Unmanned aerial vehicle active protection method and system for power system aiming at security threats

Abstract

The invention discloses a security threat-oriented unmanned aerial vehicle active protection method and system for a power system, and the method comprises the following steps: S1, collecting data through active patrol and a plurality of sensors by a defense unmanned aerial vehicle, and carrying out the detection of a surrounding environment through AI recognition; and S2, after the target is detected, the defense unmanned aerial vehicle performs intelligent behavior analysis on the detected unmanned aerial vehicle, and judges whether the unmanned aerial vehicle is a real security threat or not. According to the method, collaborative detection of multiple sensors is combined with AI recognition, the threat flyer can be effectively recognized, the false alarm rate is reduced, the detection accuracy is improved, in addition, intelligent behavior analysis is combined, the threat level is effectively evaluated, the risk of false alarm and misjudgment is reduced, automatic interception is conducted according to the evaluation result, manual intervention is not needed, and the risk of risk of threat detection is reduced. The safety protection effect is improved, and the power system safety is effectively guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of active protection of unmanned aerial vehicles (UAVs), and in particular to a method and system for active protection of UAVs against power system security threats. Background Art

[0002] With the rapid development of drone technology, drones have been widely used in power inspection, emergency repair and other fields. However, the intrusion of illegal drones may pose a threat to the safe operation of the power system, such as stealing sensitive data, maliciously damaging power facilities and even launching network attacks. Traditional power security systems only rely on cameras and radars for monitoring, and their detection accuracy is low. As a result, it is easy to misjudge the type of drone during the detection process, and the false alarm rate is high. In addition, the existing power security system can only perform simple alarms and cannot intelligently judge the threat level. There is a risk of false alarms and misjudgments, which affects the protection effect. Based on the above situation, this application proposes a drone active protection method and system for power systems against security threats. Summary of the Invention

[0003] Based on the technical problems existing in the background technology, the present invention proposes a method and system for active protection of power systems against security threats using drones.

[0004] The present invention proposes a method for actively protecting a power system from security threats using a drone, which is characterized by comprising the following steps:

[0005] S1: Defense drones collect data through active patrols and multiple sensors, and detect the surrounding environment through AI recognition;

[0006] S2: After the target is detected, the defense drone performs intelligent behavior analysis on the detected drone to determine whether it is a real security threat;

[0007] S3: Once a drone is assessed as a high-risk target, the system must take prompt action to intercept it and prevent the power system from being attacked;

[0008] S4: Feedback and data analysis of each drone intrusion incident to optimize the defense system's future threat identification and response speed.

[0009] Preferably, in S1, the multiple sensors include millimeter-wave radar, photoelectric imaging, radio spectrum analysis and acoustic sensors. The millimeter-wave radar is used for all-weather target detection, photoelectric imaging is used for target identification, radio spectrum analysis is used to monitor UAV remote control signals, and acoustic sensors are used to detect low-noise UAVs. The target feature data of the UAV is obtained through collaborative detection of multiple types of sensors.

[0010] Preferably, the specific logical steps of S1 are as follows:

[0011] S101: Acquire UAV target feature data through multiple sensors;

[0012] S102: Collect data from different sensors and perform intelligent temporal and spatial fusion of the detection data from different sensors based on a Kalman filter data fusion algorithm;

[0013] The formula used in its Kalman filter data fusion algorithm is as follows:

[0014] The current state estimation of the drone is predicted based on the state of the previous moment and the system dynamic model: =A k-1+ ;

[0015] in is the predicted state estimate, A is the state transfer matrix, k-1 is the state estimate at the previous moment, B is the control input matrix, is the control input at the current moment;

[0016] According to the actual measurement results, combined with the prediction results, the current state estimate is updated. + ;

[0017] in is the updated state estimate, is the state estimate for the prediction, is the actual measurement value at the current moment;

[0018] S103: Use deep learning algorithms to train a drone detection model. The target classification network in the drone detection model identifies drone targets and accurately distinguishes drones from other flying objects, including birds, aircraft, or windblown debris.

[0019] The specific steps for training the drone detection model are as follows:

[0020] S1031: Construct a dataset containing different categories of flying objects. This dataset includes image data and labels of the target objects. Enhance the images by rotating, flipping, cropping, and scaling. Label each image with the position and category of the object to form a labeled training dataset.

[0021] S1032: Extracting image features using convolutional neural networks;

[0022] S1033: Use cross entropy loss function for target classification;

[0023] S1034: Bounding box regression using smooth L1 loss function;

[0024] S1035: Train the target classification network, combining classification and regression losses to train the drone detection model;

[0025] During the recognition process, the trained drone detection model analyzes the input image, automatically detects objects in the image, and outputs the category and location of each object. The formula used in the recognition analysis process is as follows:

[0026] ;

[0027] in is the prediction score of category i, C is the total number of categories, is the predicted probability that the object belongs to category i; by calculating the probability that each object belongs to each category, the category with the maximum probability is output as the recognition result.

[0028] Preferably, the specific logical steps of S2 are as follows:

[0029] S201: The defensive drone uses a behavioral analysis algorithm to analyze whether the target is aggressive or has abnormal behavior based on the flight trajectory and behavior pattern of the flying object;

[0030] S202: Through multi-dimensional data analysis, threat level assessment is conducted, with conventional aircraft being low risk, unauthorized aircraft being medium risk, and potential offensive drones being high risk;

[0031] S203: The defense drone triggers a defense strategy based on the threat level. When the risk is low, a warning is issued and recorded. When the risk is medium, the power operation and maintenance personnel are notified to confirm and further track. When the risk is high, interception measures are immediately initiated to prevent damage to power grid facilities.

[0032] Preferably, in said S201, the use of a behavior analysis algorithm to determine whether there is aggressive or abnormal behavior includes trajectory change analysis and movement pattern analysis;

[0033] The trajectory change analysis uses the velocity, acceleration, and position changes of a trajectory to assess whether the behavior of a flying object conforms to a normal flight pattern. Aggressive behavior is manifested by a sudden change in speed or course, indicating a change in intent. When a flying object approaches a power facility target, it may be a potential threat.

[0034] By calculating the acceleration of a flying object, we can determine whether it has a sudden change in speed. The formula used is:

[0035] ;

[0036] in is the velocity vector at time t, Δv(t) is the velocity change, which is used to detect sudden acceleration or deceleration;

[0037] The threat level is assessed by calculating the distance of a flying object approaching a sensitive area. The formula used is:

[0038] (t)=min(|p(t)− ∣), where It is the center of the power facilities in the target area. (t) is the shortest distance between the flying object and the power facilities in the target area. When the distance is less than a certain threshold, it indicates that the flying object is a potential threat;

[0039] The motion law analysis is to determine whether the flying object conforms to the normal flight mode based on its motion trajectory. The formula used is:

[0040] , where p(t) is the position coordinate of the flying object at time t, p(t−1) is the position coordinate of the flying object at the previous time t−1, and Δp(t) is the position change between time t and t−1.

[0041] Preferably, in said S202, when performing threat level assessment through multi-dimensional data analysis, the weight of each indicator is set and adjusted according to its importance in the threat assessment, and the formula used is as follows:

[0042] T= , where T is the overall threat level, is the weight of the i-th evaluation indicator, 0≤ ≤1 and =1, is the value of the i-th evaluation indicator;

[0043] Threat Level= ,

[0044] If T is less than a certain threshold , it is judged as "low risk", if T is within a certain range, it is judged as "medium threat", if T is greater than a certain threshold , it is judged as “high threat”.

[0045] Preferably, the specific logical steps of S3 are as follows:

[0046] S301: After confirming a threat, the drone continuously tracks the target drone using multiple sensors to ensure accurate knowledge of its location and trajectory, accurately locking onto its position and flight path.

[0047] S302: Selecting an appropriate interception strategy based on the characteristics of the target, including radio jamming, physical capture, and laser strike;

[0048] S303: After successful interception, the defense system is activated and the power grid equipment is automatically dispatched for protection to ensure the normal operation of the power system.

[0049] Preferably, the specific steps of S4 are as follows:

[0050] S401: Record the process and effect of each interception and update the threat database;

[0051] S402: Clean and annotate the data recorded in S401, remove invalid or duplicate data, label each piece of data, and use a convolutional neural network classification model to detect and classify flying objects;

[0052] S403: Defending against drones continuously optimizes the drone detection model using deep learning algorithms combined with the data processed in S402;

[0053] The specific steps are as follows:

[0054] S4031: Perform data labeling to identify target category labels, threat labels, and interception strategy effectiveness labels.

[0055] S4032: Input the labeled data into the drone detection model, perform incremental training based on the convolutional neural network (CNN), and adjust the loss function to adjust the parameters of the drone detection model to achieve the purpose of optimizing the drone detection model;

[0056] S403: Conduct a detailed analysis of the recorded interception data to analyze the relationship between weather conditions, flying object types, and interception methods, identify which factors have the greatest impact on the interception success rate, and adjust the interception strategy and equipment deployment based on the results of the data analysis to ensure a more efficient response next time.

[0057] The present invention also proposes a drone active protection system for power systems against security threats, including a drone detection module, a data storage and analysis module, a threat assessment and intelligent judgment module, a drone response and interception module, a self-learning and optimization module, and a power system linkage module;

[0058] The drone detection module is responsible for active detection and target identification of drones;

[0059] The data storage and analysis module is used to store and analyze data collected by various sensors, as well as all interception events, alarm information, and feedback data, providing a basis for subsequent optimization;

[0060] The threat assessment and intelligent judgment module is used to perform intelligent behavior analysis on detected drones to determine whether they are real security threats;

[0061] The drone response and interception module is used to take action to intercept the drone when it is detected as a high-risk target to prevent the power system from being attacked;

[0062] The self-learning and optimization module is used to provide feedback and data analysis for each drone intrusion incident, optimizing the efficiency and capabilities of the defense system;

[0063] The power system linkage module is used to link with the power system when intercepting threats to ensure the normal operation of the power system.

[0064] Compared with the existing technology, the beneficial effects of the present invention are:

[0065] By combining collaborative detection with multiple sensors and AI recognition, the present invention can effectively identify threatening flying objects, reduce false alarm rates, and improve detection accuracy. In addition, combined with intelligent behavior analysis, it can effectively assess the threat level, reduce the risk of false alarms and misjudgments, and automatically intercept based on the assessment results without the need for human intervention, thereby improving safety protection effects and effectively ensuring the safety of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is a flow chart of the UAV active protection method for power systems against security threats proposed by the present invention;

[0067] Figure 2 This is a block diagram of an active protection system for unmanned aerial vehicles (UAVs) against security threats, which is applied to power systems and proposed by the present invention. DETAILED DESCRIPTION

[0068] By calculating the position change at consecutive moments, it is possible to determine whether the flying object's movement suddenly accelerates, decelerates, or deviates from the normal path (if Δp(t) is significantly greater than the historical average, it may indicate that the drone is rapidly approaching sensitive areas (such as power facilities)). Normal behavior: uniform speed or smooth trajectory (Δp(t) changes gently); abnormal behavior: sudden position changes (such as a sudden increase in Δp(t)), wandering, or repeated approach to critical facilities.

[0069] For example, the position coordinates of a drone at adjacent times are: p(t−1)=(100 m, 200 m) at time t−1 and p(t)=(120 m, 210 m) at time t. Then the displacement change is: Δp(t)= If the historical average displacement is 10 m, this proves that the current position is abnormal.

[0070] The present invention will be further explained below with reference to specific embodiments.

[0071] Example

[0072] Reference Figure 1-2 This embodiment proposes a method for actively protecting a power system from security threats using drones, including the following steps:

[0073] S1: Defense drones collect data through active patrols and multiple sensors, and detect the surrounding environment through AI recognition;

[0074] The various sensors include millimeter-wave radar, optoelectronic imaging (visible light + infrared), radio spectrum analysis, and acoustic sensors. Millimeter-wave radar is used for all-weather target detection, optoelectronic imaging is used for target identification, radio spectrum analysis is used to monitor drone remote control signals, and acoustic sensors are used to detect low-noise drones. Multi-sensor collaborative detection obtains drone target feature data.

[0075] The specific logical steps are as follows:

[0076] S101: Acquire UAV target feature data through multiple sensors;

[0077] S102: Collect data from different sensors and perform intelligent temporal and spatial fusion of the detection data from different sensors based on a Kalman filter data fusion algorithm;

[0078] The formula used in its Kalman filter data fusion algorithm is as follows:

[0079] The current state estimation of the drone is predicted based on the state of the previous moment and the system dynamic model: =A k-1+ ;

[0080] in is the predicted state estimate, A is the state transfer matrix, k-1 is the state estimate at the previous moment, B is the control input matrix, is the control input at the current moment;

[0081] According to the actual measurement results, combined with the prediction results, the current state estimate is updated. + ;

[0082] in is the updated state estimate, is the state estimate for the prediction, is the actual measurement value at the current moment;

[0083] S103: Use deep learning algorithms to train a drone detection model. The target classification network in the drone detection model identifies drone targets and accurately distinguishes drones from other flying objects, including birds, aircraft, or windblown debris.

[0084] The specific steps for training the drone detection model are as follows:

[0085] S1031: Construct a dataset containing different categories of flying objects (drones, birds, aircraft, and wind-blown debris). This data includes image data and labels of the target objects. Enhance the images by rotating, flipping, cropping, and scaling. Annotate each image with the position and category of the object to form a labeled training dataset.

[0086] S1032: Extracting image features using convolutional neural networks (CNNs);

[0087] S1033: Use cross entropy loss function for target classification;

[0088] S1034: Bounding box regression using smooth L1 loss function;

[0089] S1035: Train the target classification network, combining classification and regression losses to train the drone detection model;

[0090] During the recognition process, the trained drone detection model analyzes the input image, automatically detects objects in the image, and outputs the category (drone, bird, aircraft, wind-blown foreign object, etc.) and location (bounding box) of each object. The formula used in the recognition analysis process is as follows:

[0091] ;

[0092] in is the prediction score of category i, C is the total number of categories, is the predicted probability that the object belongs to category i; by calculating the probability that each object belongs to each category, and outputting the category with the maximum probability as the recognition result;

[0093] S2: After the target is detected, the defense drone performs intelligent behavior analysis on the detected drone to determine whether it is a real security threat;

[0094] The specific logical steps are as follows:

[0095] S201: The defensive drone uses a behavioral analysis algorithm to analyze whether the target is aggressive or has abnormal behavior based on the flight trajectory and behavior pattern of the flying object;

[0096] When using behavioral analysis algorithms to determine whether there is aggressive or abnormal behavior, it includes trajectory change analysis and movement pattern analysis;

[0097] ① Trajectory change analysis uses trajectory velocity, acceleration, and position changes to assess whether an object's behavior conforms to conventional flight patterns. Aggressive behavior is manifested by sudden changes in speed or course, indicating a change in intent. When an object approaches a power facility, it may be a potential threat.

[0098] By calculating the acceleration of a flying object, we can determine whether it has a sudden change in speed. The formula used is:

[0099] ;

[0100] in is the velocity vector at time t, Δv(t) is the velocity change, which is used to detect sudden acceleration or deceleration;

[0101] The threat level is assessed by calculating the distance of a flying object approaching a sensitive area. The formula used is:

[0102] (t)=min(|p(t)− ∣), where It is the center of the power facilities in the target area. (t) is the shortest distance between the flying object and the power facilities in the target area. When the distance is less than a certain threshold, it indicates that the flying object is a potential threat;

[0103] ② Motion pattern analysis is to determine whether the flying object's trajectory conforms to the normal flight pattern. Certain behaviors that deviate significantly from normal flying object behavior may be abnormal. The formula used is:

[0104] , where p(t) is the position information of the object and Δp(t) is the position change;

[0105] S202: Through multi-dimensional data analysis, threat level assessment is conducted, with conventional aircraft being low risk, unauthorized aircraft being medium risk, and potential offensive drones being high risk;

[0106] When conducting threat level assessment through multidimensional data analysis, the weight of each indicator is set and adjusted according to its importance in the threat assessment. The formula used is as follows:

[0107] T= , where T is the overall threat level, is the weight of the i-th evaluation indicator, 0≤ ≤1 and =1, is the value of the i-th evaluation indicator;

[0108] Threat Level= ,

[0109] If T is less than a certain threshold , it is judged as "low risk", if T is within a certain range, it is judged as "medium threat", if T is greater than a certain threshold , it is judged as “high threat”;

[0110] S203: Defense strategies are triggered based on the threat level of the drone. Warnings are issued and recorded when the risk is low. Power operation and maintenance personnel are notified for confirmation and further tracking when the risk is medium. Interception measures are immediately initiated when the risk is high to prevent damage to power grid facilities.

[0111] S3: Once a drone is assessed as a high-risk target, the system must take prompt action to intercept it and prevent the power system from being attacked;

[0112] The specific logical steps are as follows:

[0113] S301: After confirming a threat, the drone continuously tracks the target drone using multiple sensors to ensure accurate knowledge of its location and trajectory, accurately locking onto its position and flight path.

[0114] S302: Selecting an appropriate interception strategy based on the target's characteristics (flight pattern, threat level), including radio jamming, physical capture, and laser strike.

[0115] The interception strategy is selected based on the following target characteristics:

[0116] (1) Threat level (S202 assessment result): For high-risk targets (such as offensive drones), laser strikes (rapid destruction) are preferred; for medium-risk targets (such as unauthorized drones), radio jamming (non-destructive expulsion) is preferred; for low-risk targets (such as birds or stray aircraft), warnings are issued or ignored, and no interception is triggered;

[0117] (2) Target flight pattern and behavioral characteristics: For high-speed maneuvering targets, choose laser strike (instantaneous hit); for low-altitude, slow-moving targets, choose physical capture (precisely control the landing point to avoid crash risk); for cluster targets, choose radio jamming (simultaneously jamming multiple aircraft);

[0118] (3) Target technical characteristics: small drones for physical capture, large drones for laser strikes;

[0119] It makes a selection through a decision tree in the prior art, inputting threat level, target trajectory, and sensor data into the decision tree. The program code it runs is as follows:

[0120] If Threat Level = High Risk & Target Speed ​​> Threshold & Environment = Open Area:

[0121] Select Laser Strike

[0122] elif Threat Level = Medium & Target Relies on GPS:

[0123] Select Radio Interference

[0124] else:

[0125] Physical capture + manual confirmation;

[0126] For example, a drone suddenly accelerates (Δv(t)>10 m / s2) and approaches a substation (Dthreat(t)<50 m). This is assessed as a high risk and the data can be input into a decision tree. The decision tree output is: select laser strike, which can quickly destroy the substation and avoid hitting it.

[0127] S303: After successful interception, the defense system is activated and automatically dispatches power grid equipment for protection to ensure the normal operation of the power system;

[0128] S4: Feedback and data analysis of each drone intrusion incident will optimize the efficiency and capabilities of the defense system and improve the speed of future threat identification and response;

[0129] The specific steps are as follows:

[0130] S401: Record the process and effect of each interception and update the threat database;

[0131] S402: Clean and annotate the data recorded in S401, remove invalid or duplicate data, and label each piece of data (whether it is a threat, interception success or failure, target type). Use a convolutional neural network classification model to detect and classify flying objects.

[0132] S403: Defending against drones continuously optimizes the drone detection model using deep learning algorithms combined with the data processed in S402;

[0133] S403: Conduct a detailed analysis of the recorded interception data to analyze the relationship between weather conditions, flying object types, and interception methods, identify which factors have the greatest impact on the interception success rate, and adjust the interception strategy and equipment deployment based on the results of the data analysis to ensure a more efficient response next time.

[0134] This embodiment also proposes an active drone protection system for power systems and security threats, including a drone detection module, a data storage and analysis module, a threat assessment and intelligent judgment module, a drone response and interception module, a self-learning and optimization module, and a power system linkage module;

[0135] The drone detection module is responsible for active detection and target identification of drones;

[0136] The data storage and analysis module is used to store and analyze data collected by various sensors, as well as all interception events, alarm information, and feedback data, providing a basis for subsequent optimization;

[0137] The threat assessment and intelligent judgment module is used to perform intelligent behavior analysis on detected drones to determine whether they are real security threats;

[0138] The drone response and interception module is used to take action to intercept drones when they are detected as high-risk targets to prevent the power system from being attacked;

[0139] The self-learning and optimization module is used to provide feedback and data analysis for each drone intrusion incident, optimizing the efficiency and capabilities of the defense system;

[0140] The power system linkage module is used to link with the power system when intercepting threats to ensure the normal operation of the power system;

[0141] This embodiment uses a combination of collaborative detection by multiple sensors and AI recognition to effectively identify threatening flying objects, reduce false alarm rates, and improve detection accuracy. In addition, combined with intelligent behavior analysis, it effectively assesses the threat level, reduces the risk of false alarms and misjudgments, and automatically intercepts based on the assessment results without the need for human intervention, thereby improving safety protection effects and effectively ensuring the safety of the power system.

[0142] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for active protection of power systems against security threats using drones, characterized in that: The following steps are involved: S1: Defense drones collect data through active patrols and multiple sensors, and detect the surrounding environment through AI recognition; S2: After the target is detected, the defense drone performs intelligent behavior analysis on the detected drone to determine whether it is a real security threat; S3: Once a drone is assessed as a high-risk target, the system must take prompt action to intercept it and prevent the power system from being attacked; S4: Feedback and data analysis of each drone intrusion incident to optimize the defense system's future threat identification and response speed.

2. The method for active drone protection against security threats in a power system according to claim 1, characterized in that: In S1, the multiple sensors include millimeter-wave radar, optoelectronic imaging, radio spectrum analysis and acoustic sensors. The millimeter-wave radar is used for all-weather target detection, optoelectronic imaging is used for target identification, radio spectrum analysis is used to monitor UAV remote control signals, and acoustic sensors are used to detect low-noise UAVs. The target feature data of UAVs is obtained through the collaborative detection of multiple types of sensors.

3. The method for active protection of power system against security threats by drones according to claim 2, characterized in that: The specific logical steps of S1 are as follows: S101: Acquire UAV target feature data through multiple sensors; S102: Collect data from different sensors and perform intelligent temporal and spatial fusion of the detection data from different sensors based on a Kalman filter data fusion algorithm; The formula used in its Kalman filter data fusion algorithm is as follows: The current state estimation of the drone is predicted based on the state of the previous moment and the system dynamic model: =A k-1+ ; in is the predicted state estimate, A is the state transfer matrix, k-1 is the state estimate at the previous moment, B is the control input matrix, is the control input at the current moment; According to the actual measurement results, combined with the prediction results, the current state estimate is updated. + ; in is the updated state estimate, is the state estimate for the prediction, is the actual measurement value at the current moment; S103: Use deep learning algorithms to train a drone detection model. The target classification network in the drone detection model identifies drone targets and accurately distinguishes drones from other flying objects, including birds, aircraft, or windblown debris. The specific steps for training the drone detection model are as follows: S1031: Construct a dataset containing different categories of flying objects. This dataset includes image data and labels of the target objects. Enhance the images by rotating, flipping, cropping, and scaling. Label each image with the position and category of the object to form a labeled training dataset. S1032: Extracting image features using convolutional neural networks; S1033: Use cross entropy loss function for target classification; S1034: Bounding box regression using smooth L1 loss function; S1035: Train the target classification network, combining classification and regression losses to train the drone detection model; During the recognition process, the trained drone detection model analyzes the input image, automatically detects objects in the image, and outputs the category and location of each object. The formula used in the recognition analysis process is as follows: ； in is the prediction score of category i, C is the total number of categories, is the predicted probability that the object belongs to category i; by calculating the probability that each object belongs to each category, the category with the maximum probability is output as the recognition result.

4. The method for active protection of power system against security threats by drones according to claim 3, characterized in that: The specific logical steps of S2 are as follows: S201: The defensive drone uses a behavioral analysis algorithm to analyze whether the target is aggressive or has abnormal behavior based on the flight trajectory and behavior pattern of the flying object; S202: Through multi-dimensional data analysis, threat level assessment is conducted, with conventional aircraft being low risk, unauthorized aircraft being medium risk, and potential offensive drones being high risk; S203: The defense drone triggers a defense strategy based on the threat level. When the risk is low, a warning is issued and recorded. When the risk is medium, the power operation and maintenance personnel are notified to confirm and further track. When the risk is high, interception measures are immediately initiated to prevent damage to power grid facilities.

5. The method for active protection of power system against security threats by drones according to claim 4, characterized in that: In S201, the use of a behavior analysis algorithm to determine whether there is aggressive or abnormal behavior includes trajectory change analysis and movement pattern analysis; The trajectory change analysis uses the velocity, acceleration, and position changes of a trajectory to assess whether the behavior of a flying object conforms to a normal flight pattern. Aggressive behavior is manifested by a sudden change in speed or direction of a flying object, indicating a change in its intentions. When flying objects approach power facilities, they may be potential threats; By calculating the acceleration of a flying object, we can determine whether it has a sudden change in speed. The formula used is: ； in is the velocity vector at time t, Δv(t) is the velocity change, which is used to detect sudden acceleration or deceleration; The threat level is assessed by calculating the distance of a flying object approaching a sensitive area. The formula used is: (t)=min(|p(t)− ∣), where It is the center of the power facilities in the target area. (t) is the shortest distance between the flying object and the power facilities in the target area. When the distance is less than a certain threshold, it indicates that the flying object is a potential threat; The motion law analysis is to determine whether the flying object conforms to the normal flight mode based on its motion trajectory. The formula used is: , where p(t) is the position coordinate of the flying object at time t, p(t−1) is the position coordinate of the flying object at the previous time t−1, and Δp(t) is the position change between time t and t−1.

6. The method for active drone protection against security threats in a power system according to claim 4 or 5, characterized in that: In S202, when performing threat level assessment through multi-dimensional data analysis, the weight of each indicator is set and adjusted according to its importance in the threat assessment. The formula used is as follows: T= , where T is the overall threat level, is the weight of the i-th evaluation indicator, 0≤ ≤1 and =1, is the value of the i-th evaluation indicator; Threat Level= , If T is less than a certain threshold , it is judged as "low risk", if T is within a certain range, it is judged as "medium threat", if T is greater than a certain threshold , it is judged as "high threat".

7. The UAV for protecting power system from security threats according to claim 6 The active protection method is characterized in that: The specific logical steps of S3 are as follows: S301: After confirming a threat, the drone continuously tracks the target drone using multiple sensors to ensure accurate knowledge of its location and trajectory, accurately locking onto its position and flight path. S302: Selecting an appropriate interception strategy based on the characteristics of the target, including radio jamming, physical capture, and laser strike; S303: After successful interception, the defense system is activated and the power grid equipment is automatically dispatched for protection to ensure the normal operation of the power system.

8. The method for active protection of power system against security threats by using drones according to claim 7, characterized in that: The specific steps of S4 are as follows: S401: Record the process and effect of each interception and update the threat database; S402: Clean and annotate the data recorded in S401, remove invalid or duplicate data, label each piece of data, and use a convolutional neural network classification model to detect and classify flying objects; S403: Defending against drones continuously optimizes the drone detection model using deep learning algorithms combined with the data processed in S402; The specific steps are as follows: S4031: Perform data labeling to identify target category labels, threat labels, and interception strategy effectiveness labels. S4032: Input the labeled data into the drone detection model, perform incremental training based on the convolutional neural network (CNN), and adjust the loss function to adjust the parameters of the drone detection model to achieve the purpose of optimizing the drone detection model; S403: Conduct a detailed analysis of the recorded interception data to analyze the relationship between weather conditions, flying object types, and interception methods, identify which factors have the greatest impact on the interception success rate, and adjust the interception strategy and equipment deployment based on the results of the data analysis to ensure a more efficient response next time.

9. An unmanned aerial vehicle active protection system for power systems against security threats, used to implement the method according to any one of claims 1 to 8, characterized in that: It includes drone detection module, data storage and analysis module, threat assessment and intelligent judgment module, drone response and interception module, self-learning and optimization module, and power system linkage module; The drone detection module is responsible for active detection and target identification of drones; The data storage and analysis module is used to store and analyze data collected by various sensors, as well as all interception events, alarm information, and feedback data, providing a basis for subsequent optimization; The threat assessment and intelligent judgment module is used to perform intelligent behavior analysis on detected drones to determine whether they are real security threats; The drone response and interception module is used to take action to intercept the drone when it is detected as a high-risk target to prevent the power system from being attacked; The self-learning and optimization module is used to provide feedback and data analysis for each drone intrusion incident, optimizing the efficiency and capabilities of the defense system; The power system linkage module is used to link with the power system when intercepting threats to ensure the normal operation of the power system.

Images