Man-machine mixed transformer substation collision-free intelligent inspection system

Abstract

The invention relates to the technical field of unmanned aerial vehicle electric power inspection, and discloses a man-machine mixed transformer substation collision-free intelligent inspection system, which comprises the steps of transformer substation inspection personnel intention recognition: performing operation intention judgment according to the propulsion amount and swing degree of an operator control rod and the heart rate of an operator input by a multi-sensor system; and collision-free flight under a man-machine hybrid control algorithm: according to the intention recognition result of the inspection personnel, performing convolution operation on the input data of the millimeter wave radar carried by the unmanned aerial vehicle and the motion state of the unmanned aerial vehicle, and judging the autonomous auxiliary flight takeover degree of the unmanned aerial vehicle in combination with the flight track prediction of the unmanned aerial vehicle to achieve collision-free flight in the inspection process. By applying the unmanned aerial vehicle, a common substation inspector can be allowed to fly safely without collision in the process of operating the unmanned aerial vehicle to inspect, and compared with an existing passive defense unmanned aerial vehicle, a full-automatic unmanned aerial vehicle inspection system which can only plan a route in advance and other methods, the unmanned aerial vehicle inspection system has the advantages of being low in operation threshold, high in safety, high in maneuverability and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) power inspection, and in particular to a collision-free intelligent inspection system for a substation under human-machine hybrid conditions. Background Art

[0002] The power system is the lifeblood of the national economy, and ensuring the stable operation of its various components is crucial. As a key node in the power system, the substation is responsible for transmitting electricity from the power station to various power consumption areas. The equipment in the substation, such as transformers and switchgear, is exposed to the external environment for a long time and is easily eroded and damaged by natural factors. Once these equipment fail, they may cause large-scale power outages, causing serious impacts on the national economy and social life. Therefore, regular inspection and maintenance of substation equipment is a necessary measure to ensure the safe operation of the power system.

[0003] In recent years, with the rapid development of drone technology, the application of drones in substation inspection has gradually attracted attention. Traditional substation inspection methods usually use manual inspection or fixed camera monitoring. These methods have problems such as low work efficiency, limited detection range, and great influence by environmental factors. The introduction of drone inspection system provides new solutions to these problems. Traditional drone inspection control methods mainly include the following:

[0004] 1. Manual control: The operator manually controls the flight of the drone through a remote control device and uses the camera on the drone to monitor the status of the substation equipment in real time. This method relies on the operator's experience and skills and can flexibly cope with complex environments, but it has high requirements for the operator and the operation process is time-consuming.

[0005] 2. Preset path control: Before the inspection task begins, the operator presets the flight path and inspection target of the drone through the ground station. The drone completes the inspection task autonomously according to the set path. This method can improve the inspection efficiency, but it is difficult to make timely adjustments in the face of emergencies or environmental changes, and the inspection flexibility is poor.

[0006] 3. Visual servo control: The drone obtains image information of substation equipment through the onboard camera, and uses visual feedback signals to perform closed-loop control of the drone's position to achieve automatic inspection of the equipment. This method relies on machine vision technology and can reduce human intervention to a certain extent, but in complex environments such as strong light, rain and fog, visual sensors are easily disturbed, which may affect the accuracy and safety of inspections.

[0007] The present invention combines the human-machine hybrid control algorithm and trajectory prediction algorithm on the basis of the traditional inspection system, so that the operator can intervene in real time during the inspection process. At the same time, the UAV can automatically adjust the flight strategy according to the operator's intention to ensure collision-free flight. Through multi-sensor data, the system can accurately judge the operator's intention and autonomously adjust the response to the control signal, greatly reducing the dependence on the operator's experience, while improving the safety and maneuverability of the UAV in complex environments. Summary of the invention

[0008] Traditional substation inspection methods usually use manual inspections or fixed camera monitoring. These methods have problems such as low work efficiency, limited detection range, and great influence of environmental factors. Traditional drone inspection control methods mainly include the following: 1. Manual control, 2. Preset path control, 3. Visual servo control. In order to solve the above technical problems, the present invention provides a collision-free intelligent inspection system for substations under human-machine hybrid. On the basis of the traditional inspection system, it combines the human-machine hybrid control algorithm and the trajectory prediction algorithm, so that the operator can intervene in real time during the inspection process. At the same time, the drone can automatically adjust the flight strategy according to the operator's intention to ensure collision-free flight. Through multi-sensor data, the system can accurately judge the operator's intention and autonomously adjust the response to the control signal, which greatly reduces the dependence on the operator's experience and improves the safety and maneuverability of the drone in complex environments.

[0009] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0010] A collision-free intelligent inspection system for substations under human-machine hybrid, comprising:

[0011] Input layer: After the drone takes off, the input layer obtains data through the multi-sensor system and the drone system, including obtaining the drone's handle control stick thrust, the drone's handle control stick swing, and the operator's heart rate through the multi-sensor, and obtaining millimeter-wave radar sampling data and drone motion state data through the drone system; construct the drone's spatial state matrix M based on the millimeter-wave radar sampling data; and construct the drone's motion state matrix K through the drone's motion state data;

[0012] Processing layer: The processing layer performs weighted summation of the data obtained by the multi-sensor system through the control system judgment logic, determines the possibility of an imminent collision of the drone, and predicts the flight trajectory of the drone; performs convolution operation on the drone's spatial state matrix and the drone's motion state matrix through the collision-free flight technology processing logic to obtain the convolution result matrix D; and locates the maximum value in the convolution result matrix to obtain the direction in which the drone has the highest collision risk when in motion;

[0013] Decision-making layer: During the flight of the drone, the decision-making layer analyzes the control behavior and psychological state of the operator, determines the operator's control level, adjusts the weight of the drone's autonomous flight level based on the operator's control level, and generates drone control instructions based on the predicted flight trajectory of the drone and the direction in which the drone has the highest collision risk when in motion;

[0014] Execution layer: The execution layer converts the instructions generated by the decision layer into unmanned flight actions and controls the drone.

[0015] Furthermore, the inspection system also includes an output layer; the output layer is used to generate inspection reports, flight path records and obstacle avoidance logs, and when danger or abnormality is found, an alarm is issued and emergency measures are initiated.

[0016] Furthermore, in the input layer, the data is obtained through the multi-sensor system and the UAV system in the input layer, including obtaining the UAV handle control stick thrust, the UAV handle control stick swing, and the operator's heart rate through the multi-sensor, and obtaining the millimeter-wave radar sampling data and the UAV motion state data through the UAV system; constructing the UAV spatial state matrix M based on the millimeter-wave radar sampling data; and constructing the UAV motion state matrix K through the UAV motion state data, specifically including:

[0017] The amount of thrust T of the drone's handlebar reflects the current acceleration of the drone, and is T∈[0,1], where 0 means no thrust, i.e. the drone is hovering or stationary, and 1 means full thrust, i.e. the drone maintains maximum acceleration in flight;

[0018] The swing degree S of the drone's handle control stick reflects the operator's panic level, and takes S∈[0,1], where 0 means the control stick is completely stable and 1 means the control stick swings to the maximum extent;

[0019] The operator's heart rate H reflects the operator's psychological stress and panic level, and is expressed by a value normalized relative to the baseline heart rate. H∈[0,1] is taken, where 0 represents a normal heart rate and 1 represents an extremely high heart rate;

[0020] The drone spatial state matrix M is used to represent the millimeter-wave radar sampling data. The middle position of the drone spatial state matrix M represents the drone itself, and the values ​​of the surrounding positions represent the existence of obstacles. The element M[i][j]=0 in M ​​indicates that there is no obstacle at the position (i, j); M[i][j]=1 indicates that there is an obstacle at the position (i, j).

[0021] The drone motion state matrix K is used to represent the drone motion state data. The middle position of the drone motion state matrix K represents the drone itself, and the values ​​of the surrounding positions represent the movement speed of the drone in the corresponding direction; the element K[i][j] in K represents the direction and speed of the drone at position (i, j).

[0022] Furthermore, the processing layer performs weighted summation of the data acquired by the multi-sensor system through the control system judgment logic to judge the possibility of the UAV being about to collide, specifically including:

[0023] Set an indicator C to evaluate the probability that the drone is about to collide:

[0024] C=w T ·T+w S ·S+w H H;

[0025] Among them, w T 、w S 、w H is the weight coefficient, satisfying w T +w S +w H =1;

[0026] The closer the index C is to 1, the higher the possibility of a UAV collision is, and the closer the index C is to 0, the lower the possibility of a UAV collision is.

[0027] Furthermore, the prediction of the flight trajectory of the drone specifically includes:

[0028] The drone state vector x is defined as:

[0029]

[0030] Among them, x k ,y k They represent the horizontal and vertical coordinates of the position of the UAV at time k, v xk ,v yk They represent the lateral and longitudinal components of the UAV’s velocity at time k respectively;

[0031] The process noise covariance matrix Q in the filter is dynamically adjusted according to the speed, acceleration and motion state matrix K of the drone:

[0032] Q k =Q0+γ·‖K‖;

[0033] Where: Q0 is the initial process noise covariance matrix, γ is the adjustment factor, ‖K‖ represents the norm of the UAV motion state matrix K at the current moment; Q krepresents the process noise covariance matrix at time k;

[0034] A weight matrix W is introduced to perform weighted processing on the millimeter-wave radar sampling data and the data obtained by the multi-sensor system; the value of the weight matrix W is dynamically adjusted according to the operator's heart rate, the thrust change of the handle, etc.:

[0035] W k =α·G+β·(1-G);

[0036] Among them, W k Represents the weight matrix at time k; α and β are weight factors that can be dynamically adjusted according to the operator's heart rate H and the swing degree S of the drone's handle control stick: H max is the maximum heart rate, S max is the maximum swing value of the joystick control rod; G is the observation matrix;

[0037] Predict the state of the drone at the next moment based on the estimated state at the current moment and the state transfer equation:

[0038]

[0039] Among them, λ is the fusion factor, A and B represent the state transfer matrix and the control input matrix respectively. represents the predicted state from time k to time k+1, represents the estimated state at time k, u k represents the control input at time k, z k represents the millimeter wave radar sampling data at time k;

[0040] Prediction covariance:

[0041] P k+1|k =AP k A T +Q k ;

[0042] P k+1|k Represents the state covariance matrix, which gives the error of the prediction result based on time k at time k+1; updates the prediction state according to the dynamic environment weight adjustment and multimodal data fusion:

[0043] L k+1 =P k+1|k W k T (W k P k+1|k W k T +R) -1 ;

[0044]

[0045] Update the covariance:

[0046] P k+1 =(IL k+1 W k ) k+1|k

[0047] Among them, P k+1 represents the state covariance matrix at time k+1, which represents the error of the estimated state at time k+1, L k+1 The Kalman gain represents the time k+1, and R is the observation noise covariance matrix;

[0048] For the current moment k, whenever the millimeter-wave radar sampling data z at the next moment k+1 When the drone arrives, the above content is used for prediction, which can infer the specific location of the drone at the next moment (x k+1 ,y k+1 ) and speed (v x(k+1) ,v y(k+1) ).

[0049] Furthermore, the collision-free flight technology processing logic performs a convolution operation on the drone space state matrix and the drone motion state matrix to obtain a convolution result matrix D; and locates the maximum value in the convolution result matrix to obtain the direction in which the drone has the highest collision risk in the motion state, specifically including:

[0050] The convolution operation is:

[0051]

[0052] Among them, D[i,j] represents the element in the convolution result matrix, and the corresponding value represents the collision risk of the drone at position (i,j);

[0053] By locating the maximum value in the convolution result matrix, the direction in which the UAV has the highest collision risk when in motion is obtained.

[0054] Compared with the prior art, the beneficial technical effects of the present invention are:

[0055] 1. Introduction of human-machine hybrid control mode:

[0056] Traditional drone inspection systems are mainly divided into two modes: fully automatic and manually operated. Fully automatic systems usually rely on pre-planned routes and lack flexibility; while manual operation is flexible, it has high requirements for operators, especially in complex environments, and collisions are prone to occur. The present invention combines the operator's intention recognition with the autonomous flight capability of the drone through a human-machine hybrid control mode, so that the system can retain the flexibility of manual operation while improving the safety and intelligence of operation.

[0057] 2. Human-machine intention recognition based on multi-sensor fusion:

[0058] Traditional operator control mainly relies on the operation of the handle. The present invention innovatively introduces a multi-sensor system to identify the operator's intention by collecting data such as the operator's joystick thrust, swing, and heart rate, combined with a weighted algorithm and a trajectory prediction algorithm. This method not only takes into account the operator's operating behavior, but also combines physiological reactions (such as heart rate), so as to more accurately judge the operator's intention and emergency response during the inspection process.

[0059] 3. Collision-free flight technology based on convolution operation:

[0060] The present invention uniquely performs convolution operations on the data of millimeter-wave radar and the motion state of the drone to determine the position of the drone in the environment and the possible collision risk in real time. Through this innovative calculation method, the drone can flexibly adjust the flight path in a complex environment, avoid obstacles, and achieve autonomous collision-free flight. This method is more intelligent and accurate than traditional obstacle avoidance technology, and can dynamically adjust the flight path instead of relying solely on pre-set rules.

[0061] 4. Intelligent assisted flight and autonomous takeover:

[0062] Combined with the above inspection personnel's intention recognition results, the drone can autonomously take over the flight operation and perform intelligent assisted flight under appropriate circumstances. In this way, even if the operator makes an operational error in a complex environment, the drone can intervene in time to make corrections and adjustments to ensure the safety and reliability of the inspection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a schematic diagram of the overall structure of the present invention. DETAILED DESCRIPTION

[0064] A preferred embodiment of the present invention is described in detail below with reference to the accompanying drawings.

[0065] The present invention proposes a collision-free intelligent inspection system for substations under human-machine hybrid operation, and the human-machine hybrid operation is realized in the following manner:

[0066] 1. Intent recognition and feedback mechanism:

[0067] The multi-sensor system collects the operator's control signals and physiological data, performs real-time weighted processing, and predicts the flight trajectory of the drone. The system can identify the operator's intentions, such as judging that the operator is in a state of tension and may take emergency actions through the rapid advancement of the control stick and the increase in heart rate.

[0068] The system dynamically adjusts the drone's flight mode based on the intent recognition results, such as increasing the drone's degree of autonomous flight takeover in emergency situations and reducing reliance on manual control by the operator.

[0069] 2. Convolution calculation of millimeter wave radar and motion state:

[0070] The millimeter-wave radar carried by the drone collects surrounding environment data in real time and generates a spatial state matrix. The motion state of the drone is obtained through the motion sensor and forms a motion state matrix.

[0071] The system performs convolution operations on these two matrices to analyze the relationship between the drone's movement trend in space and the surrounding environment and identify potential collision risks. The results of the convolution calculation can not only be used to adjust the current flight path, but also to determine whether to take over autonomously based on the operator's intention recognition results.

[0072] 3. Intelligent inspection process under human-machine collaboration:

[0073] During the actual inspection process, the system always monitors the operator's intentions and environmental conditions in the background. When the operator performs routine operations, the drone mainly flies according to the operator's instructions, but in an emergency, the system automatically evaluates the safety of the current flight path through intention recognition and convolution calculation, and autonomously adjusts the flight strategy to avoid collision.

[0074] This hybrid human-machine control method enables even ordinary inspection personnel to safely operate drones in complex substation environments, greatly reducing the operating threshold while improving inspection efficiency and safety.

[0075] like Figure 1 As shown, the inspection system used in the present invention includes:

[0076] 1. The input layer uses a combination of a multi-sensor system and a drone system to collect the drone handle control stick thrust, handle control stick swing, operator heart rate and millimeter wave radar sampling data, and drone motion status data.

[0077] Select T to represent the thrust of the drone handle joystick, reflecting the current drone acceleration, and take T∈[0,1], where 0 represents no thrust, that is, the drone is hovering or stationary, and 1 represents full thrust, that is, the drone maintains maximum acceleration flight.

[0078] Select S to represent the swing degree of the drone handle control stick, which reflects the operator's panic level. Take S∈[0,1], where 0 represents complete stability and 1 represents severe swing.

[0079] H is selected to represent the operator's heart rate, which reflects the operator's psychological stress and panic level. It is represented by a value normalized relative to a baseline heart rate (such as resting heart rate). H∈[0,1] is taken, where 0 represents a normal heart rate and 1 represents an extremely high heart rate.

[0080] The matrix M is used to represent the millimeter wave radar sampling data. The middle position of the matrix represents the drone itself, and the values ​​of the surrounding positions represent the existence of obstacles:

[0081] M[i][j] = 0 means that the position (i, j) is open and has no obstacles;

[0082] M[i][j]=1 indicates that there is an obstacle at position (i, j).

[0083] For example,

[0084] Among them, the middle position represents the drone itself, and the 1 in the remaining positions represents obstacles.

[0085] The matrix K is used to represent the motion state matrix of the drone. The middle position of the matrix still represents the drone itself, and the values ​​of the surrounding positions represent the speed and direction of the drone in that direction:

[0086] The value of K[i][j] represents the speed of the drone at position (i, j). The larger the value, the faster the speed. The position of K[i][j] in the matrix represents the direction of the drone.

[0087] For example, The UAV motion state matrix K represents the current motion state of the UAV. The larger the value, the faster the UAV is in that direction.

[0088] 2. Processing layer: The control system judgment logic performs weighted judgment on the multi-sensor data and performs trajectory prediction; the collision-free flight technology processing logic performs convolution operations on the drone's spatial state matrix and the drone's motion state matrix to determine the spatial characteristics of the drone's next state.

[0089] (1) Weighting function:

[0090] Set an index C to evaluate the possibility of an impending collision. This index C can be calculated by the following weighting function:

[0091] C=w T ·T+w S ·S+wH H;

[0092] Among them, w T 、w S 、w H is the weight coefficient of the corresponding variable, satisfying w T +w S +w H =1.

[0093] When C approaches 1, it means that the operator may be in a highly nervous state and the drone has a high acceleration, which means that there is a greater possibility of collision. When C is lower, it means that the operation is relatively stable and the risk of collision is low.

[0094] The weighting function and variable design can be adjusted according to actual test results to ensure more accurate prediction of collision risk in various operating scenarios.

[0095] (2) UAV flight trajectory prediction:

[0096] UAV flight trajectory prediction mainly includes the following aspects:

[0097] Adaptive process noise adjustment: According to the UAV's motion state and environmental changes, the process noise covariance matrix Q is dynamically adjusted to make the filter more adaptable to changing environmental and control conditions.

[0098] Dynamic environment weight adjustment: An environment weight matrix W is introduced to incorporate the importance factors of millimeter-wave radar sampling data and handle control stick data into the state estimation process.

[0099] Multimodal data fusion: Combine multiple data sources such as millimeter-wave radar sampling data, joystick data, operator heart rate, etc. to perform multimodal fusion on state prediction to obtain more accurate and robust trajectory prediction.

[0100] The drone state vector x is defined as:

[0101]

[0102] Among them, (x k ,t k ) represents the position of the UAV at time k, which is inferred from the UAV spatial state matrix M and the data of the control system. (v xk ,v yk ) represents the speed of the UAV at time k, which is inferred from the data of the motion state matrix K.

[0103] The process noise covariance matrix Q in the traditional Kalman filter is fixed, but in a complex drone operating environment, the noise level may change with time and environment. We can dynamically adjust Q based on the drone's speed, acceleration, and radar feedback data:

[0104] Q k =Q0+γ·‖K‖;

[0105] Where: Q0 is the initial process noise covariance matrix. γ is the adjustment factor, and ‖K‖ represents the norm of the UAV motion state matrix K at the current moment. When the UAV speed is high or the acceleration changes dramatically, increase Q k , making the filter more sensitive to fast changes.

[0106] In order to achieve dynamic environment weight adjustment, a new weight matrix W is introduced, which performs weighted processing on the millimeter wave radar data and the control input data. The value of the matrix W is dynamically adjusted according to the operator's heart rate, the thrust change of the handle, etc.:

[0107] W k =α·G+β·(1-G);

[0108] Among them, α and β are weight factors, which can be dynamically adjusted according to the operator's heart rate H and the joystick swing S:

[0109]

[0110] G is the observation matrix. After the element values ​​of G are adjusted, the dynamic weight adjustment of the matrix W can be used to enhance the filter's dependence on the observation data under high-risk conditions.

[0111] In the prediction state update, multimodal data such as millimeter-wave radar sampling data, handle control lever data, and operator's physiological signals are combined, and the prediction accuracy is improved through collaborative updates.

[0112] Predict the state of the drone at the next moment based on the current state estimation and state transfer equation:

[0113]

[0114] Among them, λ is the fusion factor, which is used to balance the weights of prediction and multimodal observation.

[0115] Prediction covariance:

[0116] P k+1|k =AP k A T +Q k ;

[0117] Update the prediction status based on dynamic environment weight adjustment and multimodal data fusion:

[0118] L k+1 =P k+1|k W k T (W k P k+1|k W k T +R) -1 ;

[0119]

[0120] Update the covariance:

[0121] P k+1 =(IL k+1 W k ) k+1|k ;

[0122] Among them, L k+1 is the Kalman gain, and R is the observation noise covariance matrix.

[0123] Therefore, each new radar observation value z k+1 When the drone arrives, the above update steps are used to predict the specific location of the drone at the next moment (x k+1 ,y k+1 ) and speed (v xk+1 ,v yk+1 ).

[0124] This method introduces more dynamic adjustment mechanisms than the traditional Kalman filter, making it perform better in complex drone control and obstacle avoidance environments. Adaptive process noise adjustment, dynamic weight observation fusion, and the introduction of multimodal data enable drones to more accurately predict their own trajectories in rapidly changing environments and reduce collision risks. It is particularly suitable for obstacle avoidance and path planning of drones in highly dynamic environments.

[0125] Predicting the flight trajectory of a drone is not only used to predict the future movement trend of the drone, but also provides a basis for the drone's obstacle avoidance and decision-making:

[0126] Optimize the autonomous flight of drones in conjunction with the decision-making layer;

[0127] The predicted results of the flight trajectory can be directly fed back to the execution layer to adjust the real-time flight actions of the drone, such as adjusting speed and direction, and automatically taking over;

[0128] Combined with millimeter-wave radar data, it forms a closed loop and dynamically adjusts the path.

[0129] (3) Convolution operation:

[0130] The convolution operation is used to determine the direction in which the drone is most likely to collide. The convolution operation can use the drone's motion state matrix K as the convolution kernel to operate on the millimeter wave radar sampling matrix M with a step size of 3.

[0131] The convolution operation is defined as:

[0132]

[0133] Among them, D[i,j] represents an element in the convolution result matrix, and the value of the element represents the collision risk at that position.

[0134] The direction represented by the maximum value D[i,j] is the direction in which the current drone is most likely to collide with the millimeter-wave radar sampling data and motion state. By locating the maximum value in the convolution result matrix D, it is possible to know which direction has the highest collision risk when the drone is in motion.

[0135] The convolution result matrix D is used to guide the obstacle avoidance action of the drone. It plays the following specific roles in the subsequent decision-making and execution process:

[0136] The decision layer determines whether to adjust the flight path or speed by evaluating D. This directional risk value will become the key input of the drone's obstacle avoidance algorithm, guiding the drone to avoid obstacles in which direction during flight.

[0137] After the execution layer receives the result of D, the drone will adjust its flight trajectory based on this information. If the system identifies a direction with the highest risk of collision, the drone will actively change its flight direction to avoid collision, such as slowing down or stopping, turning or detouring.

[0138] The result of D can also be combined with indicator C (comprehensively evaluating the possibility of an imminent collision of the drone) to jointly make decisions for the system. For example, when C exceeds a certain preset threshold, such as C warning , and when the collision risk in a certain direction D is the highest, the system can avoid obstacles by automatically taking over the flight.

[0139] 3. Decision-making layer: The decision-making layer is the core judgment and decision-making link of the system. Its main task is to receive and integrate data from the intention recognition algorithm, convolution operation and other perception modules to conduct a comprehensive assessment of the current environment and inspection operations. Specifically, the decision-making layer first evaluates the operator's operational intelligence level by analyzing the operator's control behavior and psychological state (such as heart rate, joystick operation mode). This assessment includes identifying whether the operator is operating in a state of tension or fatigue, as well as the rationality and effectiveness of his operation.

[0140] The judgment criteria in the decision-making layer define different risk levels according to the value of C:

[0141] C≤C safe , the drone is in a safe state and no additional adjustments are required; C safe The safety threshold is set;

[0142] C safe <C≤C warning , the drone enters a warning state, which can trigger operator prompts or enhance automatic assisted flight; C warning The risk thresholds to be set;

[0143] C>C warning , the drone enters a high-risk state, and the system automatically takes over the flight and performs obstacle avoidance operations.

[0144] At the same time, the decision-making layer will judge the degree of autonomous flight that the drone should take in the current environment based on the drone's autonomous obstacle avoidance algorithm and path planning algorithm. If the inspection personnel's operation is evaluated as unstable or there is a risk of collision, the system will prioritize the drone's autonomous flight weight and reduce reliance on manual operation to ensure flight safety.

[0145] The key to the decision layer is also to judge the best, that is, to select the best control strategy according to the weight and priority of various inputs. By comprehensively considering the operator's intention, the autonomous capabilities of the drone, and environmental perception data, the decision layer will generate a set of optimal flight and obstacle avoidance instructions, and decide whether to take the operator's instructions as the main one or to let the drone take over autonomous flight.

[0146] 4. The execution layer is the actual execution unit of the system instructions, which is responsible for converting the instructions generated by the decision layer into specific flight actions and operations. After receiving the results of the decision layer, the execution layer will map these instructions into physical control signals according to the current state and environmental conditions of the drone, and directly act on the flight control system of the drone.

[0147] The work of the execution layer includes not only simple command execution, but also complex flight control, such as dynamically adjusting the speed, heading, and attitude of the drone. Especially in emergency situations, the execution layer needs to quickly respond to the instructions of the decision layer and perform necessary obstacle avoidance actions or autonomous flight takeover to ensure that the drone can safely avoid obstacles or dangerous areas.

[0148] In addition, the execution layer is responsible for real-time monitoring of the status of the drone and environmental changes, and feeding this information back to the decision-making layer for further optimization and adjustment. Through this closed-loop control mechanism, the execution layer ensures the stability and response speed of the system, ensuring that the drone can safely perform inspection tasks in the complex substation environment.

[0149] 5. Output layer: The output layer is the final result display and feedback unit of the entire system. It is not only responsible for presenting the flight results of the drone to the operator or other end users, but also plays a role in safety assurance. Through the output layer, the system can generate inspection reports, flight path records, and obstacle avoidance logs. These data can help operators and managers review and analyze the inspection performance of drones.

[0150] A key task of the output layer is to ensure that no safety accidents occur during the inspection process. It continuously monitors the flight status of the drone by integrating multiple sensor data and decision results, and issues an alarm or immediately initiates emergency measures when potential dangers or anomalies are found. At the same time, the output layer can also record the system's operation logs and flight data, providing an important basis for subsequent analysis, optimization, and maintenance.

[0151] Ultimately, the logic processing and data presentation of the output layer will ensure the safety and reliability of the entire system and achieve the goal of intelligent non-contact inspection of substations. Through the close collaboration of all levels of the system, the output layer can not only ensure the successful completion of the drone inspection mission, but also provide valuable reference data for future inspection missions, further improving the intelligence level of the system.

[0152] The innovation of this solution is that it allows inspectors with poor operating experience to directly use drones to inspect substations. In terms of active safety collision avoidance, it combines the dual input of the operator's psychological condition and the drone's flight status. In this way, even if the operator makes an operating error in a complex environment, the drone can intervene in time to correct and adjust, ensuring the safety and reliability of the inspection process.

[0153] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0154] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A collision-free intelligent inspection system for substations under human-machine hybrid, characterized in that: include: Input layer: After the drone takes off, the input layer obtains data through the multi-sensor system and the drone system, including obtaining the drone's handle control stick thrust, the drone's handle control stick swing, and the operator's heart rate through the multi-sensor, and obtaining millimeter-wave radar sampling data and drone motion state data through the drone system; construct the drone's spatial state matrix M based on the millimeter-wave radar sampling data; and construct the drone's motion state matrix K through the drone's motion state data; Processing layer: The processing layer uses the control system judgment logic to perform weighted summation on the data obtained by the multi-sensor system, determine the possibility of an imminent collision of the drone, and predict the flight trajectory of the drone; Through the collision-free flight technology processing logic, the spatial state matrix of the drone and the motion state matrix of the drone are convolved to obtain the convolution result matrix D; by locating the maximum value in the convolution result matrix, the direction in which the drone has the highest collision risk in the motion state is obtained; Decision-making layer: During the flight of the drone, the decision-making layer analyzes the control behavior and psychological state of the operator, determines the operator's control level, adjusts the weight of the drone's autonomous flight level based on the operator's control level, and generates drone control instructions based on the predicted flight trajectory of the drone and the direction in which the drone has the highest collision risk when in motion; Execution layer: The execution layer converts the instructions generated by the decision layer into unmanned flight actions and controls the drone.

2. The collision-free intelligent inspection system for substations under human-machine hybrid according to claim 1 is characterized in that: The inspection system also includes an output layer; the output layer is used to generate inspection reports, flight path records and obstacle avoidance logs, and to issue an alarm and initiate emergency measures when danger or abnormality is found.

3. The collision-free intelligent inspection system for substations under human-machine hybrid according to claim 1 is characterized in that: In the input layer, the data is obtained through the multi-sensor system and the UAV system, including obtaining the thrust amount of the UAV's handle control stick, the swing degree of the UAV's handle control stick, and the operator's heart rate through the multi-sensor, and obtaining millimeter-wave radar sampling data and UAV motion status data through the UAV system; Construct the UAV spatial state matrix M based on millimeter wave radar sampling data; The UAV motion state matrix K is constructed through the UAV motion state data, including: The amount of thrust T of the drone's handlebar reflects the current acceleration of the drone, and is T∈[0,1], where 0 means no thrust, i.e. the drone is hovering or stationary, and 1 means full thrust, i.e. the drone maintains maximum acceleration in flight; The swing degree S of the drone's handle control stick reflects the operator's panic level, and takes S∈[0,1], where 0 means the control stick is completely stable and 1 means the control stick swings to the maximum extent; The operator's heart rate H reflects the operator's psychological stress and panic level, and is expressed by a value normalized relative to the baseline heart rate. H∈[0,1] is taken, where 0 represents a normal heart rate and 1 represents an extremely high heart rate; The drone spatial state matrix M is used to represent the millimeter-wave radar sampling data. The middle position of the drone spatial state matrix M represents the drone itself, and the values ​​of the surrounding positions represent the existence of obstacles. The element M[i][j]=0 in M ​​indicates that there is no obstacle at the position (i, j); M[i][j]=1 indicates that there is an obstacle at the position (i, j). The drone motion state matrix K is used to represent the drone motion state data. The middle position of the drone motion state matrix K represents the drone itself, and the values ​​of the surrounding positions represent the movement speed of the drone in the corresponding direction; the element K[i][j] in K represents the direction and speed of the drone at position (i, j).

4. The collision-free intelligent inspection system for substations under human-machine hybrid according to claim 3 is characterized in that: The processing layer performs weighted summation of the data acquired by the multi-sensor system through the control system judgment logic to judge the possibility of the UAV being about to collide, specifically including: Set an indicator C to evaluate the probability that the drone is about to collide: C=w T ·T+w S ·S+w H ·H; Among them, w T 、w S 、w H is the weight coefficient, satisfying w T +w s +w H =1; The closer the index C is to 1, the higher the possibility of a UAV collision is, and the closer the index C is to 0, the lower the possibility of a UAV collision is.

5. The collision-free intelligent inspection system for substations under human-machine hybrid according to claim 3 is characterized in that: The prediction of the flight trajectory of the UAV specifically includes: The drone state vector x is defined as: Among them, x k ,y k They represent the horizontal and vertical coordinates of the position of the UAV at time k, respectively, and v xk ,v yk They represent the lateral and longitudinal components of the UAV’s velocity at time k respectively; The process noise covariance matrix Q in the filter is dynamically adjusted according to the speed, acceleration and motion state matrix K of the drone: Q k =Q0+γ·‖K‖; Where: Q0 is the initial process noise covariance matrix, γ is the adjustment factor, ‖K‖ represents the norm of the UAV motion state matrix K at the current moment; Q k represents the process noise covariance matrix at time k; A weight matrix W is introduced to perform weighted processing on the millimeter-wave radar sampling data and the data obtained by the multi-sensor system; the value of the weight matrix W is dynamically adjusted according to the operator's heart rate, the thrust change of the handle, etc.: W k =α·G+β·(1-G); Among them, W k Represents the weight matrix at time k; α and β are weight factors that can be dynamically adjusted according to the operator's heart rate H and the swing degree S of the drone's handle control stick: H max is the maximum heart rate, S max is the maximum swing value of the joystick control rod; G is the observation matrix; Predict the state of the drone at the next moment based on the estimated state at the current moment and the state transfer equation: Among them, λ is the fusion factor, A and B represent the state transfer matrix and the control input matrix respectively. represents the predicted state from time k to time k+1, represents the estimated state at time k, u k represents the control input at time k, z k represents the millimeter wave radar sampling data at time k; Prediction covariance: P k+1|k =AP k From T +Q k ; P k+1|k Represents the state covariance matrix, which gives the error of the prediction result based on time k at time k+1; updates the prediction state according to the dynamic environment weight adjustment and multimodal data fusion: L k+1 =P k+1|k W k T (W k P k+1|k W k T +R) -1 ; Update the covariance: P k+1 =(I-L k+1 W k )P k+1|k Among them, P k+1 represents the state covariance matrix at time k+1, which represents the error of the estimated state at time k+1, L k+1 The Kalman gain represents the time k+1, and R is the observation noise covariance matrix; For the current time k, whenever the millimeter-wave radar sampling data z at the next time k+1 When the drone arrives, the above content is used for prediction, which can infer the specific location of the drone at the next moment (x k+1 ,y k+1 ) and speed (v x(k+1) ,v y(k+1) ).

6. The collision-free intelligent inspection system for substations under human-machine hybrid according to claim 3 is characterized in that: The collision-free flight technology processing logic performs convolution operation on the drone space state matrix and the drone motion state matrix to obtain a convolution result matrix D; and locates the maximum value in the convolution result matrix to obtain the direction in which the drone has the highest collision risk in the motion state, specifically including: The convolution operation is: Among them, D[i,j] represents the element in the convolution result matrix, and the corresponding value represents the collision risk of the drone at position (i,j); By locating the maximum value in the convolution result matrix, the direction in which the UAV has the highest collision risk when in motion is obtained.

Images