation according to the aggregation degree value, wherein the smaller the aggregation degree value is, the longer the duration of the first time period is; and the larger the aggregation degree value is, the shorter the duration of the first time period is; The reference prediction value is adjusted inversely according to the development degree value; the smaller the development degree value is, the larger the reference prediction value is; and the larger the development degree value is, the smaller the reference prediction value is.

[0013] By implementing this technical solution, when the concentration value is low, the first time period, T1, is extended. This longer time window smoothes data fluctuations and avoids occasional isolated fires triggering false alarms. When the concentration value is high, T1 is shortened, improving the response speed to rapidly developing fires. When the development value is low (fire intensity is weakening), the reference prediction value is increased, triggering an alert only when the risk increases significantly, avoiding excessive response. When the development value is high (fire intensity is increasing), the reference prediction value is reduced, making it more sensitive to potential risks.

[0014] Optionally, the method further comprises the following steps: The optical fiber temperature measuring device is arranged by an unmanned device that drives the mobile camera to move. The optical fiber temperature measuring device includes an optical fiber line and an optical fiber sensor provided at the end of the optical fiber line. The optical fiber sensor is used to generate the detected temperature value and transfer the optical fiber data on the optical fiber line. The optical fiber sensor has a wireless communication module and an optical fiber communication module. The unmanned equipment has an optical fiber communication module and a wireless communication module connected to the optical fiber sensor; In an initial state, the unmanned device establishes an optical fiber communication connection with the optical fiber sensor; The unmanned device obtains a first delivery instruction from the background through a wireless communication module, and forwards the first delivery instruction to the optical fiber sensor; The optical fiber sensor establishes a wireless connection with the backend in response to the first placement instruction, the unmanned device places one optical fiber sensor and establishes an optical fiber communication connection with another optical fiber sensor that has not been placed; The unmanned device obtains a second delivery instruction from the background and delivers another optical fiber sensor.

[0015] By adopting the above technical solution, unmanned equipment deploys fiber optic sensors in stages according to instructions, which can quickly build distributed temperature measurement points in high-risk areas, improve laying efficiency, and adapt to the needs of rapid spread of wildfires; the fiber optic sensor body has both optical fiber and wireless communication modules. Under normal circumstances, it directly connects to the drone through optical fiber to ensure low-latency data transmission. After deployment, it switches to wireless communication to maintain connection with the background. The dual-link design avoids communication interruption.

[0016] Optionally, the step of the unmanned device acquiring the second delivery instruction from the background through the wireless communication module further includes the following sub-steps: If the unmanned device is connected to the backend via a wireless communication module, the signal strength of the wireless connection is detected. If the signal strength is greater than or equal to a preset reference strength value, the unmanned device obtains the second delivery instruction via the wireless communication module; If the unmanned device is not connected to the background through the wireless communication module or the signal strength is less than the preset reference strength value, the unmanned device establishes a connection with the background through the wireless temperature measurement module, and the background sends the second delivery instruction to the wirelessly connected and deployed optical fiber sensor body, and the optical fiber sensor body forwards the second delivery instruction to the optical fiber sensor body to be deployed through the optical fiber line body, and the optical fiber sensor body to be deployed forwards the second delivery instruction to the unmanned device through the optical fiber communication module.

[0017] By adopting the above technical solution and building a multi-link redundancy mechanism of wireless communication and fiber optic relay, the reliability of command transmission in complex environments is significantly improved. When the wireless signal strength between the unmanned equipment and the background meets the standard, the second delivery command is directly received wirelessly to ensure a low-latency response. If the wireless connection is interrupted or the signal is weak, it automatically switches to the fiber optic relay link and uses the deployed fiber optic sensor as a signal transfer station, breaking the wireless communication distance and terrain limitations and improving the success rate of command transmission.

[0018] Optionally, the step of placing another optical fiber sensor further includes the following sub-steps: If the unmanned device is not connected to the background through the wireless communication module or the signal strength is less than the preset reference strength value, the unmanned device establishes a wireless connection with the optical fiber sensor through the wireless communication module, disconnects the optical fiber communication module, and then deploys another optical fiber sensor.

[0019] By adopting the above technical solution, the risk of optical fiber line entanglement is avoided and the success rate of deployment is improved.

[0020] In a second aspect, the present application provides a transmission line wildfire prediction device based on multimodal data fusion, which adopts the following technical solutions: A multimodal data fusion transmission line wildfire prediction device includes a processor, wherein the processor executes the steps of any one of the multimodal data fusion transmission line wildfire prediction methods described above.

[0021] In summary, this application includes at least one of the following beneficial technical effects: By stitching multi-angle images from fixed cameras, a high-definition monitoring image covering the entire transmission line area is constructed. Combined with a deep learning algorithm to accurately identify global wildfire characteristics, this technology can capture the subtle sparking phenomenon in the early stages of a transmission line fire. Furthermore, temperature data from a fiber-optic temperature measurement device is introduced to achieve multimodal data fusion of visual and temperature perception, addressing the blind spots of single-view monitoring and significantly improving wildfire prediction accuracy. Furthermore, quantitative calculations of wildfire characteristics, such as dynamic analysis of generation and disappearance rates and development levels, enable real-time tracking of fire evolution trends.

[0022] A dual verification system, built from global early warning to local precision judgment, first calculates a fire prediction value by identifying global wildfire features using a fixed full-frame image. When the predicted value exceeds a threshold, the mobile camera is activated to focus on high-risk areas for local feature recognition. Using the constraint that the movement vector of the clustered area and the movement vector of the mobile camera are in opposite directions, combined with the overlap judgment of the local clustered area with the center position, a secondary confirmation of the fire is achieved, significantly reducing the false alarm rate caused by environmental interference or feature misjudgment. Furthermore, during the temperature data verification phase, the authenticity of the fire is further verified by comparing the temperature ratio of the fiber optic temperature measurement device with the reference ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a step diagram of a multimodal data fusion method for predicting wildfires on power transmission lines.

[0024] Figure 2 This is a graph of the substeps before comparing wildfire predictions. DETAILED DESCRIPTION

[0025] Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings.

[0026] Throughout this specification, reference to the terms "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0027] The present application embodiment discloses a method for predicting wildfires on power transmission lines by fusion of multimodal data. Figure 1 , including the following steps: Multiple fixed cameras deployed at fixed locations, such as towers and mountaintops, collect multi-angle image data in real time. Using an image stitching algorithm, feature points (such as corners and edges) are extracted from the images, matching the features of overlapping areas. Then, using a transformation matrix to align the images, the scattered fixed images are seamlessly merged into a complete, fixed, full-frame image. To ensure image stitching quality, image parameters such as resolution and color depth are uniformly calibrated during the acquisition process.

[0028] Using computer vision and deep learning techniques, global wildfire characteristics (such as those caused by power transmission lines, which initially ignite due to sporadic electrical sparks that flicker and flicker) are identified from fixed full-frame images. These global fire characteristics and their corresponding global fire locations are output, and a clustering value is calculated for each location. This clustering value directly reflects the spatial concentration trend of wildfire risk. The generation and disappearance rates of these global fire characteristics are calculated within a predefined first time period (e.g., the past 30 minutes). Using time series analysis, the number of occurrences and disappearances of these global fire characteristics is counted. The development level is calculated using the ratio of the generation and disappearance rates to reveal the dynamic evolution of the fire. A weighted average of the clustering and development levels is used to calculate the fire prediction value: α × clustering value + β × development value, where α and β are weight coefficients optimized based on historical data.

[0029] If the wildfire prediction value is greater than the preset reference prediction value, the centroid algorithm is used (by taking a weighted average of the coordinates of all wildfire location points to quickly locate the core risk area) to calculate the center position of the global wildfire location aggregation area, and a mobile camera (such as a drone equipped with a high-definition camera) is called to move to the center position.

[0030] During the flight, the mobile camera continuously captures moving area images at multiple different locations. The locations of these moving area images are located within the area corresponding to the fixed full-frame image, focusing on the high-risk areas in the fixed full-frame image to capture local wildfire details with higher resolution.

[0031] Based on the moving area image, local wildfire characteristics, such as flame color changes and the type of burning material, are further identified and the corresponding local wildfire location is determined. By analyzing the local wildfire locations in multiple moving area images, local clusters of local wildfire characteristics are calculated. Based on the position changes of these local clusters, the cluster movement vector is derived. The direction of the cluster movement vector is opposite to the movement vector of the mobile camera. Only when the local cluster coincides with the previously calculated center position and the center position lies on the cluster movement vector, is the wildfire threat finally confirmed and an early warning issued.

[0032] Image stitching technology from a cluster of fixed cameras creates a wide-area monitoring field of view, combined with quantitative analysis of local wildfire characteristics to provide an initial risk assessment. When the risk value exceeds a threshold, mobile cameras are automatically triggered to conduct targeted monitoring of the core area, acquiring high-precision data through local feature recognition and dynamic analysis of clustered areas. A superimposed location overlap verification mechanism reduces false alarm rates. The collaborative working mode of fixed equipment and mobile terminals enables dynamic scheduling of monitoring resources, saving operation and maintenance costs.

[0033] Reference Figure 2 In order to reduce the limitations of image recognition, such as poor detection at night and smoke obstruction, the following sub-steps are included before comparing the wildfire prediction values: Searches for fiber-optic temperature measuring devices within a preset distance from the center location and obtains the detected temperature values generated by the fiber-optic temperature measuring devices. Fiber-optic temperature measuring devices can sense ambient temperature changes in real time and are particularly suitable for scenarios where image recognition is difficult to use. At night or in smoky environments, when fixed and mobile cameras cannot clearly capture the flames on power lines, fiber-optic temperature measuring devices can accurately detect temperature anomalies around the lines through highly sensitive fiber-optic sensors. For example, YOFC's DTS-MMF optical fiber optimizes attenuation and bandwidth at 1550nm, enabling simultaneous temperature measurement and high-speed communication.

[0034] The center temperature is calculated based on the detected temperature value, and the temperature ratio between the center temperature and the preset reference temperature value is calculated. By converting the detected temperature value into a quantifiable temperature ratio, the wildfire risk assessment becomes more scientific and universal. First, the center temperature of the area covered by the optical fiber temperature measuring device is calculated, and then it is compared with the preset reference temperature value (ambient temperature) to form a temperature ratio. Compared with a single temperature threshold judgment, this ratio calculation method can flexibly adapt to the temperature baseline differences in different seasons and regions. For example, in a high temperature environment in summer, even if the ambient base temperature is high, the abnormal temperature increase can still be accurately identified through ratio calculation; in spring and autumn, this mechanism can also avoid misjudgments caused by ambient temperature fluctuations, so that temperature analysis can maintain a high degree of accuracy in various scenarios.

[0035] If the temperature ratio is greater than a preset reference ratio, the fire prediction value is adjusted based on the positive correlation with the temperature ratio. If the temperature ratio is much higher than the reference value, indicating a significant high temperature anomaly within the monitoring area, the system will significantly increase the fire prediction value, accelerating the subsequent mobile camera monitoring and warning process, and shortening response time.

[0036] The method further comprises the steps of: Calculate the shortest distance between the optical fiber temperature measuring device and the center position.

[0037] The reference ratio is adjusted in direct correlation with the minimum distance. The shorter the minimum distance, the smaller the reference ratio; the longer the minimum distance, the larger the reference ratio. When the minimum distance between the fiber optic temperature measuring device and the center of the wildfire concentration area is short, such as less than 50 meters, the system automatically reduces the reference ratio (using a table lookup), for example, from 1.5 to 1.2, effectively lowering the trigger threshold for temperature anomalies. For example, if the reference temperature is 50°C, a reference ratio of 1.2 corresponds to a trigger temperature of 60°C. At this point, the temperature measuring device detects 58°C, triggering the temperature ratio adjustment and preemptively detecting potential fires in the vicinity. Close-range temperature measurement improves monitoring reliability. The closer the distance, the smaller the temperature conduction loss. Therefore, by lowering the threshold, early warning sensitivity can be increased, facilitating detection of fires in their infancy.

[0038] The distance range is adjusted in a positive correlation with the development level value; smaller development levels reduce the distance range, while larger development levels increase the distance range. When the temperature measuring device is farther from the central location, for example, exceeding 200 meters, the system positively increases the reference ratio, for example, from 1.5 to 1.8, raising the threshold for detecting temperature anomalies. For example, long-distance temperature measurement can be affected by factors such as ambient airflow and obstruction by vegetation, and the error can increase to ±3°C. If the reference temperature is 50°C, a reference ratio of 1.8 corresponds to a trigger temperature of 90°C, filtering out false alarms caused by long-distance temperature measurement errors, such as localized high temperatures caused by direct sunlight. By using logic that increases the threshold with increasing distance, the temperature verification standard is dynamically aligned with the actual temperature measurement accuracy, reducing the rate of false positives at long distances.

[0039] The calculation steps of the generation speed value, disappearance speed value and development degree value include the following sub-steps: Within a preset first time period T1 (e.g., 10 minutes), the number of new fire signatures (Q1) and the number of new fire signatures (Q2) are counted, and the generation rate (V1) and the disappearance rate (V2) are calculated, respectively. For example, if five new bright sparks appear within 10 minutes, V1 = 0.5 / minute, directly reflecting the sudden intensity of the fire. If the disappearance rate (V2) = 0.3 / minute, the fire intensity has increased by 0.2 / minute during that time period.

[0040] The development level value is calculated by V1 / V2, eliminating unit differences and making trend analysis in different scenarios comparable. For example: When V1 / V2=3, it indicates that the generation speed is three times the disappearance speed, and the fire is in a rapid spreading stage; When V1 / V2=0.8, the fire tends to weaken and the disappearance speed exceeds the generation speed.

[0041] The calculation steps of the aggregation degree value include the following sub-steps: With the center of the global wildfire location cluster area as the center of the circle, count the number of global wildfire features Q3 within the neighborhood radius R (e.g., 50 meters) of the circle center. When Q3 exceeds the preset threshold (e.g., 3), the aggregation degree value is calculated by the formula = Q3 / (π×R 2 Calculate the density of fire points per unit area, such as "number / square meter". For example, when R=50 meters and Q3=10, the cluster density is 10 / (3.14×50 2 )≈0.00127 pieces / m 2 , with higher values indicating more concentrated risk. A preset threshold (e.g., Q3 ≥ 3) is used to filter out isolated points to prevent occasional misjudgments (such as single thermal imaging errors) from interfering with the results. For example, if a point has only one data point in its neighborhood, even if the density formula result is non-zero, it will be excluded as it does not meet the threshold, reducing the spatial misjudgment rate to below 5%. Density is only calculated when a sufficient number of global wildfire features are present in the neighborhood, ensuring that the concentration value reflects the true risk.

[0042] The method further comprises the following sub-steps: The duration of the first time period is adjusted inversely based on the concentration value. The lower the concentration value, the longer the first time period; the higher the concentration value, the shorter the first time period. When the concentration value is low, the sparks disperse, and the first time period T1 is automatically extended. Using a table lookup, for example, from 10 minutes to 30 minutes, data fluctuations are smoothed over a longer time window. For example, if one or two fire points appear occasionally, using a short T1 period might misidentify them as fire initiation. However, if the number of fire points does not continue to increase after extending to 30 minutes, it is considered an isolated incident, thus avoiding false alarms.

[0043] When the concentration level is high, such as when five or more sparks appear within a 5-meter radius, the system automatically shortens T1 to 5 minutes, or even 2 minutes, using a table lookup, improving its response to rapidly developing fires. For example, a high density of fire points often indicates an imminent fire outbreak, and shortening the time window allows for real-time capture of new fire points. If two new sparks appear within two minutes, the generation rate V1 immediately reaches 60 per hour, triggering an emergency alert.

[0044] The reference prediction value is adjusted inversely based on the fire severity value; the smaller the fire severity value, the larger the reference prediction value; the larger the fire severity value, the smaller the reference prediction value. When the fire severity value (V1 / V2) is less than 1.0, indicating that the fire is weakening or stabilizing, the system automatically increases the reference prediction value, for example, from 70 to 85, raising the threshold for triggering an early warning.

[0045] When the fire spread ratio (V1 / V2) exceeds 1.5, the system automatically lowers the reference prediction value to 55, becoming more sensitive to potential risks. For example, in windy weather, the fire spread ratio could rise sharply from 1.2 to 2.0. Lowering the warning threshold would allow the system to issue an alert at a fire prediction ratio of 55.

[0046] When sparks are scattered, the statistical period T1 is extended to filter out interference from occasional fire points and reduce false alarms. When fire points are concentrated, T1 is shortened to capture sudden changes in fire intensity and improve response time. At the same time, the warning threshold is dynamically adjusted based on the intensity of the fire: when the fire weakens, the reference prediction value is increased to avoid frequent warnings; when the fire intensifies, the reference prediction value is lowered to achieve dynamic adjustment.

[0047] In order to further improve the timeliness of wildfire monitoring, the unmanned equipment works in conjunction with the optical fiber temperature measurement device. The method also includes the following steps: Unmanned devices (such as drones) carry multiple fiber optic sensors and are deployed in phases according to backend commands, rapidly establishing distributed temperature measurement points in high-risk areas. For example, upon receiving the first deployment command, the drone deploys the first fiber optic sensor 50 meters from the center of the fire cluster, forming a core temperature measurement node. When the fire severity exceeds 1.5, it receives a second deployment command and deploys a second fiber optic sensor 200 meters downwind, establishing a fiber-optic temperature measurement module.

[0048] After the first fiber optic sensor is deployed, it connects to the backend data center via wireless or wired communication. A second fiber optic sensor on the unmanned device is directly connected to the device via a fiber optic communication module, enabling low-latency transmission of control data from the unmanned device. The fiber optic cable can withstand the traction of the drone during flight and is resistant to electromagnetic interference. It can maintain control data transmission even near ultra-high voltage transmission lines, significantly improving reliability compared to traditional wireless communications.

[0049] The first fiber optic sensor deployed can serve as a relay node for communication between the background and unmanned equipment, forwarding subsequent deployment instructions and data.

[0050] When the drone enters a wireless signal blind spot, such as the bottom of a valley, the background can forward the second deployment instruction to the fiber optic sensor to be deployed through the fiber optic line through the first deployed fiber optic sensor, forming a relay link and expanding the communication coverage from 3 kilometers of traditional wireless communication to more than 10 kilometers.

[0051] When the second fiber optic sensor is deployed, the second fiber optic sensor and the unmanned equipment automatically switch to the wireless communication module connection, and the unmanned equipment is indirectly connected to the background; so that the fiber optic sensor can continue to transmit data after it leaves the drone.

[0052] Multiple fiber optic sensors are connected in series through optical fiber lines to build a chain temperature measurement network. For example, four or more fiber optic sensors can be deployed on both sides of a transmission line corridor of 4 kilometers or longer to form a linear temperature monitoring belt to capture temperature changes along the line in real time.

[0053] The optical fiber sensor runs the optical fiber and wireless communication modules at the same time to form a hot backup link, that is, the wireless communication transmission link and the optical fiber relay transmission link serve as hot backups for each other.

[0054] Under normal circumstances, optical fiber communication is used first. When a break in the optical fiber line is detected, it automatically switches to wireless communication to ensure that data is not lost.

[0055] The wireless communication module has built-in Beidou positioning function, which automatically sends back location information after deployment, and the background can calibrate the temperature measurement point position in real time.

[0056] Unmanned equipment can recycle the fiber optic sensor after the mission is completed: The fiber optic sensor is controlled by wireless commands to release the anchoring device, which is then grabbed by the drone and brought back to the base.

[0057] Through the deep collaboration between unmanned equipment and fiber-optic sensors, the traditional static temperature measurement network is transformed into a dynamically reconfigurable intelligent monitoring system. Its phased deployment mechanism enables precise layout of temperature measurement points, and the dual communication link design ensures reliable data transmission in complex environments.

[0058] The step of the unmanned device obtaining the second delivery instruction from the background through the wireless communication module further includes the following sub-steps: Once the drone establishes a connection with the backend via the wireless communication module, the system monitors the wireless signal strength in real time. If the signal strength reaches or exceeds a preset reference threshold, the drone receives the second delivery command directly through the wireless communication module, ensuring a low-latency response under ideal communication conditions. If the wireless connection is interrupted or the signal strength is insufficient, the system automatically activates the fiber-optic relay transmission link. At this point, the unmanned device reconnects to the backend via the wireless temperature measurement module. The backend then sends a second deployment command to the deployed fiber-optic sensor that is already wirelessly connected. The deployed fiber-optic sensor relays the command to the next fiber-optic sensor via the fiber-optic line. The next fiber-optic sensor then passes the command to the unmanned device via the fiber-optic communication module. The multi-link redundancy mechanism dynamically switches transmission paths, effectively overcoming the impact of complex environments on command transmission. Wireless direct connection mode ensures efficient communication in common scenarios, while fiber-optic relay links overcome the limitations of wireless signal coverage and terrain conditions, improving the stability and success rate of command transmission.

[0059] The step of deploying another optical fiber sensor also includes the following sub-steps: If the unmanned device is not connected to the backend via the wireless communication module or the signal strength is below a preset reference strength, the device will establish a wireless connection with the fiber optic sensor through the wireless communication module, then disconnect the module and deploy another fiber optic sensor. This switch to wireless communication effectively avoids interruptions in unmanned device control caused by a sudden disconnection of another fiber optic sensor during deployment. By dynamically adjusting the communication method, the device ensures safe operation in complex environments.

[0060] An embodiment of the present application further discloses a transmission line wildfire prediction device using multimodal data fusion, comprising a processor that executes the steps of the transmission line wildfire prediction method using multimodal data fusion as described above.

[0061] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for predicting wildfires on power transmission lines based on multimodal data fusion, characterized in that: The steps include: Acquire multiple fixed images based on multiple fixed cameras, where the fixed cameras are located at fixed positions in different locations; stitch and fuse the multiple fixed images into a fixed full-frame image; Identifying global wildfire features from the fixed full-frame image, outputting the identified global wildfire features and corresponding global wildfire locations, and calculating a concentration level value for the global wildfire locations; calculating a generation speed value and a disappearance speed value of the global wildfire features within a preset first time period, and calculating a development level value based on the generation speed value and the disappearance speed value; Calculating a wildfire prediction value based on the concentration degree value and the development degree value; If the wildfire prediction value is greater than a preset reference prediction value, the center position of the global wildfire location gathering area is calculated, and a mobile camera is called to move to the center position; The mobile camera captures a plurality of moving area images at different positions when moving, and the positions of the moving area images are located in the area corresponding to the fixed full-frame image; Identifying local wildfire features based on the moving area image, outputting the identified local wildfire features and corresponding local wildfire locations, and calculating a localized concentration area of the local wildfire features based on the local wildfire locations; Calculating a plurality of local gathering areas based on the plurality of moving area images, and calculating a gathering area movement vector according to positions of the plurality of local gathering areas; wherein the direction of the gathering area movement vector is opposite to the direction of the movement vector of the moving camera; If the local gathering area corresponds to the center position and the center position is on the moving vector of the gathering area, a wildfire warning is issued.

2. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 1, characterized in that: Before comparing the wildfire prediction values, the following sub-steps are also included: Searching for an optical fiber temperature measuring device within a preset distance range from the center position, and obtaining a detected temperature value generated by the optical fiber temperature measuring device; Calculating a core temperature based on the detected temperature value, and calculating a temperature ratio between the core temperature and a preset reference temperature value; If the temperature ratio is greater than a preset reference ratio, the wildfire prediction value is adjusted according to the positive correlation of the temperature ratio.

3. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 2, characterized in that: The method further comprises the steps of: Calculating the shortest distance between the optical fiber temperature measuring device and the center position; The reference ratio is adjusted in a positive correlation according to the shortest distance value; the smaller the shortest distance, the smaller the reference ratio; The longer the shortest distance is, the larger the reference ratio is; The distance range is adjusted in a positive correlation with the development level value; the smaller the development level value, the smaller the distance range; and the larger the development level value, the larger the distance range.

4. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 1, wherein: The step of calculating the generation speed value, the disappearance speed value, and the development degree value includes the following sub-steps: The number of new global wildfire features in the first preset time period T1 is counted as Q1, and the speed value V1 is calculated: V1=Q1 / T1; The number of global wildfire characteristics reduced in the first preset time period T1 is Q2, and the disappearance speed value V2 is calculated: V2=Q2 / T1; Development level value = V1 / V2; The step of calculating the aggregation degree value includes the following sub-steps: Taking the center of the clustered area of the global wildfire location as the center of the circle, count the number of data points Q3 within the set neighborhood radius R. If the number of data points Q3 exceeds the preset number threshold, calculate the clustering degree value: Clustering degree value = Q3 / (π×R 2 )).

5. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 1, characterized in that: The method further comprises the following sub-steps: adjusting the duration of the first time period in an anti-correlated manner according to the aggregation degree value, wherein the smaller the aggregation degree value is, the longer the duration of the first time period is; The greater the aggregation degree value, the shorter the first time period; The reference prediction value is adjusted inversely according to the development degree value; the smaller the development degree value is, the larger the reference prediction value is; and the larger the development degree value is, the smaller the reference prediction value is.

6. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 2, characterized in that: The method further comprises the steps of: The optical fiber temperature measuring device is arranged by an unmanned device that drives the mobile camera to move. The optical fiber temperature measuring device includes an optical fiber line and an optical fiber sensor provided at the end of the optical fiber line. The optical fiber sensor is used to generate the detected temperature value and transfer the optical fiber data on the optical fiber line. The optical fiber sensor has a wireless communication module and an optical fiber communication module. The unmanned equipment has an optical fiber communication module and a wireless communication module connected to the optical fiber sensor; In an initial state, the unmanned device establishes an optical fiber communication connection with the optical fiber sensor; The unmanned device obtains a first delivery instruction from the background through a wireless communication module, and forwards the first delivery instruction to the optical fiber sensor; The optical fiber sensor establishes a wireless connection with the backend in response to the first placement instruction, the unmanned device places one optical fiber sensor and establishes an optical fiber communication connection with another optical fiber sensor that has not been placed; The unmanned device obtains a second delivery instruction from the background and delivers another optical fiber sensor.

7. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 6, characterized in that: The step of the unmanned device acquiring the second delivery instruction from the background through the wireless communication module further includes the following sub-steps: If the unmanned device is connected to the backend via a wireless communication module, the signal strength of the wireless connection is detected. If the signal strength is greater than or equal to a preset reference strength value, the unmanned device obtains the second delivery instruction via the wireless communication module; If the unmanned device is not connected to the background through the wireless communication module or the signal strength is less than the preset reference strength value, the unmanned device establishes a connection with the background through the wireless temperature measurement module, and the background sends the second delivery instruction to the wirelessly connected and deployed optical fiber sensor body, and the optical fiber sensor body forwards the second delivery instruction to the optical fiber sensor body to be deployed through the optical fiber line body, and the optical fiber sensor body to be deployed forwards the second delivery instruction to the unmanned device through the optical fiber communication module.

8. The method for predicting wildfires on power transmission lines using multimodal data fusion according to claim 7, characterized in that: The step of placing another optical fiber sensor further includes the following sub-steps: If the unmanned device is not connected to the background through the wireless communication module or the signal strength is less than the preset reference strength value, the unmanned device establishes a wireless connection with the optical fiber sensor through the wireless communication module, disconnects the optical fiber communication module, and then deploys another optical fiber sensor.

9. A multimodal data fusion transmission line wildfire prediction device, characterized in that: The method comprises a processor, wherein the processor executes the steps of the transmission line wildfire prediction method based on multimodal data fusion as described in any one of claims 1 to 8.

Images