A power transmission line fire alarm method and system based on multi-source data fusion

Abstract

The application discloses a power transmission line fire alarm method and system based on multi-source data fusion, the method comprising: monitoring fire points according to ground monitoring device monitoring data, unmanned aerial vehicle monitoring data and satellite monitoring data of a target area where the power transmission line is located; when it is judged that there is a fire point according to the satellite monitoring data and multiple satellites monitor the fire point, fusing the fire point information and monitoring data of the multiple satellites, generating time series data corresponding to the satellites, then acquiring geographical data and meteorological data of the target area, and respectively extracting geographical features corresponding to the geographical data, time series data and time series features corresponding to the meteorological data, splicing the geographical features and the time series features, obtaining a feature vector, identifying the fire point through a preset classification algorithm, judging whether the fire point monitored by the satellite is a real fire point, and if so, alarming, thereby improving the accuracy of fire point alarm.

Description

Technical Field

[0001] This invention relates to the field of power engineering technology, and in particular to a method and system for fire alarm of transmission lines based on multi-source data fusion. Background Technology

[0002] Wildfires are a common natural disaster. When wildfires occur, the high electrical conductivity of the fire, as well as the high temperature and smoke particles produced by the fire, can reduce the air insulation level of operating power transmission lines, causing power line faults and affecting the stable operation of the power system. Furthermore, the continuous burning of wildfires makes it impossible to repair faulty power transmission lines in a timely manner, increasing the loss costs of the power system.

[0003] In existing technologies, methods for monitoring wildfires in power grids include meteorological satellites, drones, and ground-based online monitoring devices. However, meteorological satellite monitoring suffers from low resolution and inaccurate positioning and alarms. While drone-based line inspection monitoring can quickly inspect large areas of lines, it is easily affected by weather conditions, and its monitoring time and cost are high. Although ground-based online monitoring devices can provide real-time images of the fire, they are limited by installation location and field of view, which can easily create blind spots in monitoring.

[0004] Therefore, existing monitoring methods are limited in scope, accuracy, and comprehensive analysis capabilities, and are constrained by single technologies. When using multi-source detection, the heterogeneity of multi-source monitoring data leads to inconsistencies and reduces the accuracy of the data. Furthermore, neither single-technology monitoring nor multi-source collaborative monitoring in existing technologies fully considers the terrain and meteorological characteristics of hotspot locations when issuing wildfire warnings, resulting in a large number of missed and false alarms. This not only increases the cost of warnings but also reduces the accuracy of wildfire warnings. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention discloses a method and system for alarming fires on power transmission lines based on multi-source data fusion, which improves the accuracy of wildfire monitoring and issues timely alarms to reduce power grid losses.

[0006] To achieve the above objectives, this invention discloses a method for fire alarm of transmission lines based on multi-source data fusion, comprising:

[0007] Based on the monitoring data from the ground monitoring device and the monitoring data from the drone, it is determined whether the preset ground monitoring device or drone in the target area where the transmission line is located has detected a fire point. If the fire point is detected, the first fire point information of the fire point is obtained, and an alarm is issued based on the first fire point information.

[0008] If the fire point is not detected, the satellite monitoring data obtained is used to determine whether the satellite corresponding to the target area has detected the fire point. If the fire point is detected, the meteorological data and geographical data of the target area are obtained, and the first satellite that currently detects the fire point is determined based on the satellite monitoring data.

[0009] When the first satellite is a single satellite, the second fire point information obtained by the single satellite from observing the fire point and the first monitoring data within the first time period before the single satellite detects the fire point are obtained. Based on the second fire point information and the first monitoring data, the first time series data corresponding to the single satellite is obtained.

[0010] When the first satellite is multiple satellites, the third fire point information obtained by each satellite from observing the fire point and the second monitoring data of each satellite in the first time period before the fire point is detected are obtained. The third fire point information and the second monitoring data of the multiple satellites are fused according to a preset weighted average processing method to obtain the fourth fire point information after the fusion of the third fire point information of the multiple satellites and the third monitoring data after the fusion of the second monitoring data. The second time series data corresponding to the multiple satellites is obtained based on the fourth fire point information and the third monitoring data.

[0011] Feature extraction is performed on the geographic data, meteorological data, first time series data and second time series data to obtain the first feature corresponding to the geographic data, the first time series feature corresponding to the meteorological data and the second time series feature corresponding to the first time series data or the second time series data;

[0012] Convolution, vectorization, and concatenation operations are performed on the first feature, the first temporal feature, and the second temporal feature respectively to obtain the feature vector corresponding to the concatenation of the first feature, the first temporal feature, and the second temporal feature;

[0013] The feature vector is identified according to a preset classification algorithm to determine whether the fire point is a real fire point. If the fire point is a real fire point, an alarm is issued based on the second fire point information or the fourth fire point information.

[0014] This invention discloses a method for fire alarm of transmission lines based on multi-source data fusion. First, multi-source monitoring data of the target area where the transmission line is located is collected. This multi-source monitoring data is then fused from multiple angles to improve monitoring accuracy. Next, preliminary monitoring is conducted using ground monitoring devices and drones to determine the presence of fire points. This combines close-range monitoring by ground devices with wide-area patrols by drones to ensure comprehensive monitoring of thermal power plants, thus improving monitoring accuracy and reducing drone patrol costs. When neither the ground monitoring devices nor the drones detect a fire point, to avoid missed fire detection due to blind spots of the ground monitoring devices and the patrol intervals of the drones, monitoring data from satellites located in the target area is acquired. The method first determines whether a fire exists based on this monitoring data. In wildfire detection, when fire points are detected in the data, it is determined whether the fire point was detected by a single satellite or multiple satellites. If multiple satellites are involved, the data from different satellites need to be fused to ensure that the processed fire point information has higher reliability. Furthermore, to address the potential influence of various factors on satellite observations, the fire point information and monitoring data over a period of time are fused together as a time series, serving as the judgment data for subsequent fire point identification, further improving the accuracy of fire point monitoring. During fire point monitoring, meteorological data and geographical environment of the target area can also lead to errors in fire point detection. Therefore, features of real-time meteorological and ground factors at the fire point location are extracted for fire point identification, and combined features are used with the corresponding time series data for fire point recognition. This approach considers past monitoring results during the monitoring period, improving the accuracy of fire point monitoring and identification.

[0015] As a preferred example, in determining whether a fire point has been detected by a satellite corresponding to the target area based on the acquired satellite monitoring data, and if the fire point has been detected, meteorological and geographical data of the target area are acquired, and the first satellite currently detecting the fire point is determined based on the satellite monitoring data, including:

[0016] Acquire first satellite monitoring data detected by polar-orbiting satellites, and perform regional extraction on the first satellite monitoring data to obtain second satellite monitoring data corresponding to the target region;

[0017] Based on the second satellite monitoring data, it is determined whether the polar-orbiting satellite has detected a fire point in the target area. If the polar-orbiting satellite has detected a fire point, the first polar-orbiting satellite that detected the fire point is determined based on the second satellite monitoring data.

[0018] If the polar-orbiting satellite does not detect any fire points in the target area, then the third satellite monitoring data detected by the geostationary satellite is obtained, and the third satellite monitoring data is used to extract the region to obtain the fourth satellite monitoring data corresponding to the target area.

[0019] Based on the monitoring data of the fourth satellite, it is determined whether the geostationary satellite has detected a fire point in the target area. If the geostationary satellite detects a fire point, the first geostationary satellite that detected the fire point is determined based on the monitoring data of the fourth satellite. The first satellite is either the first polar-orbiting satellite or the first geostationary satellite.

[0020] This invention utilizes polar-orbiting satellites and geostationary satellites to perform comprehensive fire monitoring of the target area, thereby improving the accuracy of fire alarms. Furthermore, since the number and coverage of polar-orbiting satellites are greater than those of geostationary satellites, monitoring data from polar-orbiting satellites are acquired first for fire identification, which reduces the amount of data processing, increases the rate of fire identification, and thus improves the alarm rate.

[0021] As a preferred example, the process of fusing the third fire point information and the second monitoring data from multiple satellites according to a preset weighted average processing method to obtain the fourth fire point information after fusing the third fire point information from multiple satellites and the third monitoring data after fusing the second monitoring data includes:

[0022] The weight of each satellite is determined based on the evaluation level of its observation performance.

[0023] Based on the weights and the third fire point information and second monitoring data corresponding to each satellite, a weighted average is performed on the multiple satellites to obtain the fourth fire point information corresponding to the fusion of the third fire point information of the multiple satellites and the third monitoring data corresponding to the fusion of the second monitoring data of the multiple satellites.

[0024] This invention takes into account that when multiple satellites simultaneously observe a fire point, it is necessary to fuse the observation data from different satellites. The processing method is to perform a weighted average of the fire point information and monitoring data obtained by the multiple satellites according to the weights corresponding to different satellites, so as to ensure that the processed fire point information has higher reliability and thus improve the accuracy of fire point identification.

[0025] As a preferred example, the feature extraction from the geographic data, meteorological data, first time-series data, and second time-series data includes:

[0026] The geographic data is subjected to feature extraction according to a preset feature extraction algorithm to obtain the first feature corresponding to the geographic data;

[0027] Based on a preset convolutional neural network, time-series features are extracted in the time dimension of the meteorological data, the first time-series data, and the second time-series data by using a sliding convolution kernel, so as to obtain the first time-series feature corresponding to the meteorological data and the second time-series feature corresponding to the first time-series data or the second time-series data.

[0028] This invention considers the influence of real-time meteorological and ground factors on fire point identification. Therefore, it extracts the features of the real-time meteorological and ground factors to assist in subsequent fire point identification. Then, it uses time-series data to extract the corresponding time-series features, taking into account past monitoring results during the monitoring period, thereby improving the accuracy of final fire point monitoring and identification.

[0029] As a preferred example, the convolution operation, vectorization operation, and concatenation operation performed on the first feature, the first temporal feature, and the second temporal feature respectively include:

[0030] Convolution operations are performed on the first feature, the first temporal feature, and the second temporal feature respectively. The convolution kernel slides on the feature matrices corresponding to the first feature, the first temporal feature, and the second temporal feature respectively to obtain the first feature map corresponding to the first feature, the first temporal feature map corresponding to the first temporal feature, and the second temporal feature map corresponding to the second temporal feature.

[0031] All elements in the first feature map, the first time-series feature map, and the second time-series feature map are connected by rows or columns respectively, and flattened to obtain the first vector corresponding to the first feature map, the first time-series vector corresponding to the first time-series feature map, and the second time-series vector corresponding to the second time-series feature map;

[0032] By concatenating the first vector, the first time-series vector, and the second time-series vector, the feature vector corresponding to the fire point is obtained.

[0033] This invention performs convolution, vectorization, and splicing operations on the features corresponding to the geographical factors, meteorological data, and time-series data, respectively. This results in a feature vector that not only considers real-time meteorological and geographical factors of the fire location for fire point identification, but also utilizes time-series data to take into account past monitoring results during the monitoring period, thereby improving the accuracy of subsequent fire point monitoring and identification.

[0034] On the other hand, the present invention also discloses a transmission line fire alarm system based on multi-source data fusion. The system includes an initial monitoring module, a satellite monitoring module, a single processing module, a multi-source processing module, a feature extraction module, a feature splicing module, and a fire alarm module.

[0035] The initial monitoring module is used to determine whether the preset ground monitoring device or drone in the target area where the transmission line is located has detected a fire point based on the monitoring data of the ground monitoring device and the monitoring data of the drone. If the fire point is detected, the first fire point information of the fire point is obtained and an alarm is issued based on the first fire point information.

[0036] The satellite monitoring module is used to determine whether the satellite corresponding to the target area has detected the fire point if the fire point is not detected, based on the acquired satellite monitoring data. If the fire point is detected, the module acquires the meteorological and geographical data of the target area and determines the first satellite that currently detects the fire point based on the satellite monitoring data.

[0037] The single processing module is used to obtain, when the first satellite is a single satellite, the second fire point information obtained by the single satellite from observing the fire point and the first monitoring data within a first time period before the single satellite detects the fire point, and obtain the first time series data corresponding to the single satellite based on the second fire point information and the first monitoring data.

[0038] The multi-source processing module is used to, when the first satellite is multiple satellites, acquire the third fire point information obtained by each satellite from observing the fire point and the second monitoring data within a first time period before each satellite detects the fire point, fuse the third fire point information and the second monitoring data of the multiple satellites according to a preset weighted average processing method, obtain the fourth fire point information after the fusion of the third fire point information of the multiple satellites and the third monitoring data after the fusion of the second monitoring data, and obtain the second time series data corresponding to the multiple satellites according to the fourth fire point information and the third monitoring data.

[0039] The feature extraction module is used to extract features from the geographic data, meteorological data, first time series data and second time series data to obtain the first feature corresponding to the geographic data, the first time series feature corresponding to the meteorological data and the second time series feature corresponding to the first time series data or the second time series data.

[0040] The feature concatenation module is used to perform convolution, vectorization and concatenation operations on the first feature, the first temporal feature and the second temporal feature respectively to obtain the feature vector corresponding to the concatenation of the first feature, the first temporal feature and the second temporal feature;

[0041] The fire alarm module is used to identify the feature vector according to a preset classification algorithm, determine whether the fire point is a real fire point, and if the fire point is a real fire point, then issue an alarm based on the second fire point information or the fourth fire point information.

[0042] This invention discloses a transmission line fire alarm system based on multi-source data fusion. First, multi-source monitoring data of the target area where the transmission line is located is collected. This multi-source monitoring data is then fused from multiple angles to improve monitoring accuracy. Next, preliminary monitoring is conducted using ground monitoring devices and drones to determine the presence of fire points. This combines close-range monitoring by ground devices with wide-area patrols by drones to ensure comprehensive monitoring of thermal power plants, thus improving monitoring accuracy and reducing drone patrol costs. When neither the ground monitoring devices nor the drones detect a fire point, to avoid missed fire detection due to blind spots of the ground monitoring devices and the patrol intervals of the drones, monitoring data from satellites located in the target area is acquired. The system first determines whether a fire exists based on this monitoring data. In wildfire detection, when fire points are detected in the data, it is determined whether the fire point was detected by a single satellite or multiple satellites. If multiple satellites are involved, the data from different satellites need to be fused to ensure that the processed fire point information has higher reliability. Furthermore, to address the potential influence of various factors on satellite observations, the fire point information and monitoring data over a period of time are fused together as a time series, serving as the judgment data for subsequent fire point identification, further improving the accuracy of fire point monitoring. During fire point monitoring, meteorological data and geographical environment of the target area can also lead to errors in fire point detection. Therefore, features of real-time meteorological and ground factors at the fire point location are extracted for fire point identification, and combined features are used with the corresponding time series data for fire point recognition. This approach considers past monitoring results during the monitoring period, improving the accuracy of fire point monitoring and identification.

[0043] As a preferred example, the satellite monitoring module includes a polar-orbiting monitoring unit and a geostationary monitoring unit;

[0044] The polar-orbiting monitoring unit is used to acquire first satellite monitoring data detected by the polar-orbiting satellite, and to extract the region from the first satellite monitoring data to obtain second satellite monitoring data corresponding to the target region; based on the second satellite monitoring data, it is determined whether the polar-orbiting satellite has detected a fire point in the target region; if the polar-orbiting satellite has detected a fire point, the first polar-orbiting satellite that detected the fire point is determined based on the second satellite monitoring data.

[0045] The geostationary monitoring unit is used to acquire third satellite monitoring data detected by the geostationary satellite if the polar-orbiting satellite does not detect a fire point in the target area, and to extract the region from the third satellite monitoring data to obtain fourth satellite monitoring data corresponding to the target area; based on the fourth satellite monitoring data, it determines whether the geostationary satellite detects a fire point in the target area; if the geostationary satellite detects a fire point, it determines the first geostationary satellite that detected the fire point based on the fourth satellite monitoring data; the first satellite is the first polar-orbiting satellite or the first geostationary satellite.

[0046] This invention utilizes polar-orbiting satellites and geostationary satellites to perform comprehensive fire monitoring of the target area, thereby improving the accuracy of fire alarms. Furthermore, since the number and coverage of polar-orbiting satellites are greater than those of geostationary satellites, monitoring data from polar-orbiting satellites are acquired first for fire identification, which reduces the amount of data processing, increases the rate of fire identification, and thus improves the alarm rate.

[0047] As a preferred example, the multi-source processing module includes a weight determination unit and a data fusion unit;

[0048] The weight determination unit is used to determine the weight of each of the multiple satellites based on the evaluation level of the observation performance of each of the multiple satellites.

[0049] The data fusion unit is used to perform weighted average processing on the multiple satellites according to the weights and the third fire point information and second monitoring data corresponding to each satellite, to obtain the fourth fire point information corresponding to the fusion of the third fire point information of the multiple satellites and the third monitoring data corresponding to the fusion of the second monitoring data of the multiple satellites.

[0050] This invention takes into account that when multiple satellites simultaneously observe a fire point, it is necessary to fuse the observation data from different satellites. The processing method is to perform a weighted average of the fire point information and monitoring data obtained by the multiple satellites according to the weights corresponding to different satellites, so as to ensure that the processed fire point information has higher reliability and thus improve the accuracy of fire point identification.

[0051] As a preferred example, the feature extraction module includes geographic feature units and temporal feature units;

[0052] The geographic feature unit is used to extract features from the geographic data according to a preset feature extraction algorithm to obtain the first feature corresponding to the geographic data;

[0053] The temporal feature unit is used to extract temporal features in the time dimension of the meteorological data, the first temporal data, and the second temporal data respectively by sliding convolution kernels according to a preset convolutional neural network, so as to obtain the first temporal feature corresponding to the meteorological data and the second temporal feature corresponding to the first temporal data or the second temporal data.

[0054] This invention considers the influence of real-time meteorological and ground factors on fire point identification. Therefore, it extracts the features of the real-time meteorological and ground factors to assist in subsequent fire point identification. Then, it uses time-series data to extract the corresponding time-series features, taking into account past monitoring results during the monitoring period, thereby improving the accuracy of final fire point monitoring and identification.

[0055] As a preferred example, the feature concatenation module includes a convolutional unit, a vector unit, and a concatenation unit;

[0056] The convolutional unit is used to perform convolution operations on the first feature, the first temporal feature and the second temporal feature respectively. By sliding the convolutional kernel on the feature matrix corresponding to the first feature, the first temporal feature and the second temporal feature respectively, a first feature map corresponding to the first feature, a first temporal feature map corresponding to the first temporal feature and a second temporal feature map corresponding to the second temporal feature are obtained.

[0057] The vector unit is used to connect all elements in the first feature map, the first time-series feature map and the second time-series feature map by rows or columns, and flatten them to obtain the first vector corresponding to the first feature map, the first time-series vector corresponding to the first time-series feature map and the second time-series vector corresponding to the second time-series feature map;

[0058] The splicing unit is used to splice the first vector, the first time-series vector, and the second time-series vector to obtain the feature vector corresponding to the fire point.

[0059] This invention performs convolution, vectorization, and splicing operations on the features corresponding to the geographical factors, meteorological data, and time-series data, respectively. This results in a feature vector that not only considers real-time meteorological and geographical factors of the fire location for fire point identification, but also utilizes time-series data to take into account past monitoring results during the monitoring period, thereby improving the accuracy of subsequent fire point monitoring and identification. Attached Figure Description

[0060] Figure 1 : A flowchart illustrating a transmission line fire alarm method based on multi-source data fusion provided in an embodiment of the present invention;

[0061] Figure 2 : A schematic diagram of the structure of a transmission line fire alarm system based on multi-source data fusion provided in an embodiment of the present invention;

[0062] Figure 3 : A flowchart illustrating a transmission line fire alarm method based on multi-source data fusion, as provided in another embodiment of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Example 1

[0065] This embodiment discloses a transmission line fire alarm method based on multi-source data fusion. The specific implementation process of the alarm method can be referred to... Figure 1 It mainly includes steps 101 to 107, the steps being:

[0066] Step 101: Based on the acquired monitoring data from the ground monitoring device and the monitoring data from the drone, determine whether the preset ground monitoring device or drone in the target area where the transmission line is located has detected a fire point. If the fire point is detected, obtain the first fire point information of the fire point and issue an alarm based on the first fire point information.

[0067] Step 102: If the fire point is not detected, determine whether the satellite corresponding to the target area has detected the fire point based on the acquired satellite monitoring data. If the fire point is detected, acquire the meteorological data and geographical data of the target area, and determine the first satellite that currently detects the fire point based on the satellite monitoring data.

[0068] In this embodiment, the steps mainly include: acquiring first satellite monitoring data detected by a polar-orbiting satellite, and performing region extraction on the first satellite monitoring data to obtain second satellite monitoring data corresponding to the target area; determining whether the polar-orbiting satellite has detected a fire point in the target area based on the second satellite monitoring data; if the polar-orbiting satellite has detected a fire point, determining the first polar-orbiting satellite that detected the fire point based on the second satellite monitoring data; if the polar-orbiting satellite has not detected a fire point in the target area, acquiring third satellite monitoring data detected by a geostationary satellite, and performing region extraction on the third satellite monitoring data to obtain fourth satellite monitoring data corresponding to the target area; determining whether the geostationary satellite has detected a fire point in the target area based on the fourth satellite monitoring data; if the geostationary satellite has detected a fire point, determining the first geostationary satellite that detected the fire point based on the fourth satellite monitoring data; the first satellite is either the first polar-orbiting satellite or the first geostationary satellite.

[0069] In this embodiment, this step utilizes the polar-orbiting satellites and the geostationary satellites to conduct comprehensive fire point monitoring of the target area, thereby improving the accuracy of fire point alarms. Furthermore, since the number and coverage of polar-orbiting satellites are larger than those of geostationary satellites, the monitoring data of polar-orbiting satellites are acquired first to identify fire points, so as to reduce the amount of data processing, increase the fire point identification rate, and thus increase the alarm rate.

[0070] Step 103: When the first satellite is a single satellite, the second fire point information obtained by the single satellite from observing the fire point and the first monitoring data within the first time period before the single satellite detects the fire point are obtained. Based on the second fire point information and the first monitoring data, the first time series data corresponding to the single satellite is obtained.

[0071] Step 104: When the first satellite consists of multiple satellites, the third fire point information obtained by each satellite from observing the fire point and the second monitoring data within the first time period before each satellite detects the fire point are obtained. The third fire point information and the second monitoring data of the multiple satellites are fused according to a preset weighted average processing method to obtain the fourth fire point information after the fusion of the third fire point information of the multiple satellites and the third monitoring data after the fusion of the second monitoring data. Based on the fourth fire point information and the third monitoring data, the second time series data corresponding to the multiple satellites are obtained.

[0072] In this embodiment, the step mainly includes: determining the weight of each satellite among the multiple satellites based on the evaluation level of the observation performance of each satellite; and performing a weighted average processing on the multiple satellites based on the weights and the third fire point information and second monitoring data corresponding to each satellite to obtain the fourth fire point information corresponding to the fusion of the third fire point information of the multiple satellites and the third monitoring data corresponding to the fusion of the second monitoring data of the multiple satellites.

[0073] In this embodiment, this step takes into account that if multiple satellites simultaneously observe a fire point, it is necessary to fuse the observation data from different satellites. The processing method is to perform weighted averaging of the fire point information and monitoring data obtained by the multiple satellites according to the weights corresponding to different satellites, so as to ensure that the processed fire point information has higher credibility and thus improve the accuracy of fire point identification.

[0074] Step 105: Perform feature extraction on the geographic data, meteorological data, first time series data and second time series data to obtain the first feature corresponding to the geographic data, the first time series feature corresponding to the meteorological data and the second time series feature corresponding to the first time series data or the second time series data.

[0075] In this embodiment, the step mainly includes: extracting features from the geographic data according to a preset feature extraction algorithm to obtain a first feature corresponding to the geographic data; and extracting time-series features from the meteorological data, the first time-series data, and the second time-series data respectively by using a preset convolutional neural network and sliding convolution kernels to obtain a first time-series feature corresponding to the meteorological data and a second time-series feature corresponding to the first time-series data or the second time-series data.

[0076] In this embodiment, this step considers the influence of real-time meteorological and ground factors on the fire point identification. Therefore, the features of the real-time meteorological and ground factors are extracted to assist in the subsequent fire point identification. Then, the corresponding time-series features are extracted using time-series data, taking into account the past monitoring results of the monitoring period, which improves the accuracy of the final fire point monitoring and identification.

[0077] Step 106: Perform convolution, vectorization, and concatenation operations on the first feature, the first temporal feature, and the second temporal feature respectively to obtain the feature vector corresponding to the concatenation of the first feature, the first temporal feature, and the second temporal feature.

[0078] In this embodiment, the step mainly includes: performing convolution operations on the first feature, the first temporal feature, and the second temporal feature respectively, and sliding the convolution kernel on the feature matrices corresponding to the first feature, the first temporal feature, and the second temporal feature respectively to obtain a first feature map corresponding to the first feature, a first temporal feature map corresponding to the first temporal feature, and a second temporal feature map corresponding to the second temporal feature; concatenating all elements in the first feature map, the first temporal feature map, and the second temporal feature map by rows or columns respectively, and flattening them to obtain a first vector corresponding to the first feature map, a first temporal vector corresponding to the first temporal feature map, and a second temporal vector corresponding to the second temporal feature map; and concatenating the first vector, the first temporal vector, and the second temporal vector to obtain the feature vector corresponding to the fire point.

[0079] In this embodiment, this step performs convolution, vectorization, and splicing operations on the features corresponding to the geographical factors, meteorological data, and time series data, respectively. This results in a feature vector that not only considers the real-time meteorological and geographical factors of the fire location for fire point identification, but also utilizes time series data to consider past monitoring results during the monitoring period, thereby improving the accuracy of subsequent fire point monitoring and identification.

[0080] Step 107: Identify the feature vector according to the preset classification algorithm, and determine whether the fire point is a real fire point. If the fire point is a real fire point, issue an alarm based on the second fire point information or the fourth fire point information.

[0081] On the other hand, this embodiment also discloses a transmission line fire alarm system based on multi-source data fusion. For the specific structural composition of the system, please refer to... Figure 2 The system includes an initial monitoring module 201, a satellite monitoring module 202, a single processing module 203, a multi-source processing module 204, a feature extraction module 205, a feature splicing module 206, and a fire alarm module 207.

[0082] The initial monitoring module 201 is used to determine whether the preset ground monitoring device or drone in the target area where the transmission line is located has detected a fire point based on the monitoring data of the ground monitoring device and the monitoring data of the drone. If the fire point is detected, the module obtains the first fire point information of the fire point and issues an alarm based on the first fire point information.

[0083] The satellite monitoring module 202 is used to determine whether the satellite corresponding to the target area has detected the fire point if the fire point is not detected, based on the acquired satellite monitoring data. If the fire point is detected, the module acquires the meteorological and geographical data of the target area and determines the first satellite that currently detects the fire point based on the satellite monitoring data.

[0084] The single processing module 203 is used to obtain, when the first satellite is a single satellite, the second fire point information obtained by the single satellite from observing the fire point and the first monitoring data within a first time period before the single satellite detects the fire point, and obtain the first time series data corresponding to the single satellite based on the second fire point information and the first monitoring data.

[0085] The multi-source processing module 204 is used to, when the first satellite is multiple satellites, acquire the third fire point information obtained by each satellite from observing the fire point and the second monitoring data within a first time period before each satellite detects the fire point, fuse the third fire point information and the second monitoring data of the multiple satellites according to a preset weighted average processing method, obtain the fourth fire point information after the fusion of the third fire point information of the multiple satellites and the third monitoring data after the fusion of the second monitoring data, and obtain the second time series data corresponding to the multiple satellites according to the fourth fire point information and the third monitoring data.

[0086] The feature extraction module 205 is used to extract features from the geographic data, meteorological data, first time series data and second time series data to obtain a first feature corresponding to the geographic data, a first time series feature corresponding to the meteorological data and a second time series feature corresponding to the first time series data or the second time series data.

[0087] The feature concatenation module 206 is used to perform convolution, vectorization and concatenation operations on the first feature, the first temporal feature and the second temporal feature respectively to obtain the feature vector corresponding to the concatenation of the first feature, the first temporal feature and the second temporal feature.

[0088] The fire alarm module 207 is used to identify the feature vector according to a preset classification algorithm, determine whether the fire point is a real fire point, and if the fire point is a real fire point, then issue an alarm according to the second fire point information or the fourth fire point information.

[0089] In this embodiment, the satellite monitoring module 202 includes a polar orbit monitoring unit and a geostationary monitoring unit.

[0090] The polar-orbiting monitoring unit is used to acquire first satellite monitoring data detected by the polar-orbiting satellite, and to extract the region from the first satellite monitoring data to obtain second satellite monitoring data corresponding to the target region; based on the second satellite monitoring data, it is determined whether the polar-orbiting satellite has detected a fire point in the target region; if the polar-orbiting satellite has detected a fire point, the first polar-orbiting satellite that detected the fire point is determined based on the second satellite monitoring data.

[0091] The geostationary monitoring unit is used to acquire third satellite monitoring data detected by the geostationary satellite if the polar-orbiting satellite does not detect a fire point in the target area, and to extract the region from the third satellite monitoring data to obtain fourth satellite monitoring data corresponding to the target area; based on the fourth satellite monitoring data, it determines whether the geostationary satellite detects a fire point in the target area; if the geostationary satellite detects a fire point, it determines the first geostationary satellite that detected the fire point based on the fourth satellite monitoring data; the first satellite is the first polar-orbiting satellite or the first geostationary satellite.

[0092] In this embodiment, the multi-source processing module 204 includes a weight determination unit and a data fusion unit.

[0093] The weight determination unit is used to determine the weight of each of the multiple satellites based on the evaluation level of the observation performance of each of the multiple satellites.

[0094] The data fusion unit is used to perform weighted average processing on the multiple satellites according to the weights and the third fire point information and second monitoring data corresponding to each satellite, to obtain the fourth fire point information corresponding to the fusion of the third fire point information of the multiple satellites and the third monitoring data corresponding to the fusion of the second monitoring data of the multiple satellites.

[0095] In this embodiment, the feature extraction module 205 includes a geographic feature unit and a temporal feature unit.

[0096] The geographic feature unit is used to extract features from the geographic data according to a preset feature extraction algorithm to obtain the first feature corresponding to the geographic data.

[0097] The temporal feature unit is used to extract temporal features in the time dimension of the meteorological data, the first temporal data, and the second temporal data respectively by sliding convolution kernels according to a preset convolutional neural network, so as to obtain the first temporal feature corresponding to the meteorological data and the second temporal feature corresponding to the first temporal data or the second temporal data.

[0098] In this embodiment, the feature splicing module 206 includes a convolution unit, a vector unit, and a splicing unit.

[0099] The convolutional unit is used to perform convolution operations on the first feature, the first temporal feature and the second temporal feature respectively. By sliding the convolutional kernel on the feature matrix corresponding to the first feature, the first temporal feature and the second temporal feature respectively, a first feature map corresponding to the first feature, a first temporal feature map corresponding to the first temporal feature and a second temporal feature map corresponding to the second temporal feature are obtained.

[0100] The vector unit is used to connect all elements in the first feature map, the first time-series feature map, and the second time-series feature map by rows or columns, and flatten them to obtain the first vector corresponding to the first feature map, the first time-series vector corresponding to the first time-series feature map, and the second time-series vector corresponding to the second time-series feature map.

[0101] The splicing unit is used to splice the first vector, the first time-series vector, and the second time-series vector to obtain the feature vector corresponding to the fire point.

[0102] This embodiment discloses a method for fire alarm of transmission lines based on multi-source data fusion. First, multi-source monitoring data of the target area where the transmission line is located is collected. This multi-source monitoring data is then used to perform fusion monitoring of the transmission line from multiple angles, improving monitoring accuracy. Next, preliminary monitoring is conducted using ground monitoring devices and drones to determine the presence of fire points. This combines close-range monitoring by ground devices with wide-area patrols by drones, ensuring comprehensive monitoring of thermal power plants. This improves monitoring accuracy while reducing drone patrol costs. When neither the ground monitoring devices nor the drones detect a fire point, to avoid missed fire detection due to blind spots of the ground monitoring devices and the patrol intervals of the drones, monitoring data from satellites located in the target area is acquired. The first step is to determine whether a fire exists based on this monitoring data. In wildfire detection, when fire points are detected in the data, it is determined whether the fire point was detected by a single satellite or multiple satellites. If multiple satellites are involved, the data from different satellites need to be fused to ensure that the processed fire point information has higher reliability. Furthermore, to address the potential influence of various factors on satellite observations, the fire point information and monitoring data over a period of time are fused together as a time series, serving as the judgment data for subsequent fire point identification, further improving the accuracy of fire point monitoring. During fire point monitoring, meteorological data and geographical environment of the target area can also lead to errors in fire point detection. Therefore, features of real-time meteorological and ground factors at the fire point location are extracted for fire point identification, and combined features are used with the corresponding time series data for fire point recognition. This approach considers past monitoring results during the monitoring period, improving the accuracy of fire point monitoring and identification.

[0103] Example 2

[0104] This embodiment discloses another method for fire alarm of transmission lines based on multi-source data fusion. The specific implementation process of the method can be referred to... Figure 3 It mainly includes steps 301 to 306, the steps being:

[0105] Step 301: Obtain ground device monitoring data and UAV monitoring data of the target area where the transmission line is located, and determine whether the ground device monitoring device or the UAV has detected a fire point based on the ground device monitoring data and the UAV monitoring data. If a fire point is detected, obtain the fire point information of the fire point based on the ground device monitoring device or the UAV, and issue an alarm based on the fire point information.

[0106] In this embodiment, the step is as follows: First, the monitoring data transmitted back by the ground monitoring device and the drone, which are used to lay out the transmission line according to the optimal layout principle, are obtained to monitor the target area where the transmission line is located in real time. Based on the monitoring data, it is determined whether the ground monitoring device or the drone has detected a fire point. When the ground monitoring device or the drone detects a fire point, the fire point information detected by the ground monitoring device or the drone is output, and an alarm is issued to the transmission line based on the fire point information.

[0107] Step 302: If the ground monitoring device or the UAV does not detect the fire point, then the monitoring data monitored by the satellite corresponding to the target area is obtained, and the first satellite that detected the fire point is determined based on the monitoring data.

[0108] In this embodiment, the step is as follows: acquiring monitoring data obtained when a polar-orbiting satellite or a geostationary satellite monitors the target area, and determining whether the polar-orbiting satellite or the geostationary satellite has detected a fire point based on the monitoring data; if the polar-orbiting satellite or the geostationary satellite has detected a fire point, then determining the first satellite that detected the fire point based on the monitoring data.

[0109] Furthermore, in order to improve the comprehensiveness of satellite monitoring while reducing the amount of data processing, the target area can be monitored based on a single type of satellite, either polar-orbiting or geostationary. Specifically, referring to the satellites currently accessed by China Southern Power Grid, there are 9 polar-orbiting satellites and 2 geostationary satellites. The spatiotemporal resolution of the polar-orbiting satellites and the geostationary satellites is:

[0110]

[0111]

[0112] As shown in the spatiotemporal resolution table above, polar-orbiting satellites are more numerous and cover a wider time period than geostationary satellites, resulting in greater monitoring capabilities. Therefore, we can first obtain the first satellite monitoring data detected by the polar-orbiting satellites, and then extract the region from the first satellite monitoring data to obtain the second satellite monitoring data corresponding to the target region. Based on the second satellite monitoring data, we can determine whether the polar-orbiting satellites have detected a fire point in the target region. If the polar-orbiting satellites have detected a fire point, we can determine the first polar-orbiting satellite that detected the fire point based on the second satellite monitoring data. If the polar-orbiting satellites have not detected a fire point in the target region, we can obtain the third satellite monitoring data detected by the geostationary satellites, and then extract the region from the third satellite monitoring data to obtain the fourth satellite monitoring data corresponding to the target region. Based on the fourth satellite monitoring data, we can determine whether the geostationary satellites have detected a fire point in the target region. If the geostationary satellites have detected a fire point, we can determine the first geostationary satellite that detected the fire point based on the fourth satellite monitoring data. The first satellite is either the first polar-orbiting satellite or the first geostationary satellite.

[0113] Step 303: Check whether the first satellite is a single satellite or multiple satellites. When the first satellite is a single satellite, obtain the fire point information observed by the single satellite and the first monitoring data within the first time period before the fire point is detected. Based on the fire point information and the first monitoring data, obtain the first time series data corresponding to the single satellite.

[0114] In this embodiment, the step specifically involves checking whether each polar-orbiting satellite or the geostationary satellite has detected a fire point. If a single polar-orbiting satellite or a single geostationary satellite observes a fire point, then the fire point information detected by the single polar-orbiting satellite or the single geostationary satellite, together with all the monitoring data from the previous half-hour monitoring period, constitutes time-series data. The fire point information refers to the geographical location of the wildfire, the time of the wildfire, and the pixel size of the fire. The monitoring data refers to the data emitted by each monitoring device.

[0115] Step 304: When the first satellite is multiple satellites, the first fire point information obtained by each satellite is fused to obtain the second fire point information, and the second monitoring data obtained by each satellite is fused to obtain the third monitoring data. Based on the second fire point information and the third monitoring data, the second time series data corresponding to the multiple satellites is obtained.

[0116] In this embodiment, the step is as follows: obtain the third fire point information obtained by each satellite observing the fire point and the second monitoring data within the first time period before each satellite detects the fire point; fuse the third fire point information and the second monitoring data of the multiple satellites according to a preset weighted average processing method to obtain the fourth fire point information after the fusion of the third fire point information of the multiple satellites and the third monitoring data after the fusion of the second monitoring data.

[0117] Specifically, if multiple polar-orbiting or geostationary satellites pass overhead simultaneously and all observe a fire point, it is necessary to fuse the observation data from different satellites. The fusion processing method includes weighted averaging of the detected fire point information and monitoring data to ensure that the processed fire point information and monitoring data have higher reliability.

[0118] The basic idea behind weighted averaging is to assign a weight to each observation, reflecting its importance or accuracy. The weights can be chosen based on the accuracy, reliability, and timeliness of the observing satellites. First, the weight of each satellite's observation data needs to be determined, which requires evaluating the satellite's performance and observation conditions. For example, if a satellite's observation equipment is more accurate, or its observation time is closer to the time of the fire, then it should be given a higher weight. After assigning weights, the weighted average is calculated. Specifically, this involves multiplying each observation by its weight, summing the results, and finally dividing by the sum of the weights. Finally, the weighted average is checked for reasonableness. For example, it is checked whether it falls within the range of all observations or whether it is consistent with information from other sources.

[0119] The processed data on the current fire point of the polar-orbiting or geostationary satellite, together with all the monitoring data from the previous half hour, constitutes time-series data.

[0120] Step 305: Obtain meteorological and geographical data of the target area, and extract the geographical features corresponding to the geographical data and the first time-series features and the second time-series features or the second time-series features corresponding to the first time-series data, respectively. Then, fuse the geographical features, the first time-series features and the second time-series features to obtain the feature vector corresponding to the fire point.

[0121] In this embodiment, the step includes: the data types involved in the acquired geographic data, meteorological data, and time-series data mainly include two categories: non-time-series data and time-series data, as shown in the table below:

[0122]

[0123] As shown in the table, for non-time-series data, features such as vegetation, geography, and anthropogenic factors are directly extracted from the table above; while for time-series data, a convolutional neural network (CNN) is used to extract time-series features by sliding convolution kernels. After feature extraction, various parameters generate corresponding feature matrices. A specific example of time-series feature extraction is provided below.

[0124] Suppose we have a one-dimensional time series: [1,0,0,1,1,1,0,1,0]. We use a convolutional kernel of size 3 to slide from left to right to extract the time series features.

[0125] The first slide yields the feature [1,0,0]. The second slide yields the feature [0,0,1]. This process continues until the last slide, resulting in the following feature sequence: [1,0,0], [0,0,1], [0,1,1], [1,1,1], [1,1,0], [1,0,1], [0,1,0]. This feature sequence can be organized into a feature matrix as shown below:

[0126]

[0127] This is used to extract time-series features, where t-2, t-1, and t represent the data from the two previous times, the one previous time, and the current time, respectively.

[0128] Next, the generated feature matrix is ​​convolved again. The convolution kernel slides across the feature matrix, performing element-wise multiplication and summation with the feature matrix obtained in the first step to obtain a new feature value. Through the convolution operation, the data is transformed into a new set of feature representations, with each convolution kernel generating a feature map. The convolution operation is as follows:

[0129]

[0130] In the first step of the example, the first three rows are used as the feature matrix. A convolution kernel is defined as follows: [0.1, 0.2, 0.3]. The convolution kernel is slid across the feature matrix, and each time it is multiplied element-wise with the corresponding part of the feature matrix and summed.

[0131] For the first row: 0.1*1 + 0.2*0 + 0.3*0 = 0.1;

[0132] For the second row: 0.1*0 + 0.2*0 + 0.3*1 = 0.3;

[0133] For the third row: 0.1*0 + 0.2*1 + 0.3*1 = 0.5;

[0134] This convolution operation yields a new set of feature representations: [0.1, 0.3, 0.5]. This results in a new feature map. With multiple convolution kernels, multiple such feature maps can be obtained. These feature maps are then stacked along the depth direction to form a new feature matrix.

[0135] All elements in the feature map are concatenated by rows or columns and flattened into a vector. These vectors are then concatenated to form a longer feature matrix. This concatenation can be performed in a fully connected layer, after which parameters are learned and adjusted to complete subsequent tasks. A specific calculation example for fully connected layer concatenation is as follows:

[0136] Suppose we have two feature maps generated from different convolution kernels, such as: feature Figure 1 [1.4, 2.0, 2.6], Features Figure 2 [1.5,2.1,2.7], concatenate these two feature maps column by column and flatten them into a long vector: [1.4,2.0,2.6,1.5,2.1,2.7].

[0137] Then, this vector is processed through a fully connected layer. A fully connected layer can be viewed as a linear transformation with a set of weight parameters inside. Assume the fully connected layer has the following weight matrix: [0.1 0.2 0.3 0.4 0.5 0.6]

[0139] Taking the dot product of the input vector and the weight matrix: 0.1*1.4 + 0.2*2.0 + 0.3*2.6 + 0.4*1.5 + 0.5*2.1 + 0.6*2.7 = 3.82, we obtain a new feature value. In reality, a fully connected layer has multiple output nodes, each with its own set of weights, thus generating a new feature vector.

[0140] Step 306: Identify the feature vector according to the preset classification algorithm, and determine whether the fire point is a real fire point. If the fire point is a real fire point, issue an alarm based on the second fire point information or the fourth fire point information.

[0141] In this embodiment, the step is as follows: The high-dimensional features, i.e., the feature vectors, are processed using a trained Support Vector Machine (SVM) algorithm. The output determines whether the fire point detected by the satellite is a real fire point. If it is a real fire point, the coordinates of the fire point detected by the satellite are output. These coordinates can be binary classification results or specific location coordinates.

[0142] This embodiment discloses a transmission line fire alarm method based on multi-source data fusion. By fusing multi-source data, the monitoring range and resolution of wildfires can be expanded. Different monitoring methods have different spatial and temporal coverage capabilities. Fusion can compensate for the shortcomings of single methods, achieving a wider monitoring range and finer spatial resolution, thus providing a more comprehensive understanding of the distribution of wildfires. It also overcomes data heterogeneity, as data from different monitoring methods are heterogeneous. By fusing these data, a unified data model and processing method are established, achieving seamless data fusion, resolving data inconsistencies, and reducing errors. When identifying fire points, real-time meteorological conditions, ground factors, and human influences at the hotspot location are comprehensively considered. Furthermore, time-series data is used to consider past monitoring results during the monitoring period, improving the accuracy of fire point monitoring and identification.

[0143] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A power transmission line fire alarm method based on multi-source data fusion, characterized in that, The application relates to a power transmission line fire point monitoring method and device. According to the obtained ground monitoring device monitoring data and unmanned aerial vehicle monitoring data, it is judged whether a preset ground monitoring device or unmanned aerial vehicle in a target region where a power transmission line is located monitors a fire point, if the fire point is monitored, first fire point information of the fire point is acquired, and alarm is given according to the first fire point information; If the fire point is not monitored, according to the obtained satellite monitoring data, it is judged whether a satellite corresponding to the target region monitors a fire point, if the fire point is monitored, meteorological data and geographical data of the target region are acquired, and a first satellite currently monitoring the fire point is determined according to the satellite monitoring data; When the first satellite is a single satellite, second fire point information obtained by the single satellite observing the fire point and first monitoring data in a first time period before the single satellite monitors the fire point are acquired, and first time sequence data corresponding to the single satellite is obtained according to the second fire point information and the first monitoring data; When the first satellite is a plurality of satellites, third fire point information obtained by each satellite observing the fire point and second monitoring data in a first time period before each satellite monitors the fire point are acquired, fourth fire point information after fusion of the third fire point information of the plurality of satellites and third monitoring data after fusion of the second monitoring data are obtained by respectively fusing the third fire point information and the second monitoring data of the plurality of satellites according to a preset weighted average processing method, and second time sequence data corresponding to the plurality of satellites is obtained according to the fourth fire point information and the third monitoring data; Feature extraction is performed on the geographical data, meteorological data, first time sequence data and second time sequence data, first features corresponding to the geographical data, first time sequence features corresponding to the meteorological data and second time sequence features corresponding to the first time sequence data or second time sequence data are obtained; Convolution operation, vectorization operation and splicing operation are respectively performed on the first features, first time sequence features and second time sequence features, and a feature vector corresponding to the first features, first time sequence features and second time sequence features after splicing is obtained; According to a preset classification algorithm, the feature vector is identified to judge whether the fire point is a real fire point, if the fire point is a real fire point, alarm is given according to the second fire point information or the fourth fire point information.

2. The power line fire alarm method based on multi-source data fusion according to claim 1, characterized in that, According to the obtained satellite monitoring data, it is judged whether a satellite corresponding to the target region monitors a fire point, if the fire point is monitored, meteorological data and geographical data of the target region are acquired, and a first satellite currently monitoring the fire point is determined according to the satellite monitoring data, including: First satellite monitoring data monitored by a polar orbit satellite is acquired, and region extraction is performed on the first satellite monitoring data to obtain second satellite monitoring data corresponding to the target region; According to the second satellite monitoring data, it is judged whether the polar orbit satellite monitors a fire point in the target region, if the polar orbit satellite monitors the fire point, a first polar orbit satellite monitoring the fire point is determined according to the second satellite monitoring data. If the polar orbit satellite does not monitor the existence of the fire point in the target area, third satellite monitoring data monitored by a geostationary satellite is acquired, and regional extraction is performed on the third satellite monitoring data to obtain fourth satellite monitoring data corresponding to the target area; According to the fourth satellite monitoring data, it is judged whether the geostationary satellite monitors the existence of the fire point in the target area, and if the geostationary satellite monitors the fire point, a first geostationary satellite monitoring the fire point is determined according to the fourth satellite monitoring data; the first satellite is the first polar orbit satellite or the first geostationary satellite.

3. The power line fire alarm method based on multi-source data fusion according to claim 1, characterized in that, The third fire point information and the second monitoring data of the plurality of satellites are fused according to the preset weighted average processing method to obtain fourth fire point information after fusion of the third fire point information of the plurality of satellites and third monitoring data after fusion of the second monitoring data, comprising: According to the evaluation level of the observation performance of each satellite in the plurality of satellites, the weight of each satellite in the plurality of satellites is determined; According to the weight and the third fire point information and the second monitoring data corresponding to each satellite, the plurality of satellites are weighted and averaged to obtain the corresponding fourth fire point information after fusion of the third fire point information of the plurality of satellites and the corresponding third monitoring data after fusion of the second monitoring data of the plurality of satellites.

4. The power line fire alarm method based on multi-source data fusion according to claim 1, characterized in that, The feature extraction on the geographic data, the meteorological data, the first time series data and the second time series data comprises: According to a preset feature extraction algorithm, the geographic data is subjected to feature extraction to obtain first features corresponding to the geographic data; According to a preset convolutional neural network, time series features are extracted on the time dimension of the meteorological data, the first time series data and the second time series data by sliding a convolution kernel to obtain first time series features corresponding to the meteorological data, the first time series data or second time series features corresponding to the second time series data.

5. The power line fire alarm method based on multi-source data fusion according to claim 1, characterized in that, The convolution operation, vectorization operation and splicing operation on the first features, first time series features and second time series features respectively comprise: The first features, first time series features and second time series features are subjected to convolution operation respectively, and a convolution kernel slides on the feature matrix corresponding to the first features, first time series features and second time series features respectively to obtain first feature maps corresponding to the first features, first time series feature maps corresponding to the first time series features and second time series feature maps corresponding to the second time series features; All elements in the first feature maps, first time series feature maps and second time series feature maps are connected by rows or columns to be flattened to obtain first vectors corresponding to the first feature maps, first time series vectors corresponding to the first time series feature maps and second time series vectors corresponding to the second time series feature maps; The first vectors, first time series vectors and second time series vectors are spliced to obtain a feature vector corresponding to the fire point.

6. A power transmission line fire alarm system based on multi-source data fusion, characterized in that, The system comprises an initial monitoring module, a satellite monitoring module, a single processing module, a multi-source processing module, a feature extraction module, a feature splicing module and a fire point alarm module. The initial monitoring module is configured to determine whether a preset ground monitoring device or unmanned aerial vehicle in a target area where the power transmission line is located monitors a fire point according to the obtained ground monitoring device monitoring data and unmanned aerial vehicle monitoring data, and if the fire point is monitored, obtain first fire point information of the fire point and perform an alarm according to the first fire point information; The satellite monitoring module is configured to, if the fire point is not monitored, determine whether a satellite corresponding to the target area monitors the fire point according to satellite monitoring data obtained by the satellite monitoring module, and if the fire point is monitored, obtain meteorological data and geographical data of the target area and determine a first satellite currently monitoring the fire point according to the satellite monitoring data; The single processing module is configured to, if the first satellite is a single satellite, obtain second fire point information obtained by the single satellite observing the fire point and first monitoring data in a first time period before the single satellite monitors the fire point, and obtain first time sequence data corresponding to the single satellite according to the second fire point information and the first monitoring data; The multi-source processing module is configured to, if the first satellite is a plurality of satellites, obtain third fire point information obtained by each satellite observing the fire point and second monitoring data in the first time period before each satellite monitors the fire point, respectively fuse the third fire point information and the second monitoring data of the plurality of satellites according to a preset weighted average processing method, obtain fourth fire point information obtained by fusing the third fire point information of the plurality of satellites and third monitoring data obtained by fusing the second monitoring data, and obtain second time sequence data corresponding to the plurality of satellites according to the fourth fire point information and the third monitoring data; The feature extraction module is configured to perform feature extraction on the geographical data, the meteorological data, the first time sequence data and the second time sequence data to obtain first features corresponding to the geographical data, first time sequence features corresponding to the meteorological data, and second time sequence features corresponding to the first time sequence data or the second time sequence data; The feature splicing module is configured to perform convolution operation, vectorization operation and splicing operation on the first features, the first time sequence features and the second time sequence features respectively to obtain a feature vector corresponding to the spliced first features, the first time sequence features and the second time sequence features; The fire point alarm module is configured to identify the feature vector according to a preset classification algorithm, determine whether the fire point is a real fire point, and if the fire point is a real fire point, perform an alarm according to the second fire point information or the fourth fire point information.

7. A power line fire alarm system based on multi-source data fusion as claimed in claim 6, wherein, The satellite monitoring module includes a polar orbit monitoring unit and a stationary monitoring unit; The polar orbit monitoring unit is configured to obtain first satellite monitoring data monitored by a polar orbit satellite, perform region extraction on the first satellite monitoring data to obtain second satellite monitoring data corresponding to the target area, determine whether the polar orbit satellite monitors a fire point in the target area according to the second satellite monitoring data, and if the polar orbit satellite monitors the fire point, determine a first polar orbit satellite monitoring the fire point according to the second satellite monitoring data. The static monitoring unit is configured to, if the polar orbit satellite does not monitor the existence of the fire point in the target region, acquire third satellite monitoring data monitored by a geostationary satellite, and perform regional extraction on the third satellite monitoring data to obtain fourth satellite monitoring data corresponding to the target region; and determine, according to the fourth satellite monitoring data, whether the geostationary satellite monitors the existence of the fire point in the target region, and if the geostationary satellite monitors the fire point, determine a first geostationary satellite that monitors the fire point according to the fourth satellite monitoring data; the first satellite is the first polar orbit satellite or the first geostationary satellite.

8. A power line fire alarm system based on multi-source data fusion as claimed in claim 6, wherein, The multi-source processing module comprises a weight determination unit and a data fusion unit. The weight determination unit is configured to determine the weight of each of the plurality of satellites according to the evaluation level of the observation performance of each of the plurality of satellites. The data fusion unit is configured to perform weighted average processing on the plurality of satellites according to the weight, the third fire point information corresponding to each of the plurality of satellites, and the second monitoring data, to obtain fourth fire point information corresponding to the third fire point information after fusion of the plurality of satellites, and third monitoring data corresponding to the second monitoring data after fusion of the plurality of satellites.

9. A power line fire alarm system based on multi-source data fusion as claimed in claim 6, wherein, The feature extraction module comprises a geographic feature unit and a time sequence feature unit. The geographic feature unit is configured to perform feature extraction on the geographic data according to a preset feature extraction algorithm to obtain first features corresponding to the geographic data. The time sequence feature unit is configured to extract time sequence features of the meteorological data, the first time sequence data, and the second time sequence data respectively in a time dimension by sliding a convolution kernel according to a preset convolutional neural network, to obtain first time sequence features corresponding to the meteorological data, the first time sequence data, or second time sequence features corresponding to the second time sequence data.

10. A power line fire alarm system based on multi-source data fusion as claimed in claim 6, wherein, The feature splicing module comprises a convolution unit, a vector unit, and a splicing unit. The convolution unit is configured to perform convolution operations on the first features, the first time sequence features, and the second time sequence features respectively by sliding a convolution kernel on feature matrices corresponding to the first features, the first time sequence features, and the second time sequence features respectively, to obtain a first feature map corresponding to the first features, a first time sequence feature map corresponding to the first time sequence features, and a second time sequence feature map corresponding to the second time sequence features. The vector unit is configured to connect all elements in the first feature map, the first time sequence feature map, and the second time sequence feature map by rows or columns respectively, and flatten to obtain a first vector corresponding to the first feature map, a first time sequence vector corresponding to the first time sequence feature map, and a second time sequence vector corresponding to the second time sequence feature map. The splicing unit is configured to splice the first vector, the first time sequence vector, and the second time sequence vector to obtain a feature vector corresponding to the fire point.

Images