Unmanned aerial vehicle cluster autonomous behavior decision-making method in high-risk police actual combat scene

Abstract

The invention discloses an unmanned aerial vehicle cluster autonomous behavior decision-making method in a high-risk police affair actual combat scene, and belongs to the technical field of police affair actual combat unmanned aerial vehicles. Starting an environment sensing system, collecting environment information through an optical camera in the multi-mode sensor unit, and carrying out noise reduction and correction on original data through a data preprocessing module; s2, the environment information obtained in the S1 is transmitted to a data fusion module, the data fusion module uses an extended Kalman filter to carry out fusion processing on data from different sensors, and a comprehensive environment model is generated; and S3, initializing a positioning and navigation system, compensating the GPS signal by a positioning correction unit based on a visual SLAM technology in combination with a geomagnetic navigation mode, and automatically switching to the geomagnetic navigation mode when the GPS signal is unavailable. According to the invention, more comprehensive environment information can be provided.

Description

Technical Field

[0001] The present invention belongs to the technical field of police combat unmanned aerial vehicles, and in particular, to a method for autonomous behavior decision-making of unmanned aerial vehicle clusters in high-risk police combat scenarios. Background Technique

[0002] In police combat scenarios, the autonomous behavior of unmanned aerial vehicle clusters refers to multiple unmanned aerial vehicles achieving a certain degree of self-decision-making and collaborative work through advanced algorithms and technologies to complete specific task objectives.

[0003] Traditional unmanned aerial vehicles usually rely on a single type of sensor (such as GPS or optical camera), and are prone to inaccurate or lost data in complex environments. Different sensors have their own advantages and disadvantages. For example, optical cameras have poor performance at night or in low visibility conditions, while infrared thermal imagers may fail in high-temperature environments. A single sensor cannot provide comprehensive environmental information, which may lead to path planning errors or target recognition failures. For example, in forest fire monitoring, relying solely on an optical camera may not be able to identify the location of the fire source in the smoke. At the same time, the original data is not effectively denoised and corrected, resulting in poor image quality, affecting the accuracy of subsequent decisions. The data collected by sensors often contains noise and distortion. Unfiltered and uncorrected data will affect the algorithm performance, and poor image quality will increase path planning errors and may even misjudge the location of obstacles. For example, uncorrected images may cause the unmanned aerial vehicle to misidentify an obstacle in front and thus change the flight path. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for autonomous behavior decision-making of unmanned aerial vehicle clusters in high-risk police combat scenarios to solve the problems raised in the background technique.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A method for autonomous behavior decision-making of unmanned aerial vehicle clusters in high-risk police combat scenarios, including the following steps:

[0006] S1; Start the environmental perception system, collect environmental information through the optical camera in the multi-modal sensor unit, and perform noise reduction and correction on the original data through the data preprocessing module;

[0007] S2: Transmit the environmental information obtained in S1 to the data fusion module, and this module uses an extended Kalman filter to fuse the data from different sensors to generate a comprehensive environmental model;

[0008] S3: Initialize the positioning and navigation system, where the positioning correction unit compensates the GPS signal based on visual SLAM technology combined with the geomagnetic navigation mode and automatically switches to the geomagnetic navigation mode when the GPS signal is unavailable;

[0009] S4: The path planning module receives the comprehensive environment model generated by S2 and applies the improved Dijkstra algorithm to calculate the optimal flight path of the UAV swarm;

[0010] S5: Activate the decision support system, and the status evaluation unit collects the status information of the UAV swarm in real time;

[0011] S6: The policy generation module trains the model parameters using the deep reinforcement learning algorithm based on the status information obtained in S5 and formulates the best action plan for the current task;

[0012] S7: When the conflict resolution unit detects potential collision risks, it adjusts the behaviors of each UAV according to the preset rules to avoid collisions with each other;

[0013] S8: Establish a data security management system, and the encryption communication unit applies blockchain technology and simultaneously encrypts the transmitted data through the data encryption module;

[0014] S9: The access control module sets a hierarchical access permission mechanism, manages the access permissions of the external to the UAV swarm system, and realizes the user identity authentication through the identity authentication unit;

[0015] S10: Enable the energy efficiency optimization system, and the energy prediction unit establishes an energy consumption prediction model based on historical flight data;

[0016] S11: The scheduling optimization module dynamically adjusts the flight speed and altitude of the UAVs according to the energy prediction results to achieve the best energy efficiency ratio, where the scheduling optimization module uses the particle swarm optimization (PSO) algorithm to select the optimal combination of working modes.

[0017] Preferably, in S1, the multi-modal sensor unit includes at least three types of sensors, and the data preprocessing module uses wavelet transform to reduce the noise of the original data and corrects the image through the geometric correction algorithm.

[0018] Preferably, in S2, during the fusion process of the extended Kalman filter, the weight assignment strategy W is determined by minimizing the mean square error as:

[0019] E = ∑(x_i - x_′i)∧2, where x_i is the actual observed value and x′ i is the predicted value.

[0020] Preferably, in S3, the adaptive Kalman filter improves the accuracy of position estimation by online adjusting the noise covariance matrices R and Q, where R represents the measurement noise covariance matrix and Q represents the process noise covariance matrix.

[0021] Preferably, in S4, the improved Dijkstra algorithm introduced a dynamic weight adjustment mechanism, and the weight update formula is W_new = W_old + α * ΔW, where α is the learning rate and ΔW is the weight increment calculated according to environmental changes.

[0022] Preferably, in S5, the state prediction unit uses a recurrent neural network to predict the future state of the UAV swarm, and the prediction model is trained by the backpropagation algorithm.

[0023] Preferably, in S6, the objective function of the double Q-learning algorithm is defined as J(Q) = E[Σγ∧tr(s_t, a_t)], where r(s_t, a_t) is the immediate reward, γ is the discount factor, and t is the time step.

[0024] Preferably, in S7, the priority obstacle avoidance strategy P is dynamically adjusted according to the task urgency and current position of the UAV. The priority calculation formula is p_i = βd_i + γt_i, where d_i is the distance from the UAV to the target, t i i is the time limit for completing the target, and β and γ are weight coefficients.

[0025] Preferably, in 58, the blockchain technology is applied to the data block chain structure, and each data block contains a timestamp, transaction data, and the hash value of the previous block.

[0026] Preferably, in S10, the energy consumption prediction model of the long short-term memory network (LSTM) optimizes the model parameters by minimizing the loss function L = Σ(y_i - y′) i )∧2, where y_i is the actual energy consumption and y″i is the model prediction value. In S11, the particle swarm optimization (PSO) algorithm finds the global optimal solution by iteratively updating the position and velocity of the particles. The update formula is v_i(t + 1) = ωv_i(t) + c_7r_1 * (pbest_i - x_(t)) + c_2r_2(gbest - x_(t)), where ω is the inertia weight, c_1 and c_2 are acceleration coefficients, r_1 and r_2 are random numbers, pbest_i is the individual optimal position, and gbest is the global optimal position.

[0027] Compared with the prior art, the technical effects and advantages of the present invention:

[0028] The method for autonomous behavior decision-making of UAV clusters in high-risk police combat scenarios can provide more comprehensive environmental information by combining optical cameras, infrared thermal imagers, and lidar. For example, in nighttime or low visibility conditions, the infrared thermal imager can complement the deficiencies of the optical camera; lidar provides precise distance information to help construct a three-dimensional map. Wavelet transform is used to denoise the original data, and a geometric correction algorithm is applied to correct the images. This not only improves the quality of the data but also enhances the reliability of subsequent processing steps. For example, by eliminating noise and correcting image distortion, the accuracy of path planning and target recognition is ensured.

[0029] When the GPS signal is unavailable, the system automatically switches to the geomagnetic navigation mode and uses the IMU to assist in improving the positioning accuracy. The AKF optimizes the position estimation by online adjusting the noise covariance matrices R and Q, enabling the system to maintain high-precision position tracking in various environments. For example, when flying in a tunnel, the UAV relies on visual SLAM and geomagnetic navigation to maintain a stable flight path and avoid getting lost due to the loss of the GPS signal.

[0030] The state prediction unit uses an RNN to predict future states and trains the model through the BPTT algorithm, enabling the model to better understand time series data. For example, based on the current battery level and flight speed, it predicts whether the UAV can complete the task on time and makes adjustments in advance to ensure the successful execution of the task. Brief Description of the Drawings

[0031] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0032] Figure 1 It is a flowchart of a method for autonomous behavior decision-making of UAV clusters in high-risk police combat scenarios of the present invention. Detailed Embodiments

[0033] In the following description, a large number of specific details are given to provide a more thorough understanding of the present invention. However, it is obvious to those skilled in the art that the present invention can be implemented without one or more of these details. In other examples, to avoid confusion with the present invention, some well-known technical features in the art are not described.

[0034] Unless otherwise defined, the directions such as up, down, left, right, front, back, inside and outside involved in this article are based on the up, down, left, right, front, back, inside and outside directions in the figures shown in the present invention. This is hereby explained together.

[0035] This embodiment provides a method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) cluster in a high-risk police actual combat scenario as shown in Figure 1 the following steps:

[0036] S1: Activate the environmental perception system, collect environmental information through the optical camera, infrared thermal imager and lidar in the multimodal sensor unit, and perform noise reduction and correction on the original data through the data preprocessing module;

[0037] S2: Transmit the environmental information obtained in S1 to the data fusion module. This module uses the Extended Kalman Filter (EKF) to fuse the data from different sensors to generate a comprehensive environmental model. The fusion process adopts a weight assignment strategy based on probability distribution W = {w_1, w_2,..., w_n}, where n is the number of sensors, and w_i represents the weight of the i-th sensor in the fusion result;

[0038] S3: Initialize the positioning and navigation system. The positioning correction unit compensates the GPS signal based on the Visual Simultaneous Localization and Mapping (Visual SLAM) technology combined with the geomagnetic navigation mode, automatically switches to the geomagnetic navigation mode when the GPS signal is unavailable, and uses the Inertial Measurement Unit (IMU) to assist in improving the positioning accuracy. The geomagnetic navigation mode uses the Adaptive Kalman Filter (AKF) for position estimation;

[0039] S4: The path planning module receives the comprehensive environmental model generated by S2 and applies the improved Dijkstra algorithm to calculate the optimal flight path of the UAV swarm. The improved Dijkstra algorithm introduces a dynamic weight adjustment mechanism to adjust the edge weights according to real-time environmental changes;

[0040] S5: Activate the decision support system. The state evaluation unit collects the state information of the UAV swarm in real time, including but not limited to parameters such as position, speed, remaining battery power, etc., and predicts the future state through the state prediction unit;

[0041] S6: The strategy generation module trains the model parameters using the deep reinforcement learning algorithm according to the state information obtained in S5, and formulates the best action plan for the current task. The deep reinforcement learning algorithm uses the Double Q-learning method to reduce the overestimation problem;

[0042] S7: When the conflict resolution unit detects a potential collision risk, it adjusts the behaviors of each UAV according to preset rules to avoid collisions between them. The preset rules include an obstacle avoidance strategy based on priorities P = {p_1, p_2, …, p_m}, where m is the number of UAVs, and p_i represents the priority of the i-th UAV;

[0043] S8: A data security management system is established. The encryption communication unit applies blockchain technology to ensure the security and integrity of data, and at the same time encrypts the transmitted data through a data encryption module;

[0044] S9: The access control module sets a hierarchical access permission mechanism to manage the access permissions of the outside to the UAV swarm system, and realizes user identity authentication through an identity authentication unit;

[0045] S10: An energy efficiency optimization system is enabled. The energy prediction unit establishes an energy consumption prediction model based on historical flight data and continuously updates the model through machine learning algorithms. The energy consumption prediction model is modeled using a long short-term memory network (LSTM);

[0046] S11: The scheduling optimization module dynamically adjusts the flight speed and altitude of the UAVs according to the energy prediction results to achieve the best energy efficiency ratio. The scheduling optimization module uses a particle swarm optimization (PSO) algorithm to select the optimal combination of working modes.

[0047] In this embodiment, in S1, the multi-modal sensor unit includes at least three types of sensors: an optical camera for capturing visible light images; an infrared thermal imager for identifying targets with large temperature differences; a lidar for creating a three-dimensional environmental map. The data preprocessing module uses wavelet transform to denoise the original data and corrects the images through a geometric correction algorithm.

[0048] In this embodiment, in S2, during the fusion process of the extended Kalman filter (EKF), the weight allocation strategy W is determined by minimizing the mean square error as:

[0049] E = ∑(x_i - x′_i)∧2, where x_i is the actual observed value and x′ i is the predicted value.

[0050] In this embodiment, in S3, the adaptive Kalman filter (AKF) improves the accuracy of position estimation by online adjusting the noise covariance matrices R and Q, where R represents the measurement noise covariance matrix and Q represents the process noise covariance matrix.

[0051] In this embodiment, in S4, the improved Dijkstra algorithm introduces a dynamic weight adjustment mechanism, and the weight update formula is W_new = W_old + α *ΔW, where α is the learning rate and ΔW is the weight increment calculated according to the environmental change.

[0052] In this embodiment, in S5, the state prediction unit (34) uses a recurrent neural network (RNN) to predict the future state of the UAV swarm, and the prediction model is trained by the Backpropagation Through Time (BPTT) algorithm.

[0053] In this embodiment, in S6, the objective function of the double Q-learning algorithm is defined as J(Q)=E[Σγ∧tr(s_t, a_t)], where r(s_t, a_t) is the immediate reward, γ is the discount factor, and t is the time step.

[0054] In this embodiment, in S7, the priority-based obstacle avoidance strategy P is dynamically adjusted according to the task urgency and current position of the UAV. The priority calculation formula is p_i = βd_i + γt_i, where d_i is the distance from the UAV to the target, t i i is the time limit for completing the target, and β and γ are weight coefficients.

[0055] In this embodiment, in S8, blockchain technology is applied to the data block chain structure. Each data block contains a timestamp, transaction data, and the hash value of the previous block. The data encryption module (43) encrypts the data using the AES-256 encryption standard.

[0056] In this embodiment, in S10, the energy consumption prediction model of the long short-term memory network (LSTM) optimizes the model parameters by minimizing the loss function L = Σ(y_i - y′) i )∧2, where y_i is the actual energy consumption and y″i is the model prediction value. In S11, the particle swarm optimization (PSO) algorithm iteratively updates the position and velocity of the particles to find the global optimal solution. The update formula is v_i(t + 1)=ωv_i(t)+c_7r_1 * (pbest_i - x_(t))+c_2r_2(gbest - x_(t)), where ω is the inertia weight, c_1 and c_2 are acceleration coefficients, r_1 and r_2 are random numbers, pbest_i is the individual optimal position, and gbest is the global optimal position.

[0057] It should be noted that in this text, relational terms such as "one" and "two" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprising", "including" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the said element.

[0058] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) cluster in a high-risk police combat scenario, characterized in that, Including the following steps: S1； Start the environmental perception system, collect environmental information through the optical camera in the multi-modal sensor unit, and perform noise reduction and correction on the original data through the data preprocessing module; S2: Transmit the environmental information obtained in S1 to the data fusion module, which uses an extended Kalman filter to fuse data from different sensors and generate a comprehensive environmental model; S3: Initialize the positioning and navigation system, where the positioning correction unit compensates the GPS signal based on visual SLAM technology combined with the geomagnetic navigation mode and automatically switches to the geomagnetic navigation mode when the GPS signal is unavailable; S4: The path planning module receives the comprehensive environmental model generated by S2 and applies an improved Dijkstra algorithm to calculate the optimal flight path of the UAV swarm; S5: Activate the decision support system, and the state evaluation unit collects the state information of the UAV swarm in real time; S6: The policy generation module trains the model parameters using a deep reinforcement learning algorithm according to the state information obtained in S5 and formulates the best action plan for the current task; S7: When the conflict resolution unit detects a potential collision risk, it adjusts the behavior of each UAV according to preset rules to avoid collisions with each other; S8: Establish a data security management system, and the encryption communication unit applies blockchain technology and encrypts the transmitted data through the data encryption module; S9: The access control module sets a hierarchical access permission mechanism to manage the access permissions of the external to the UAV swarm system and realizes user authentication through the identity authentication unit; S10: Enable the energy efficiency optimization system, and the energy prediction unit establishes an energy consumption prediction model based on historical flight data; S11: The scheduling optimization module dynamically adjusts the flight speed and altitude of the UAV according to the energy prediction result to achieve the best energy efficiency ratio. The scheduling optimization module uses the particle swarm optimization (PSO) algorithm to select the optimal combination of working modes.

2. The method for autonomous behavior decision-making of an unmanned aerial vehicle cluster in a high-risk police actual combat scenario according to claim 1, wherein: In S1, the multi-modal sensor unit includes at least three types of sensors. The data preprocessing module uses wavelet transform to reduce noise in the original data and corrects the image through a geometric correction algorithm.

3. A method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) cluster in a high-risk police actual combat scenario according to claim 1, characterized in that: In S2, during the fusion process of the extended Kalman filter, the weight allocation strategy W is obtained by minimizing the mean square error as: E = ∑(x_i - x′_i) ∧ is determined by 2, where x_i is the actual observed value, and x′ i is the predicted value.

4. A method for autonomous behavior decision-making of an unmanned aerial vehicle cluster in a high-risk police actual combat scenario according to claim 1, characterized in that: In S3, the adaptive Kalman filter improves the accuracy of position estimation by online adjusting the noise covariance matrices R and Q, where R represents the measurement noise covariance matrix and Q represents the process noise covariance matrix.

5. A method for autonomous behavior decision-making of an unmanned aerial vehicle cluster in a high-risk police combat scenario according to claim 1, characterized in that: In S4, the improved Dijkstra algorithm introduces a dynamic weight adjustment mechanism, and the weight update formula is W_new = W_old + α * ΔW, where α is the learning rate and ΔW is the weight increment calculated according to the environmental change.

6. The method for autonomous behavior decision-making of an unmanned aerial vehicle cluster in a high-risk police combat scenario according to claim 1, wherein: In S5, the state prediction unit uses a recurrent neural network to predict the future state of the UAV swarm, and the prediction model is trained through the backpropagation algorithm.

7. A method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) cluster in a high-risk police actual combat scenario according to claim 1, characterized in that: In S6, the objective function of the double Q-learning algorithm is defined as J(Q)=E[Σγ∧tr(s_t,a_t)], where r(s_t,a_t) is the immediate reward, γ is the discount factor, and t is the time step.

8. A method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) swarm in a high-risk police actual combat scenario according to claim 1, characterized in that: In S7, the obstacle avoidance strategy P of the priority is dynamically adjusted according to the task urgency and the current position of the UAV. The priority calculation formula is p_i = βd_i + γt_i, where d_i is the distance from the UAV to the target, and t i i is the time limit for the target to be completed, and β and γ are weight coefficients.

9. A method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) cluster in a high-risk police combat scenario according to claim 1, characterized in that: In S8, the blockchain technology is applied to the data block chain structure, and each data block contains a time stamp, transaction data, and the hash value of the previous block.

10. A method for autonomous behavior decision-making of an unmanned aerial vehicle (UAV) swarm in a high-risk police actual combat scenario according to claim 1, characterized in that: In S10, the energy consumption prediction model of the long short-term memory network (LSTM) optimizes the model parameters by minimizing the loss function L = Σ(yi - y′) i ) ∧ 2, where yi is the actual energy consumption and y″i is the model prediction value. In S11, the particle swarm optimization (PSO) algorithm searches for the global optimal solution by iteratively updating the positions and velocities of the particles, and the update formula is vi(t + 1) = ω v_ i(t) + c_7r_1 * (pbest_i - x_(t)) + c - 2r_2(g best - x-(t)), where ω is the inertia weight, c_1 and c_2 are acceleration coefficients, r_1 and r_2 are random numbers, pbest_i is the individual optimal position, and gbest is the global optimal position.

Images