Air-ground integrated cooperative patrol scheduling system based on dynamic task scheduling

Abstract

The application discloses an air-ground integrated cooperative patrol scheduling system based on dynamic task scheduling, which comprises a data acquisition module, a credibility evaluation module, an optimization scheduling module, a degradation detection module and a self-recovery control module. The data acquisition module carries out checking and processing on original data to generate a credible perception data packet. The credibility evaluation module fuses each score and task correlation credibility to obtain a comprehensive credibility score. The optimization scheduling module combines a column generation algorithm with a Lagrange algorithm to solve a robust scheduling function, generate an optimal task allocation matrix, corresponding scheduling credibility and a task load vector. The degradation detection module adopts a Bayesian online variable point detection algorithm to recursively analyze a perception performance index and calculate a perception health degree. The self-recovery control module executes an update adjustment strategy based on the overall health level, generates a self-recovery control strategy and executes the same. The optimization scheduling module of the system solves the problem of poor robustness of traditional deterministic scheduling in a complex environment, and can also realize global optimal efficiency.

Description

Technical Field

[0001] This invention relates to the field of air-ground patrol, specifically to an integrated air-ground collaborative patrol scheduling system based on dynamic task scheduling. Background Technology

[0002] The air-ground integrated collaborative patrol and dispatch system based on dynamic task scheduling is an intelligent patrol and control system for collaborative operations of multiple air-based and land-based platforms. The system adopts a modular and layered architecture, with five core functional modules working together to form a complete business chain from data collection, reliability assessment, task scheduling to health monitoring and self-healing regulation. The system can be widely used in various fields that require large-scale, all-weather, and intelligent patrol and control.

[0003] In practical applications, existing technologies, particularly traditional deterministic task scheduling, can only perform simple allocation based on platform location and basic load. This makes it difficult to cope with the uncertainty of credibility caused by disturbances in perception data and data pollution attacks. It is prone to problems such as untrusted platforms executing critical tasks and a sharp drop in the credibility of patrol intelligence. Furthermore, it lacks the ability to efficiently solve large-scale air-ground collaborative tasks under multiple constraints, resulting in delayed scheduling response or only achieving local optima. It cannot balance execution efficiency, energy consumption, and load balancing. At the same time, it lacks the adaptive closed-loop self-healing control capability based on the global perception health level, and cannot dynamically adjust the task entry threshold and patrol range. This can easily lead to platform overload and burnout, and missed defenses in key areas. Moreover, there is no scheduling avalanche suppression mechanism, and a single point of failure can cause a chain reaction of system paralysis, making it difficult to achieve long-term maintenance-free stable operation. Summary of the Invention

[0004] To address the shortcomings of the existing technologies, the present invention aims to provide an integrated air-ground collaborative patrol and dispatch system based on dynamic task scheduling, in order to solve the problems of existing systems being prone to untrusted platforms executing critical tasks, a sharp drop in the credibility of patrol intelligence, and a lack of adaptive closed-loop self-healing control capabilities based on global perception of health levels.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] The present invention relates to an integrated air-ground collaborative patrol and dispatch system based on dynamic task scheduling, the system comprising:

[0007] The data acquisition module acquires raw data collected by various sensors on the air-based platform and the land-based platform; it performs verification processing on the raw data to generate a reliable sensing data package;

[0008] The credibility assessment module maps the credibility-aware data packet into an embedding vector and fuses them into a fused feature vector; based on the fused feature vector, it calculates a spatial consistency score and a temporal continuity score; it performs prediction processing on the fused feature vector to obtain the task-related credibility; and it fuses the scores and the task-related credibility to obtain a comprehensive credibility score.

[0009] The optimization scheduling module constructs a confidence uncertainty set based on the comprehensive confidence score, forming a robust scheduling function in the worst case; the column generation algorithm is combined with the Lagrange algorithm to solve the robust scheduling function, generating the optimal task allocation matrix and the corresponding scheduling confidence and task load vectors;

[0010] The degradation detection module calculates the perception performance index based on the task load vector and scheduling reliability; it uses a Bayesian online change point detection algorithm to perform recursive analysis on the perception performance index and calculate the perception health; it compares the perception health with the alarm threshold and outputs a degradation alarm flag.

[0011] The self-healing control module sums up all perceived health levels to generate an overall health level; based on the overall health level, it executes an update and adjustment strategy, generates a self-healing control strategy, and executes it.

[0012] Furthermore, the verification process in the data acquisition module includes: constructing a challenge vector corresponding to the sensor and inputting the challenge vector into an identity authentication function to generate a response vector; inputting the response vector into a fingerprint generation function to generate an identity fingerprint; constructing a data signature at the sensor end based on the identity fingerprint and the original data, and calculating a verification signature at the receiving end according to the same rules; inputting the data signature and the verification signature into a verification function to generate a data authenticity identifier; inputting the original data whose signature verification has passed into a normalization function to generate standard data; summarizing the standard data, the identity fingerprint, and the data authenticity identifier, and then performing combined processing to generate a trusted perception data packet.

[0013] Furthermore, the prediction processing in the credibility assessment module includes: inputting the fused feature vector into the task prediction function to generate a predicted task vector; and inputting the predicted task vector and the real task vector together into the task association function to generate task association credibility.

[0014] Furthermore, the column generation algorithm and Lagrange algorithm in the optimized scheduling module are as follows: A constraint mechanism is introduced into the robust scheduling function; the constraint mechanism includes allocation constraints, load constraints, energy constraints, time constraints, and reliability constraints; a corresponding Lagrange multiplier is assigned to each task to construct a Lagrange relaxation function; after completing the Lagrange relaxation, an initial feasible solution set is generated, where each element is a scheduling column, and a scheduling column represents a complete patrol execution scheme corresponding to a platform; it is determined whether a task contains a scheduling column; if the task does not contain a scheduling column, the Lagrange multiplier corresponding to the task is increased for iterative processing.

[0015] Further, the iterative process includes: calculating the change value of the robust scheduling function during two rounds of iteration; comparing the change value with a preset stopping threshold, and stopping the iteration when the change value is less than the preset stopping threshold, to generate an optimal task allocation matrix and an optimal joint optimization objective value, wherein the optimal task allocation matrix is ​​composed of optimal task identifiers; calculating the scheduling reliability based on the optimal task identifiers, and summing all the optimal task identifiers to generate a task load vector.

[0016] Furthermore, the recursive analysis in the degradation detection module includes: constructing a Bayesian recursive probability based on the perception performance index; calculating the degradation probability when a change point is detected, performing nonlinear amplification on the degradation probability to obtain the degradation intensity; and inputting the degradation intensity into the perception health function to generate the perception health score.

[0017] Furthermore, the self-healing control module updates and adjusts its strategies as follows: The overall health level is input into the update function to adjust the initial confidence threshold and generate a new confidence threshold; the area of ​​the original patrol task is adjusted according to the overall health level, shrinking proportionally while maintaining the center position, resulting in a new, shrunken patrol area; a cleaning flag is generated based on the degradation alarm flag, and the platform is cleaned and regulated; the perceived health level is input into the scheduling avalanche suppression function to generate a global avalanche risk value, and a scheduling avalanche suppression factor is calculated based on the global avalanche risk value; the new confidence threshold, new patrol area, cleaning flag, and scheduling avalanche suppression factor are summarized to generate a set of self-healing control strategies.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0019] The optimized scheduling module proposed in this invention uses a robust scheduling function constructed with a set of uncertain credibility. This function can ensure the effectiveness and stability of the scheduling scheme even in the worst-case scenario where data is disturbed or some platforms are attacked. This solves the problem of poor robustness of traditional deterministic scheduling in complex environments. At the same time, it comprehensively considers multiple objectives such as task benefits, execution costs, credibility risks and response latency, and introduces five types of hard constraints: allocation, load, energy, time and credibility. This ensures that the scheduling scheme not only meets the actual physical constraints, but also achieves optimal global performance.

[0020] The self-healing control module proposed in this invention dynamically adjusts the credibility threshold based on the overall health level: when the system health declines, it automatically raises the task entry threshold, allowing only high-credibility platforms to execute tasks, thus ensuring the execution quality of core patrol tasks; when the health recovers, it lowers the threshold to fully utilize platform resources. Simultaneously, it employs an adaptive shrinking patrol area strategy, proportionally reducing the patrol range while maintaining the central location under limited system resources, prioritizing full coverage of key areas and maximizing patrol efficiency with limited resources. For degraded platforms, it triggers local self-healing operations such as cache clearing, abnormal data removal, and sensor cleaning, which can quickly restore the platform's sensing capabilities, extend equipment lifespan, and reduce manual maintenance costs. Finally, it innovatively introduces a scheduling avalanche suppression mechanism, dynamically adjusting task rewards through an avalanche suppression factor to avoid system overload caused by the concentrated transfer of tasks from faulty platforms to at least a few normal platforms, significantly improving system stability and continuous operation capabilities. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the air-ground integrated collaborative patrol scenario of the system in this invention;

[0023] Figure 2 This is a schematic diagram of the multi-source heterogeneous data credibility assessment mechanism in the system of this invention;

[0024] Figure 3 This is a visual diagram illustrating the dynamic task scheduling results of the system in this invention. Detailed Implementation

[0025] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.

[0026] Reference Figure 1 As shown, the present invention provides an integrated air-ground collaborative patrol and dispatch system based on dynamic task scheduling. The system includes:

[0027] The data acquisition module acquires raw data collected by sensors on both the air-based and land-based platforms; it verifies the raw data and generates a reliable sensing data package.

[0028] First, raw data generated in real time by various sensors on both the air-based and land-based platforms is collected. The raw data is shown below:

[0029]

[0030] in, For the first The platform One sensor in The raw data at any given moment; For platform indexing; For sensor indexing; It is a time variable;

[0031] Construct a challenge vector corresponding to the sensor, and input the challenge vector into the authentication function to generate a response vector. The authentication function is shown below:

[0032]

[0033] in, The challenge vector; For the response vector; It is a physically unclonable function;

[0034] The response vector is input into the fingerprint generation function to generate an identity fingerprint. The fingerprint generation function is shown below:

[0035]

[0036] in, For hash functions; For identity fingerprint;

[0037] Secondly, based on the identity fingerprint and the original data, a data signature is constructed at the sensor end, as shown below:

[0038]

[0039] in, For the first The platform One sensor in Data signature at any given moment; For hash functions; It is a time variable; This is a byte concatenation operator that can combine multiple data fields into a sequence in order;

[0040] The receiving end calculates the verification signature according to the same rules, and the verification signature is as follows:

[0041]

[0042] in, To verify the signature;

[0043] Input the data signature and verification signature into the verification function to generate a data authenticity identifier. The verification function is shown below:

[0044]

[0045] in, For the first The platform One sensor in Real-time data authenticity verification This indicates that the signature verification has passed and the data is authentic and reliable. This indicates that the signature verification failed, and the data has been tampered with or forged.

[0046] Finally, the original data that has passed signature verification is input into the normalization function to generate standard data. The normalization function is shown below:

[0047]

[0048] in, For the first The platform One sensor in Standard data for time; This is the average of historical data; The standard deviation of historical data;

[0049] Standard data, identity fingerprints, and data authenticity identifiers are aggregated, processed, and then combined to generate a trusted perception data package, as shown below:

[0050]

[0051]

[0052]

[0053]

[0054] in, For the first Trusted data packets of the platform; For the first All standard data from each platform; For the first All identity fingerprints across all platforms; For the first All data authenticity verification on each platform; This represents the total number of sensors; It is the transpose symbol;

[0055] Among them, the airborne platform collects data from the DJI M300RTK drone equipped with an IMU inertial sensor (BNO080), a high-definition video sensor (SONY IMX577), an infrared thermal imaging sensor (FLIRLepton3.5), and a Beidou positioning sensor (NEO-M8N).

[0056] The land-based platform includes an inspection vehicle and a fixed base station. The land-based platform is equipped with a millimeter-wave radar (AR0144), a temperature and humidity sensor (SHT30), a vibration sensor (ADXL345), and a Beidou dual-mode positioning module (ZED-F9P).

[0057] The physically unclonable PUF chip uses the NXPA1000 hardware PUF security chip, which is integrated into various sensor terminals, UAV flight control boards, and ground inspection terminals.

[0058] The edge acquisition and processing unit (local hashing, signature, and normalization operation) uses Rockchip RK3588 industrial-grade edge computing board and is deployed in the drone's airborne cabin and the electrical control box of the ground inspection equipment.

[0059] The narrowband communication transmission module (data uplink) adopts Quectel BC95-B55NB-IoT communication module + Lexing LoRaE32-433M long-range wireless module to stably upload trusted sensing data packets to the next module, ensuring the integrity of heterogeneous data transmission.

[0060] The credibility assessment module maps credibility-aware data packets into embedded vectors and fuses them into a fused feature vector; based on the fused feature vector, it calculates spatial consistency scores and temporal continuity scores; it performs prediction processing on the fused feature vector to obtain task-related credibility; and it fuses the scores and task-related credibility to obtain a comprehensive credibility score.

[0061] First, the trusted sensing data packet is input into the feature mapping function, which maps the standard data in the trusted sensing data packet to a unified feature space to generate an embedding vector. The feature mapping function is shown below:

[0062]

[0063] in, For embedding vectors; For feature mapping functions, the typical feature hashing method first concatenates the sensor standard data, identity fingerprint, and trusted identifier into a binary string, then uses the SHA-256 hash function to generate a fixed-length hash value, and finally truncates and normalizes the hash value into a floating-point embedding vector, using the C++ standard library;

[0064] All embedded vectors are fused to generate a fused feature vector, as shown below:

[0065]

[0066] in, To fuse feature vectors; For feature concatenation; This represents the total number of sensors;

[0067] The embedding vectors between different sensors are input into the cosine similarity calculation function to generate modal similarity. The cosine similarity calculation function is shown below:

[0068]

[0069] in, for and Modal similarity of embedded vectors among sensors and For sensor index, and ; For the first The platform's first One sensor in The embedding vector at time step; For the first The platform's first One sensor in The embedding vector at time step; It is an L2 norm;

[0070] Calculate the average modal similarity across all modalities within the platform to generate a modal consistency score, as shown in the function below:

[0071]

[0072] in, Modal consistency scoring; To meet The modal similarity is summed. The total number of modal pairs;

[0073] Secondly, based on the fused feature vectors, the spatial consistency score between different platforms is calculated, as shown in the function below:

[0074]

[0075] in, Score the spatial consistency. Neighborhood set The number of elements, i.e., the total number of platforms; It is an exponential function; For the first The fusion feature vector of each platform; for The fusion feature vector of each platform; and For platform indexing; For the spatial sensitivity coefficient, referencing similar sensor network research, a suitable value is first selected as the initial value. Then, under different platform spatial distribution scenarios, the initial value is adjusted to make the score accurately reflect the perception correlation, thereby determining the spatial sensitivity coefficient. It is an L2 norm;

[0076] Based on the fused feature vectors, the temporal continuity score across different platforms is calculated using the following function:

[0077]

[0078] in, Scoring for time continuity; The length of the time window; Indexed by timestamp;

[0079] The fused feature vector is input into the task prediction function to generate the predicted task vector. The task prediction function is shown below:

[0080]

[0081] in, For predicting task vectors; The task prediction function typically uses a logistic regression classifier. The input is a fused feature vector, and the output is a predicted task vector. The hidden layer uses a One-vs-Rest strategy.

[0082] The predicted task vector and the actual task vector (issued by the task scheduling system) are input together into the task association function to generate the task association confidence score. The task association function is shown below:

[0083]

[0084] in, Associating credibility with tasks; The vector represents the actual task.

[0085] All data authenticity indicators, modal consistency scores, spatial consistency scores, temporal continuity scores, and task-related credibility scores are aggregated and combined to generate a credibility feature vector, as shown below:

[0086]

[0087] in, This is the credibility feature vector;

[0088] The credibility feature vector is input into the credibility fusion function to generate a comprehensive credibility score. The credibility fusion function is shown below:

[0089]

[0090] in, A comprehensive credibility score is given. For activation functions; For the transpose of the fusion weights; The bias parameters are used to calculate the weights and bias parameters. The weights are calculated by collecting a dataset with labeled true credibility, aiming to minimize the prediction error. The weights are trained by gradient descent, and then scored by domain experts to obtain the initial weights. Finally, the objective weights are calculated based on the information entropy of the data in each dimension, and combined with the subjective weights to obtain the final value.

[0091] Finally, the comprehensive credibility score is input into the data contamination identification function to generate the credibility deviation score. The data contamination identification function is shown below:

[0092]

[0093] in, Confidence deviation;

[0094] Anomaly attack flags are generated by comparing the degree of trust deviation with a preset trust attack threshold, as shown in the function below:

[0095]

[0096] in, This is an abnormal attack indicator, when This indicates that the platform has suffered a data pollution attack. This indicates that the platform is functioning normally; To set a preset trusted attack threshold, a trusted deviation sample set is collected under normal operation and data pollution attack scenarios. The probability density distribution curves of the two types of samples are then plotted, and the optimal Bayesian decision point that minimizes the sum of the false positive rate and the false negative rate is calculated. Finally, the threshold is determined using the ROC curve at an acceptable false positive rate.

[0097] The regional aggregation gateway (data aggregation and preprocessing) adopts the Huawei AR502H industrial IoT gateway and is deployed in the patrol area central base station, field duty room, and indoor and outdoor combined environment;

[0098] The mid-range computing server (feature calculation, credibility assessment) adopts the Inspur NF5280M62U edge server and is deployed in the patrol command center computer room;

[0099] The electromagnetic environment monitoring module (auxiliary anomaly detection) adopts the Spirent GSS6300 electromagnetic interference monitoring module, which is deployed in the command center and key patrol points in the field to monitor electromagnetic interference throughout the area and distinguish between "data pollution attacks" and "environmental interference".

[0100] The scheduling module is optimized by constructing a confidence uncertainty set based on the comprehensive confidence score, forming a robust scheduling function in the worst case. The column generation algorithm is combined with the Lagrange algorithm to solve the robust scheduling function, generating the optimal task allocation matrix and the corresponding scheduling confidence and task load vectors.

[0101] First, a patrol task set is constructed, as shown below:

[0102]

[0103]

[0104] in, For the patrol mission assembly, For the first A patrol mission For patrol mission indexing; This represents the total number of patrol missions; patrol missions include mission locations. Deadline and task priority ;

[0105] Input the comprehensive credibility score into the task reward function to generate the corresponding task reward. The task reward function is shown below:

[0106]

[0107]

[0108] in, For the first The platform executes the first The rewards for each patrol mission; A comprehensive credibility score is given. For the first The platform executes the first The distance traveled for each patrol mission. The platform location was obtained through manual annotation. The mission location was determined by manual observation. It is an L2 norm; The distance attenuation coefficient is first classified according to the urgency of the task, with the larger value assigned to urgent tasks, and the nearby platform is prioritized. Then, the balance of task allocation and total benefit under different distance attenuation coefficients are evaluated by simulation to determine the optimal value.

[0109] The total task revenue is generated by summing all task rewards, as shown in the function below:

[0110]

[0111] in, Total task revenue; Total number of platforms; For task identification, ,when , indicating the first The platform is Time is allocated to the first A patrol mission; when , indicating the first The platform was not assigned any patrol tasks;

[0112] Secondly, based on the credibility deviation and anomaly attack indicators, the platform's overall risk value is calculated:

[0113]

[0114] in, For the first The overall risk value of each platform;

[0115] The overall risk values ​​of all platforms are aggregated to generate the total trusted risk, as shown in the function below:

[0116]

[0117] in, Total credible risk;

[0118] Based on the patrol task, the total task cost is calculated using the function shown below:

[0119]

[0120]

[0121]

[0122]

[0123] in, Total cost of the task; For mobile costs; This is the moving cost coefficient; The distance traveled; Energy consumption costs; Energy cost coefficient; The remaining energy is measured in real time by sensors mounted on the platform itself; For task processing costs; This is the task processing cost coefficient; Task load capacity is determined by the platform's inherent maximum load capacity; each cost coefficient is initially determined by physical cost conversion, and then the coefficients are adjusted by experts in relevant fields to balance the proportion of the three types of costs in the total execution cost;

[0124] The total execution cost is generated by summing the total task costs across all platforms, as shown in the function below:

[0125]

[0126] in, Total execution cost;

[0127] Based on the comprehensive credibility score, a credibility uncertainty set is constructed, as shown below:

[0128]

[0129] in, This is a set with uncertain credibility. The credibility score is a noisy observation. The Wasserstein distance (bulldozer distance) is used to measure the degree of difference between two confidence scores, typically... ; The confidence perturbation radius is determined based on the statistical characteristics of measurement noise according to the confidence score;

[0130] Based on the distance traveled, the response time is calculated as the time required for the platform to move from its current location to the task location, as shown in the function below:

[0131]

[0132] in, For the first The platform moves from its current position to the... Response time for each task location; The distance traveled; The speed is the platform's real-time status data, measured in real time by sensors mounted on the platform itself.

[0133] The response times of all platforms are aggregated to generate an average response latency, as shown in the function below:

[0134]

[0135] in, This represents the average response time.

[0136] Input the total task benefit, total credible risk, total execution cost, and average response latency into the objective function to generate a joint optimization objective, as shown below:

[0137]

[0138] in, To jointly optimize objectives; This is the cost weighting coefficient; This refers to the risk weighting coefficient. This is the time delay weighting coefficient;

[0139] Finally, a robust scheduling function is constructed based on the joint optimization objective.

[0140]

[0141]

[0142] in, Find the worst case in all possible uncertainty sets of confidence, and then maximize the objective function in that worst case to ensure the robustness of the scheduling scheme; Assign a matrix to the initial task;

[0143] By introducing a constraint mechanism into the robust scheduling function and then combining the column generation algorithm and the Lagrange relaxation algorithm, the robust scheduling function is solved to obtain the optimal task scheduling matrix and the optimal joint optimization objective. The constraint mechanism is as follows:

[0144]

[0145]

[0146]

[0147]

[0148]

[0149] in, As an allocation constraint, it means that each task must be executed by one and only one platform; For load constraints, it means that the number of tasks that each platform can execute simultaneously cannot exceed its task load capacity; An energy constraint means that the total energy consumption of each platform in executing all assigned tasks cannot exceed its remaining energy. This is a time constraint, meaning the platform's response time for executing a task cannot exceed the task's deadline. The deadline for the task; As a credibility constraint, The initial credibility threshold is typically set within a certain range to ensure that only high-credibility platforms participate in the task. The scheduling algorithm is then run under different initial credibility thresholds to evaluate task completion rate, total revenue, and risk. Finally, the initial value with the best overall performance is selected as the benchmark for online adaptive adjustment. , ,otherwise, ;

[0150] Column generation algorithm and Lagrange relaxation algorithm:

[0151] Assign a corresponding Lagrange multiplier to each task, with the initial Lagrange multipliers being zero. Relax the assignment constraints corresponding to the tasks into the robust scheduling function to construct the Lagrange relaxation function, as shown below:

[0152]

[0153] in, Let Lagrange relaxation function be used. For the first The multiplier of the task; Assign a matrix to the initial task; The original Brussels joint optimization objective;

[0154] After completing the Lagrange relaxation, an initial feasible solution set is first generated. Each element in the initial feasible solution set is called a scheduling column. A scheduling column represents a complete patrol execution plan corresponding to a platform (for example, platform one executes tasks one and three; platform two executes tasks two and four). It is then determined whether a task contains a scheduling column. If a task does not contain a scheduling column, the Lagrange multiplier corresponding to the task is increased for iterative processing, as shown in the function below:

[0155]

[0156] in, For the first Multipliers in iteration rounds; For the first Multipliers in iteration rounds; To update the step size; for Task identifiers in iteration rounds;

[0157] The robust scheduling function is calculated to account for the changes in value during the two iterations. The function is shown below:

[0158]

[0159] in, The value is a change. For the first Joint optimization objective in iteration rounds; For the first Joint optimization objective in iteration rounds;

[0160] By comparing the change value with a preset stop threshold If a comparison is made, If the iteration stops, the optimal task allocation matrix and the optimal joint optimization objective value are generated, as shown in the function below:

[0161]

[0162]

[0163] in, Assign the optimal task matrix; The optimal joint optimization objective; The optimal task identifier. ,when , indicating the first The platform's optimal patrol task; when , indicating the first The platform was not assigned the optimal patrol task; stop threshold. Typically, based on the convergence characteristics of column generation algorithms, all values ​​within a certain range are initially selected, and then the convergence speed and solution accuracy are tested under different scale tasks and platform combinations. Finally, a threshold that balances real-time requirements and optimality is selected.

[0164] Based on the optimal task identifier, the scheduling reliability is calculated using the following function:

[0165]

[0166] in, Scheduling reliability; A comprehensive credibility score; This is the optimal task identifier;

[0167] The task load vector is generated by summing all the optimal task identifiers, as shown in the function below:

[0168]

[0169] in, For the first Platform task load vector;

[0170] The core scheduling server (main computing unit) is a Dell PowerEdge R750 high-performance computing server (dual Xeon 8375C, 512G memory, 2TB SSD) deployed in the core computer room of the command center.

[0171] The UPS uninterruptible power supply (power supply guarantee) adopts the SANTAK C10K online UPS power supply, which is deployed in the command center computer room and used in conjunction with the dispatch server;

[0172] The platform status acquisition terminal (remaining energy and load capacity acquisition) adopts Holtek HT66F70A industrial measurement and control terminal, which is deployed inside the drone and inspection device;

[0173] The location calculation unit (task and platform location matching) uses the Sinan N70 high-precision positioning calculation module.

[0174] The degradation detection module calculates perception performance indicators based on task load vectors and scheduling reliability; it uses a Bayesian online change point detection algorithm to perform recursive analysis on the perception performance indicators and calculates perception health; it compares the perception health with the alarm threshold and outputs a degradation alarm flag.

[0175] First, based on the task load vector and scheduling reliability, the perceived performance index is calculated, as shown in the function below:

[0176]

[0177] in, For perceived performance indicators; , and As performance weight coefficients, first compare the relative importance of reliability, load rate, and scheduling reliability, then use the overall system performance as the target, search for the optimal weight combination through a genetic algorithm, and finally fine-tune it according to the system design goals; This represents the maximum task load capacity, i.e., the maximum number of tasks that the platform can execute simultaneously.

[0178] Secondly, a Bayesian online change point detection algorithm is used to perform real-time recursive analysis on the sensing performance indicators. By continuously estimating whether a change point (abrupt statistical characteristic) has occurred in the sensing based on the probability distribution of the running length, the real-time recursive analysis is shown below:

[0179] Based on the perceived performance metrics, a Bayesian recursive probability is constructed, as shown below:

[0180]

[0181]

[0182] in, Let be the Bayesian recursive probability, representing the th Each platform from initial time to time All perceived performance metrics, current runtime is The conditional probability; The duration is the time length, representing the length of time the system has been running since the last change point occurred. For the first Each platform from initial time to time All perceived performance metrics;

[0183] When a change point is detected, the degradation probability is calculated using the following function:

[0184]

[0185] in, This represents the probability of degradation.

[0186] The degradation probability is nonlinearly amplified to obtain the degradation intensity, as shown in the function below:

[0187]

[0188] in, Degradation intensity; The degradation amplification factor is determined by fitting the relationship between the degradation probability and the actual degradation degree based on the historical degradation data of the sensor. Different degrees of degradation are then injected, and the degradation amplification factor is adjusted so that the degradation intensity accurately reflects the health status.

[0189] The degradation intensity is input into the perceived health function to generate the perceived health score. The perceived health function is shown below:

[0190]

[0191] in, To perceive health status;

[0192] Finally, by comparing the perceived health status with a preset alarm threshold, a degradation alarm flag is generated, as shown in the function below:

[0193]

[0194] in, For degradation warning signs, This indicates that the platform's perceived health level has reached the low alarm threshold, triggering a degradation alarm. This indicates that the platform perceives the health status to be normal. To set a preset alarm threshold, the mean and standard deviation of the perceived health level during normal operation are calculated. All values ​​between the mean and standard deviation are then used as the initial threshold. Finally, through a sensor degradation fault injection experiment, the threshold is adjusted to optimize the alarm timeliness and false alarm rate, thereby determining the final threshold.

[0195] Among them, the performance monitoring and analysis server (algorithm running unit) adopts Huawei TaiShan2280V5 balanced server, which is deployed in the command center computer room and deployed in the same rack as the scheduling server;

[0196] The equipment status acquisition terminal (hardware condition acquisition) adopts Advantech EKI-1222 remote IO acquisition module, which is deployed in the flight control cabin of airborne UAV and the electrical control box of land-based inspection vehicle to collect hardware condition data such as voltage, current and temperature of sensors, boards and motors to assist in the judgment of sudden changes in sensing performance.

[0197] The audible and visual alarm terminal (on-site alarm output) uses Anbang ABJ-100 industrial audible and visual alarm devices, which are deployed in ground inspection base stations, command center consoles, and key patrol points.

[0198] The self-healing control module summarizes and sums all perceived health levels to generate an overall health level; based on the overall health level, it executes update and adjustment strategies, generates self-healing control strategies, and executes them.

[0199] First, all perceived health scores are summed to generate the overall health level, as shown in the function below:

[0200]

[0201] in, For overall health level, ; Total number of platforms;

[0202] Secondly, the overall health level is input into the update function to adjust the initial confidence threshold and generate a new confidence threshold. The update function is shown below:

[0203]

[0204] in, The new credibility threshold is set, which increases when the overall health level declines, thereby raising the threshold for triggering the task. This is the initial credibility threshold; To adjust the gain, experts in the relevant field first obtain empirical values ​​within a certain range, and then optimize them through experiments;

[0205] The area of ​​the original patrol mission is adjusted based on the overall health level, as shown in the function below:

[0206]

[0207]

[0208] in, This refers to the area of ​​the new patrol zone after the contraction. The contraction coefficient is equal to the system's overall health. The area covered by the original patrol mission;

[0209] Based on the area of ​​the new patrol area after shrinking, the new patrol area is obtained by proportionally reducing it while keeping the center position unchanged, as shown in the function below:

[0210]

[0211] in, This is the new patrol area after the contraction; This is the initial patrol area; The set contains symbols;

[0212] Based on the degradation alarm flags, a cleaning flag is generated, and the platform is then subjected to cleaning and control processing. The function is as follows:

[0213]

[0214] in, For cleaning purposes; This indicates that the platform's perceived health level is low, triggering a cleaning process. The cleaning process includes cache clearing, abnormal data removal, and sensor cleaning. This indicates that the platform perceives the health status as normal and does not trigger cleaning processes.

[0215] The perceived health level is input into the scheduling avalanche suppression function to generate a global avalanche risk value. The scheduling avalanche suppression function is shown below:

[0216]

[0217] in, The global avalanche risk value is the average of the unhealthiness of all platforms.

[0218] Based on the global avalanche risk value, a scheduling avalanche suppression factor is calculated to adjust task rewards. The function is shown below:

[0219]

[0220]

[0221] in, To regulate avalanche suppression factors; To determine the avalanche suppression coefficient, we first simulated an avalanche scenario where a large number of platforms simultaneously degraded offline. Then, we adjusted the avalanche suppression coefficient to effectively suppress the concentrated overload of tasks and maintain scheduling stability, thereby determining the avalanche suppression coefficient. For the benefit of new tasks;

[0222] Finally, the new credibility threshold, new patrol area, cleaning marker, and scheduling avalanche suppression factor are summarized to generate a set of self-healing control strategies, as shown below:

[0223]

[0224] in, A set of self-healing control strategies;

[0225] Among them, the self-healing control all-in-one machine (policy calculation and distribution) usually adopts the Lenovo ECI-430 industrial touch all-in-one machine;

[0226] The all-domain command transmission radio (policy distribution) uses the KELX DP985 digital walkie-talkie and omnidirectional antenna;

[0227] The platform execution controller (local self-healing execution) uses the DJI FC6310 drone flight control motherboard and the Yikong YKC-200 ground controller;

[0228] The Huawei S5720-24P-LI-AC industrial Ethernet switch enables wired interconnection of all servers, gateways, and terminals.

[0229] This invention has many specific applications, and the above are only preferred embodiments of this invention. It should be noted that for those skilled in the art, several improvements can be made without departing from the principle of this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. An integrated air-ground collaborative patrol and dispatch system based on dynamic task scheduling, characterized in that, The system includes: The data acquisition module acquires raw data collected by various sensors on the air-based platform and the land-based platform; it performs verification processing on the raw data to generate a reliable sensing data package; The credibility assessment module maps the credibility-aware data packet into an embedding vector and fuses them into a fused feature vector; based on the fused feature vector, it calculates a spatial consistency score and a temporal continuity score; it performs prediction processing on the fused feature vector to obtain the task-related credibility; and it fuses the scores and the task-related credibility to obtain a comprehensive credibility score. The optimization scheduling module constructs a confidence uncertainty set based on the comprehensive confidence score, forming a robust scheduling function in the worst case; the column generation algorithm is combined with the Lagrange algorithm to solve the robust scheduling function, generating the optimal task allocation matrix and the corresponding scheduling confidence and task load vectors; The degradation detection module calculates the perception performance index based on the task load vector and scheduling reliability; it uses a Bayesian online change point detection algorithm to perform recursive analysis on the perception performance index and calculate the perception health; it compares the perception health with the alarm threshold and outputs a degradation alarm flag. The self-healing control module sums up all perceived health levels to generate an overall health level; based on the overall health level, it executes an update and adjustment strategy, generates a self-healing control strategy, and executes it.

2. The system according to claim 1, characterized in that, The verification process in the data acquisition module includes: constructing a challenge vector corresponding to the sensor and inputting the challenge vector into an identity authentication function to generate a response vector; inputting the response vector into a fingerprint generation function to generate an identity fingerprint; constructing a data signature at the sensor end based on the identity fingerprint and the original data, and calculating a verification signature at the receiving end according to the same rules; inputting the data signature and the verification signature into a verification function to generate a data authenticity identifier; inputting the original data whose signature verification has passed into a normalization function to generate standard data; summarizing the standard data, the identity fingerprint, and the data authenticity identifier, and then performing combined processing to generate a trusted perception data packet.

3. The system according to claim 1, characterized in that, The prediction processing in the credibility assessment module includes: inputting the fused feature vector into the task prediction function to generate a predicted task vector; and inputting the predicted task vector and the real task vector together into the task association function to generate task association credibility.

4. The system according to claim 1, characterized in that, The column generation algorithm and Lagrange algorithm in the optimized scheduling module are as follows: A constraint mechanism is introduced for the robust scheduling function; the constraint mechanism includes allocation constraints, load constraints, energy constraints, time constraints, and reliability constraints; a corresponding Lagrange multiplier is assigned to each task to construct a Lagrange relaxation function; after completing the Lagrange relaxation, an initial feasible solution set is generated, where each element is a scheduling column, and a scheduling column represents a complete patrol execution scheme corresponding to a platform; it is determined whether a task contains a scheduling column; if the task does not contain a scheduling column, the Lagrange multiplier corresponding to the task is increased for iterative processing.

5. The system according to claim 4, characterized in that, The iterative process includes: calculating the change value of the robust scheduling function during two rounds of iteration; comparing the change value with a preset stopping threshold, and stopping the iteration when the change value is less than the preset stopping threshold, to generate an optimal task allocation matrix and an optimal joint optimization objective value, wherein the optimal task allocation matrix is ​​composed of optimal task identifiers; calculating the scheduling reliability based on the optimal task identifiers, and summing all the optimal task identifiers to generate a task load vector.

6. The system according to claim 1, characterized in that, The recursive analysis in the degradation detection module includes: constructing a Bayesian recursive probability based on the perception performance index; calculating the degradation probability when a change point is detected, performing nonlinear amplification on the degradation probability to obtain the degradation intensity; and inputting the degradation intensity into the perception health function to generate the perception health score.

7. The system according to claim 1, characterized in that, The self-healing control module employs the following update and adjustment strategy: the overall health level is input into the update function to adjust the initial confidence threshold and generate a new confidence threshold; the area of ​​the original patrol task is adjusted according to the overall health level, with the center position remaining unchanged and the area reduced proportionally to obtain a new, shrunken patrol area. Based on the degradation alarm flags, a cleaning flag is generated, and the platform is cleaned and controlled. The perceived health status is input into the scheduling avalanche suppression function to generate a global avalanche risk value. Based on the global avalanche risk value, the scheduling avalanche suppression factor is calculated. The new credibility threshold, new patrol area, cleaning markers, and scheduling avalanche suppression factors are summarized to generate a set of self-healing control strategies.

Images