Unmanned aerial vehicle intelligent scheduling and management method based on task planning

Abstract

The application discloses a method for intelligent scheduling and management of unmanned aerial vehicles based on task planning, relates to the technical field of dynamic path planning, and comprises the following steps: collecting flight data and performing pretreatment; based on the pretreated data, a health state of the unmanned aerial vehicle is analyzed by constructing a smart patrol dynamic convolution network model, and a health state result is generated; the generated health state result is utilized, combined with environment perception data, to perform dynamic path planning through a recurrent neural network model; and according to the dynamic path planning, a task is formulated and a plan is executed by an intelligent scheduling algorithm; and the application performs hierarchical feature extraction and fusion on the pretreated flight data through the smart patrol dynamic convolution network model, generates a health state feature vector, and realizes accurate health state evaluation and abnormality detection in combination with a health state value and One-Class SVM.

Description

Technical Field

[0001] This invention relates to the field of dynamic path planning technology, and in particular to a method for intelligent scheduling and management of unmanned aerial vehicles (UAVs) based on task planning. Background Technology

[0002] With the rapid development of drone technology, its application scope has expanded from traditional aerial photography and entertainment to multiple fields such as logistics transportation, agricultural monitoring, environmental protection, and emergency rescue. However, in multi-tasking and highly complex application environments, the intelligent scheduling and management of drones faces many challenges. Existing technologies typically rely on static path planning algorithms and simple task allocation mechanisms, which lack the ability to respond in real time to dynamic environmental changes. In addition, for drone health status monitoring, traditional methods are mostly based on rule thresholds or simple statistical analysis, which are difficult to comprehensively capture the health status characteristics of complex systems, resulting in low reliability of task execution under abnormal conditions. As the complexity of task planning and the uncertainty of the drone operating environment increase, the limitations of existing technologies become increasingly apparent.

[0003] To address the aforementioned issues, existing technologies suffer from the following shortcomings: First, they lack intelligent analysis of the health status of UAVs, making it impossible to identify potential faults or abnormal states in a timely manner, which may lead to mission failure or safety hazards. Second, path planning methods do not adequately utilize environmental perception data and fail to fully consider the real-time changes of dynamic interference factors (such as obstacles and weather conditions). Third, UAV mission scheduling and execution strategies lack global optimization capabilities and fail to effectively integrate the health status of UAVs, mission requirements, and environmental conditions. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a task planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) to solve the health status detection problem.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for intelligent scheduling and management of unmanned aerial vehicles (UAVs) based on task planning, which includes: collecting flight data and preprocessing it; constructing a smart patrol dynamic convolutional network model based on the preprocessed data to analyze the health status of the UAV and generate health status results; using the generated health status results and combined with environmental perception data, performing dynamic path planning through a recurrent neural network model; and formulating tasks and executing plans based on the dynamic path planning using an intelligent scheduling algorithm.

[0008] As a preferred embodiment of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) described in this invention, the flight data includes GPS location data, speed data, acceleration data, battery power data, and environmental perception data.

[0009] As a preferred embodiment of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) described in this invention, the preprocessing includes data cleaning, removing outliers and noise, and normalizing the data.

[0010] As a preferred embodiment of the task planning-based UAV intelligent scheduling and management method of the present invention, the specific steps for constructing the intelligent patrol dynamic convolutional network model are as follows:

[0011] The input layer of the intelligent patrol dynamic convolutional network model is defined as preprocessed data;

[0012] The preprocessed data is divided into training set, validation set and test set;

[0013] Extracting local features from data using dynamic convolutional layers;

[0014] Add a pooling layer to extract the main features of the data;

[0015] Local features and main features are merged into a one-dimensional feature vector, and higher-order feature relationships are learned through a fully connected layer to complete feature fusion and generate a health status feature vector.

[0016] Define the output layer of the intelligent patrol dynamic convolutional network model, which calculates the health status value based on the health status feature vector, as expressed by:

[0017]

[0018] Where F(X) is the health status value, d is the dimension of the health status feature vector, and x i is the actual value of the i-th feature, μ is the average of all features, i is the index of the feature, and ∏ is the product sign. It is a factor that adjusts the scale of the results;

[0019] Set a judgment threshold;

[0020] If F(X)≤T, then the health status is determined to be abnormal;

[0021] If F(X) > T, then the health status is determined to be normal;

[0022] One-Class SVM is introduced to detect the health status feature vector and output the health status result;

[0023] The intelligent patrol dynamic convolutional network model is trained using the training set to optimize the classification task;

[0024] The performance of the intelligent patrol dynamic convolutional network model was evaluated using a validation set, and hyperparameters were optimized.

[0025] The Intelligent Patrol Dynamic Convolutional Network model and One-Class SVM were comprehensively evaluated using the test set.

[0026] As a preferred embodiment of the UAV intelligent scheduling and management method based on task planning described in this invention, the specific steps of introducing One-Class SVM to detect health status feature vectors and output health status results are as follows:

[0027] Based on the generated health status values ​​and the set judgment threshold, normal state feature vector samples are selected as training data for One-Class SVM;

[0028] Using the RBF radial basis kernel as the kernel function of the One-Class SVM, the One-Class SVM is trained to learn the distribution of normal state feature vector samples;

[0029] During the testing phase, the health status feature vector is input into the trained One-Class SVM;

[0030] One-Class SVM determines whether a test sample belongs to a normal state based on the distribution of normal state feature vector samples and outputs a health status result.

[0031] As a preferred embodiment of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) according to the present invention, the specific steps of using the generated health status results, combined with environmental perception data, to perform dynamic path planning through a recurrent neural network model are as follows:

[0032] Based on the flight mission, environmental perception data and health status values ​​are used as inputs for path planning;

[0033] The priority of path adjustment is calculated using the following expression:

[0034]

[0035] Where A(P) is the route adjustment priority, P is the evaluation target of the current route planning, e is the total number of environmental factors, j is the index of the environmental factor, and E j D is the urgency weight of the j-th environmental factor. j is the distance between the drone and the j-th environmental factor, k is the distance adjustment factor, and m is the result adjustment factor;

[0036] If the drone's health status is abnormal, it will be directly assigned the priority of returning to base.

[0037] If the drone is in good health, prioritize avoiding high-risk areas;

[0038] The time series data of environmental perception data is processed using a recurrent neural network model, and combined with health status values, to generate a dynamically optimized path node sequence as the result of dynamic path planning.

[0039] The dynamic path planning results are sent as navigation commands to the UAV navigation center;

[0040] The UAV navigation center receives navigation commands and parses them to generate flight parameters;

[0041] The drone executes navigation commands and adjusts flight parameters in real time to avoid obstacles;

[0042] If the environmental conditions in the environmental perception data change, the path is recalculated and the navigation instructions are updated.

[0043] As a preferred embodiment of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) described in this invention, the specific steps of processing time-series environmental perception data using a recurrent neural network model and combining it with health status values ​​to generate a dynamically optimized path node sequence as the dynamic path planning result are as follows:

[0044] Define the first layer of a recurrent neural network model, input environmental perception data and health status values, and extract time series features;

[0045] Define the intermediate layer of the recurrent neural network model, and introduce an LSTM (Long Short-Term Memory) network to capture long-term dependencies in the time series and extract global features;

[0046] Define the last layer of the recurrent neural network model, which combines the extracted time series features and global features to output a dynamically optimized path node sequence;

[0047] Define a loss function to avoid high-risk areas, optimize path length, and constrain health status;

[0048] The dynamically optimized path node sequence generated by the recurrent neural network model is used as the result of dynamic path planning.

[0049] As a preferred embodiment of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) described in this invention, the specific steps of formulating a task and executing a plan based on dynamic path planning and intelligent scheduling algorithm are as follows:

[0050] Input the flight mission, dynamic path planning results, and UAV health status, and initialize the intelligent scheduling algorithm;

[0051] Through iterative optimization, a task allocation scheme is generated, and a task allocation plan is output.

[0052] Send the mission allocation plan to the UAV navigation center and begin executing the flight mission;

[0053] The flight status during flight missions can be monitored in real time through drone communication.

[0054] After completing its mission, the drone uploads the mission results to the central processing center and updates the mission status.

[0055] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles as described in the first aspect of the present invention.

[0056] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles as described in the first aspect of the present invention.

[0057] The beneficial effects of this invention are as follows: This invention uses a smart patrol dynamic convolutional network model to perform hierarchical feature extraction and fusion on preprocessed flight data to generate a health status feature vector. Combined with health status values ​​and a One-Class SVM, it achieves accurate health status assessment and anomaly detection. This model employs dynamic convolutional layers to extract local features, pooling layers to extract main features, and fully connected layers to learn higher-order feature relationships, ensuring the comprehensiveness and accuracy of health status analysis. By quantifying health status values ​​through formulas and combining them with a One-Class SVM to efficiently handle scenarios with insufficient abnormal samples, the reliability of UAV health monitoring is significantly improved. The smart patrol dynamic convolutional network model provides accurate health status data for dynamic path planning and intelligent scheduling, effectively ensuring the safety and stability of mission execution. Attached Figure Description

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is a flowchart of the UAV intelligent scheduling and management method based on task planning in Example 1;

[0060] Figure 2 This is a schematic diagram illustrating the flight mission defined in Example 1. Detailed Implementation

[0061] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0062] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0063] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0064] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a method for intelligent scheduling and management of unmanned aerial vehicles (UAVs) based on task planning, including the following steps:

[0065] S1. Collect flight data and perform preprocessing.

[0066] Furthermore, flight data includes GPS location data, speed data, acceleration data, battery power data, and environmental perception data;

[0067] GPS location data is collected by the GPS receiver equipped on the drone to receive satellite signals and determine the drone's geographical location, including longitude, latitude, and altitude.

[0068] Speed ​​data is collected via an airspeed meter to measure the drone's speed relative to the air. It can be calculated from GPS data and provides ground speed (the drone's speed relative to the ground).

[0069] Acceleration data is collected through accelerometers, which can measure the acceleration of the drone on three axes (X, Y, and Z axes). This data can be used to calculate the dynamic changes of the drone.

[0070] Battery power data is collected through the battery management center, which monitors the battery's voltage, current, and temperature to ensure battery safety and provide power information.

[0071] Environmental perception data includes weather conditions, obstacle information, and mission area information;

[0072] Weather conditions are obtained through meteorological sensors, which collect data on temperature, wind speed and direction, and humidity.

[0073] Temperature data is used to determine whether extreme temperatures affect drone performance;

[0074] Wind speed and direction data are used to assess flight stability and the feasibility of path planning;

[0075] Humidity data is used to determine the impact of humidity on flight equipment (such as rain and fog);

[0076] Obstacle information is obtained in real time through onboard sensors (such as lidar and ultrasonic sensors) to determine the position and size of obstacles;

[0077] The distance and angle of the obstacle relative to the drone;

[0078] Obstacle size helps determine avoidance priority;

[0079] Mission area information is obtained by combining mission planning objectives with terrain features and no-fly zones acquired through topographic maps, predefined mission areas, or visual sensors.

[0080] Topographical features such as undulating terrain and steep slopes;

[0081] No-fly zones include areas restricted by law or high-risk areas;

[0082] All collected data is transmitted in real time to the central processing center via the drone's wireless communication for subsequent input.

[0083] S1.1 Preprocessing, specifically,

[0084] Preprocessing includes data cleaning, removing outliers and noise, and normalizing the data;

[0085] Specifically, Kalman filtering is used for data cleaning to remove outliers and noise, and the data is normalized.

[0086] Set the initial state estimate and the initial error covariance matrix;

[0087] Based on dynamic prediction of the next state;

[0088] Calculate the error covariance of the predicted state;

[0089] Receive new measurement values;

[0090] Calculate the Kalman gain, which is a weighted tradeoff between the predicted state and the measured value;

[0091] Update the state estimate by combining the predicted state and the measured value;

[0092] Update the error covariance matrix;

[0093] Identify and process outliers that deviate significantly from model predictions or historical data. These outliers may be caused by sensor malfunctions, external interference, or data transmission errors.

[0094] For each feature (such as velocity, acceleration, etc.), determine its minimum and maximum values ​​across the entire dataset;

[0095] For each data point, normalization is performed using a formula, expressed as:

[0096]

[0097] Where, x norm This is the normalized data, where x is the original data. max It is the maximum value of the data, x min It is the minimum value of the data;

[0098] When new data points are added, normalization is performed using the same minimum and maximum values ​​to ensure data consistency.

[0099] It should be noted that by collecting GPS location, speed, acceleration, battery level, and environmental perception data, comprehensive information on the UAV's operational status and environment is obtained, ensuring rich data dimensions and coverage of key flight factors. Through preprocessing steps, Kalman filtering is used to remove noise and outliers, significantly improving the accuracy and reliability of the data. At the same time, normalization processing unifies the data scale, enhancing the comparability of data with different characteristics and the model's adaptability. This process provides high-quality input data for subsequent health status analysis, path planning, and task scheduling, ensuring the system's stability and the accuracy of intelligent decision-making.

[0100] S2. Based on the preprocessed data, construct a smart patrol dynamic convolutional network model to analyze the health status of the drone and generate health status results.

[0101] Furthermore, the input layer of the intelligent patrol dynamic convolutional network model is defined as preprocessed data;

[0102] The preprocessed data is divided into training set, validation set and test set (70% training set, 15% validation set and 15% test set);

[0103] The training set is used to train the intelligent patrol dynamic convolutional network model;

[0104] The validation set is used for model performance evaluation and hyperparameter tuning;

[0105] The test set is used for a comprehensive evaluation of the final model;

[0106] Extracting local features from data using dynamic convolutional layers;

[0107] Adding pooling layers (such as max pooling or average pooling) can reduce dimensionality, extract the main features of the data, and reduce unnecessary noise.

[0108] The purpose of pooling layers is to reduce the dimensionality of features while retaining the most important information;

[0109] Local features and main features are merged into a one-dimensional feature vector, and higher-order feature relationships are further learned through a fully connected layer to complete feature fusion and generate a health status feature vector.

[0110] Define the output layer of the intelligent patrol dynamic convolutional network model. Based on the health state feature vector, calculate the health state value (normal or abnormal state) to quantify the health state of the UAV. The expression is:

[0111]

[0112] Where F(X) is the health status value, an abnormality score representing the degree of abnormality of data point X; the larger the value, the greater the deviation of the data point from the normal pattern. d is the dimension of the health status feature vector, i.e., the number of features. i is the actual value of the i-th feature. In drone health management applications, this could be a sample of a sensor reading (such as speed, acceleration, etc.), and μ is the average value of all features (x...). i The mean of the values ​​taken in the normal sample is used as a statistic to measure the central tendency of the dataset. i is the feature index used to iterate through each feature in the health status feature vector. ∏ is the product sign, representing the product of the squared deviations of all features. It is a factor that adjusts the scale of the results and is used to standardize the range of health status values;

[0113] This formula quantifies health status based on the product of squared deviations of features from the mean. By combining the degree of deviation of each feature, it comprehensively assesses the degree of abnormality, avoiding the limitations of single-feature abnormalities. At the same time, the product operation is used to enhance the sensitivity to abnormalities in the correlation between multidimensional features. The introduction of a normalization factor ensures that the result range is standardized, making it easier to identify abnormalities. This improves the comprehensiveness, accuracy and robustness of health status assessment.

[0114] Set the judgment threshold, T = 0.75 (the specific threshold can be set according to the actual situation);

[0115] If F(X)≤T, the health status value is too low, indicating that the sample deviates from the normal state, and the health status is judged to be abnormal.

[0116] If F(X)>T, the health status value is high, indicating that the sample is within the normal range. Therefore, the health status is determined to be normal, and its health status feature vector is input into the trained One-Class SVM for further confirmation.

[0117] For example, if a sample's health status value F(X) = 0.6, which means 0.6 ≤ 0.75, then the sample's health status is determined to be abnormal.

[0118] Suppose that the health status value of another sample is F(X) = 1.2, which means that 1.2 ≤ 0.75. Then the health status of the sample is determined to be normal, and its health status feature vector is input into One-Class SVM for further confirmation.

[0119] One-Class SVM is introduced to detect the health status feature vector and output the health status result;

[0120] One-Class SVM, as an anomaly detector, learns the distribution of normal samples to detect anomalous samples that deviate from the normal distribution. It complements the classification capabilities of dynamic convolutional networks and is used to further improve the accuracy of health status detection.

[0121] One-Class SVM is an unsupervised or one-class classification algorithm based on Support Vector Machine (SVM). It is mainly used for anomaly detection or novelty detection scenarios. Unlike traditional SVM, One-Class SVM focuses on the data distribution of a single class, attempting to identify which data points belong to that class (normal class) and which data points are outliers (abnormal class).

[0122] The intelligent patrol dynamic convolutional network model is trained using the training set to optimize the classification task (normal / abnormal state), including the dynamic convolutional network and the One-Class SVM part. The changes in the loss function are monitored to ensure model convergence.

[0123] The performance of the intelligent patrol dynamic convolutional network model is evaluated using a validation set, and hyperparameters (such as convolutional kernel size, learning rate, SVM kernel function, etc.) are optimized.

[0124] Regularly adjusting the parameters of the intelligent patrol dynamic convolutional network model (such as learning rate, batch size, etc.) can improve the model's classification accuracy and anomaly detection capabilities. This step helps prevent overfitting and find the optimal configuration of the intelligent patrol dynamic convolutional network model.

[0125] The optimal combination of hyperparameters is achieved through grid search and random search.

[0126] The tuning parameters for dynamic convolutional networks include kernel size, number of layers, learning rate, and Dropout ratio.

[0127] One-Class SVM tuning parameters include kernel function type (such as RBF kernel), regularization parameters, etc.

[0128] A comprehensive evaluation of the Intelligent Patrol Dynamic Convolutional Network model and One-Class SVM was conducted using a test set to examine their performance on unseen data. This step can provide a true estimate of the model’s generalization ability.

[0129] Record the classification performance of the intelligent patrol dynamic convolutional network model on the test set, including key indicators such as accuracy, recall, and F1 score;

[0130] Dynamic convolutional networks are used for health status classification, outputting normal or abnormal status.

[0131] The results of the health status assessment formula are used to further assist in determining the degree of abnormality;

[0132] One-Class SVM is used to detect abnormal samples with abnormal distributions, thereby improving the accuracy of anomaly detection.

[0133] S2.1. Introduce One-Class SVM to detect health status feature vectors and output health status results. Specifically,

[0134] Based on the generated health status values ​​and the set judgment threshold, normal status feature vector samples are selected as training data for One-Class SVM; it is ensured that the training data does not contain abnormal samples so that One-Class SVM can accurately learn the distribution characteristics of "normal" data;

[0135] Using the RBF radial basis function as the kernel function of One-Class SVM, the One-Class SVM is trained to learn the distribution of normal state feature vector samples. The RBF kernel function is suitable for high-dimensional data with non-linear distribution and can effectively capture complex feature patterns.

[0136] During the testing phase, the health status feature vector is input into the trained One-Class SVM. The One-Class SVM determines whether the test sample belongs to the normal state based on the distribution of the normal state feature vector samples and outputs a binary classification health status result (normal or abnormal).

[0137] It should be noted that by combining the intelligent patrol dynamic convolutional network model with One-Class SVM, accurate classification and anomaly detection of UAV health status are achieved. The dynamic convolutional network layer extracts local and high-order features to generate health status feature vectors, and quantifies the degree of anomaly through the health status value formula. The One-Class SVM further learns the normal sample distribution, effectively detecting abnormal samples that deviate from the normal distribution, making up for the limitation of insufficient abnormal samples. This significantly improves the accuracy and robustness of health status assessment, provides a reliable basis for UAV dynamic path planning and task scheduling, and enhances the safety and stability of task execution.

[0138] S3. Using the generated health status results and combined with environmental perception data, dynamic path planning is performed through a recurrent neural network model.

[0139] Furthermore, based on the flight mission, the weather conditions, obstacle information, and mission area information from environmental perception data, as well as the normal and abnormal states of health status values, are used as inputs for path planning.

[0140] The priority of path adjustment is calculated using the following expression:

[0141]

[0142] Where A(P) is the route adjustment priority, used to determine the urgency of route optimization; P is the evaluation objective of the current route planning; e is the total number of environmental factors; j is the index of the environmental factor; and E... j D is the urgency weight of the j-th environmental factor. j is the distance between the UAV and the j-th environmental factor, k is the distance adjustment factor, and m is the result adjustment factor, which is used to adjust the nonlinearity of the priority calculation result;

[0143] This formula comprehensively evaluates the priority of path adjustment by weighting the emergency weight of environmental factors with the distance of the UAV. The weights reflect the importance of the risks, and the distance factor enhances the sensitivity to high-risk factors at close range. The introduction of a nonlinear adjustment factor and the improvement of the formula's flexibility allow it to adapt to different mission requirements, thereby improving the overall accuracy of path optimization and ensuring the safety and efficiency of UAV flight.

[0144] Set the preferred adjustment threshold;

[0145] If A(P)≤A, the path adjustment has a high priority and requires immediate or instantaneous path adjustment (such as avoiding obstacles or returning to base).

[0146] If A(P) > A, the path adjustment priority is low, and the current path can be continued to execute the task;

[0147] For example (the specific setting depends on the actual situation), if A = 0.3, it is a high-risk mission, and the path needs to be adjusted immediately to ensure the safety of the drone (such as avoiding obstacles or returning to base);

[0148] If A = 0.5, it is considered a medium-risk task, requiring timely adjustment of the path, but the sensitivity to path adjustment should be appropriately reduced.

[0149] If A = 0.7, then it is a low-risk task, and we can continue on the current path to execute the task;

[0150] If the health status is abnormal or there are high-risk factors in the environment (such as obstacles being too close or wind speeds exceeding the limit), path adjustment will be triggered first.

[0151] If the drone's health status is abnormal, the return-to-base path will be given the highest priority (stop the current mission, plan the shortest time and lowest risk return-to-base path, avoid high-risk areas (such as obstacles, severe weather, no-fly zones, etc.), and ensure that the path meets the constraints of battery power and safety margin).

[0152] If the health status is normal, the path priority is calculated based on environmental perception data, prioritizing the avoidance of high-risk areas (such as environmental obstacles, such as buildings, utility poles, trees, etc., which may pose a threat to the drone's flight; areas with severe weather, such as strong winds, rain, sandstorms, etc., which may affect the stability and navigation accuracy of the drone; areas with weak wireless signals, where the communication signal between the drone and the navigation center is weak or even interrupted, which may lead to mission loss of control or navigation failure; areas with malicious interference, where there is electronic interference (such as jamming signals) or GPS signal failure, which may cause the drone to be unable to navigate normally; and no-fly zones, areas where flying is explicitly prohibited by regulations, such as near airports, over military bases, etc.).

[0153] The environmental obstacle determination criterion is the distance D between the obstacle and the drone. j Smaller (i.e., danger distance), and emergency weight E j High;

[0154] The criteria for identifying severe weather areas are weather data detected by the environmental sensing system, such as wind speed exceeding a set threshold (e.g., 10 m / s), precipitation or humidity reaching dangerous levels, etc.

[0155] The criteria for determining a weak wireless signal area is that the communication signal strength is lower than a set threshold (e.g., RSSI < -90dBm).

[0156] The criteria for determining a severe interference area are: GPS signal loss or the perceived interference signal strength exceeds a set threshold (e.g., GPS signal loss threshold C / N0 < 25dB-Hz or no signal for 5 consecutive seconds, interference signal strength threshold J / S > 15dB or interference signal power > -100dBm).

[0157] The criteria for determining no-fly zones are to combine electronic map data and no-fly zone coordinates to determine whether a drone is approaching a no-fly zone;

[0158] The time series data of environmental perception data is processed using a recurrent neural network model, and combined with health status values, to generate a dynamically optimized path node sequence as the result of dynamic path planning.

[0159] The dynamic path planning results are sent as navigation commands to the UAV navigation center;

[0160] The UAV navigation center receives navigation commands and parses them to generate flight parameters;

[0161] The drone executes navigation commands and adjusts flight parameters in real time to avoid obstacles or high-risk areas;

[0162] Static obstacles include buildings, tall buildings, bridges, trees, utility poles, streetlights, and terrain features (such as mountains and cliffs);

[0163] Dynamic obstacles include other drones or aircraft, ground vehicles, and birds or other flying or moving animals;

[0164] High-risk areas include areas with severe weather, areas with GPS signal failure, areas with electronic interference, no-fly zones, and areas with environmental obstacles;

[0165] If environmental conditions in the environmental perception data change, the path planning system will recalculate the path and update the navigation instructions by combining the health status value and the new environmental data. The navigation instructions will also be updated through the UAV navigation center. The UAV will adjust its flight parameters in real time, including speed, heading and altitude, based on the updated path to ensure the safety and efficiency of the mission.

[0166] S3.1. A recurrent neural network model is used to process the time-series data of environmental perception data, and combined with health status values, to generate a dynamically optimized path node sequence as the result of dynamic path planning. Specifically,

[0167] Define the first layer of a recurrent neural network model, input environmental perception data and health status values, and extract time series features to analyze short-term trends;

[0168] Define the intermediate layer of the recurrent neural network model, introduce an LSTM long short-term memory network to capture long-term dependencies in the time series (e.g., identify the trend of wind speed change or the dynamic pattern of obstacle movement), predict future environmental change trends (such as dynamic obstacle movement, the possibility of health deterioration), and extract global features.

[0169] Define the last layer of the recurrent neural network model, which combines the extracted time-series features and global features to output a dynamically optimized path node sequence (including Node... i (GPS coordinates, flight altitude, and flight direction);

[0170] Define a loss function to avoid high-risk areas, optimize path length, and constrain health status;

[0171] Avoiding high-risk areas means keeping path nodes as far away from obstacles or dangerous areas as possible;

[0172] Optimizing path length can reduce the distance between path nodes (saving energy);

[0173] Constraining health status ensures that path planning conforms to the drone's health status (e.g., prioritizing return in case of anomalies);

[0174] By processing with a recurrent neural network model, a set of dynamically optimized path node sequences is generated as the result of dynamic path planning.

[0175] Specifically, this includes: GPS coordinates of path nodes, flight altitude, and flight direction;

[0176] The GPS coordinates of its path nodes are used to predict the geographical location (latitude and longitude coordinates) that the drone needs to reach next.

[0177] The flight altitude is dynamically adjusted based on the location of obstacles and the health status of the drone.

[0178] Flight direction determines the flight direction between each path node, avoiding unnecessary detours;

[0179] It should be noted that by combining health status values ​​with environmental perception data (such as weather, obstacles, and mission area information), a recurrent neural network model is used to dynamically plan the UAV's flight path, achieving optimized generation of path nodes; LSTM captures time-series features to identify environmental change trends, ensuring the flexibility and accuracy of path planning; high-risk areas are prioritized for avoidance and the path is adjusted according to health status, prioritizing return to base in case of anomalies and optimizing flight altitude and direction in normal conditions to reduce mission risks; real-time updates to path planning respond to environmental changes, improving the safety, energy efficiency, and mission success rate of UAV flight, and adapting to complex dynamic environments.

[0180] S4. Based on dynamic path planning, the intelligent scheduling algorithm formulates tasks and executes plans.

[0181] Furthermore, input the flight mission, dynamic path planning results, and UAV health status to initialize the intelligent scheduling algorithm (such as a genetic algorithm or particle swarm optimization algorithm);

[0182] Genetic Algorithm:

[0183] Chromosome encoding, where each chromosome represents a task allocation scheme (the mapping relationship between drones and tasks);

[0184] The fitness function comprehensively considers task completion rate, path length, power limit, and task priority.

[0185] Optimize task allocation schemes through selection, crossover, and mutation operations;

[0186] Particle Swarm Optimization Algorithm:

[0187] Each particle represents a task allocation scheme;

[0188] Consistent with genetic algorithms, it is used to evaluate task allocation schemes;

[0189] Optimize task allocation through dynamic adjustment of particle position and velocity;

[0190] The optimal task allocation scheme is generated through iterative optimization.

[0191] Output the task allocation plan, including the mapping relationship between drones and tasks;

[0192] Send the mission allocation plan to the UAV navigation center and begin executing the flight mission (including the mission objectives, dynamic path sequence, navigation instructions, and flight parameters for each UAV);

[0193] The mission objectives of each drone, such as the specific execution location;

[0194] Flight parameters, such as altitude and speed;

[0195] Through drone communication, the flight status during flight missions can be monitored in real time (including flight status monitoring, mission progress feedback, and anomaly handling).

[0196] Flight status monitoring provides real-time updates on the drone's flight position, battery level, speed, and other status information;

[0197] Task progress feedback reports the completion progress of the task (such as the number of completed path nodes);

[0198] The abnormal handling procedure is to immediately adjust the mission plan or terminate the mission if an abnormal health status of the drone is detected (such as insufficient power or malfunction).

[0199] After completing its mission, the drone uploads the mission results (such as inspection photos, data collection records, etc.) to the central processing center and updates the mission status ("Completed").

[0200] It should be noted that the intelligent scheduling algorithm optimizes the drone task allocation scheme (such as genetic algorithm or particle swarm optimization algorithm) by using dynamic path planning results, health status and task requirements. It comprehensively considers task completion rate, path length, power limit and task priority to ensure optimal resource allocation; monitors the task execution status in real time and dynamically adjusts the task plan to deal with abnormal conditions (such as insufficient power or abnormal health), improving the flexibility and safety of task execution; and uploads the results and updates the status after the task is completed, realizing a closed loop of task management.

[0201] This embodiment also provides a computer device applicable to the intelligent scheduling and management method of UAVs based on task planning, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the intelligent scheduling and management method of UAVs based on task planning as proposed in the above embodiment.

[0202] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0203] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the task-planning-based intelligent scheduling and management method for unmanned aerial vehicles (UAVs) as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0204] In summary, this invention utilizes a smart patrol dynamic convolutional network model to perform hierarchical feature extraction and fusion on preprocessed flight data, generating a health status feature vector. This vector is then combined with health status values ​​and a One-Class SVM to achieve accurate health status assessment and anomaly detection. The model employs dynamic convolutional layers to extract local features, pooling layers to extract key features, and fully connected layers to learn higher-order feature relationships, ensuring the comprehensiveness and accuracy of health status analysis. By quantifying health status values ​​using formulas and combining them with a One-Class SVM to efficiently handle scenarios with insufficient abnormal samples, the reliability of UAV health monitoring is significantly improved. The smart patrol dynamic convolutional network model provides accurate health status data for dynamic path planning and intelligent scheduling, effectively ensuring the safety and stability of mission execution.

[0205] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent scheduling and management of unmanned aerial vehicles (UAVs) based on task planning, characterized by: include, Collect flight data and perform preprocessing; Based on the preprocessed flight data, a smart patrol dynamic convolutional network model is constructed to analyze the health status of the UAV and generate health status results. Using the generated health status results and combined with environmental perception data, dynamic path planning is performed through a recurrent neural network model. Based on dynamic path planning, intelligent scheduling algorithms formulate tasks and execute plans. The specific steps for constructing the intelligent patrol dynamic convolutional network model are as follows: The input layer of the intelligent patrol dynamic convolutional network model is defined as preprocessed data; The preprocessed data is divided into training set, validation set and test set; Extracting local features from data using dynamic convolutional layers; Add a pooling layer to extract the main features of the data; Local features and main features are merged into a one-dimensional feature vector, and higher-order feature relationships are learned through a fully connected layer to complete feature fusion and generate a health status feature vector. Define the output layer of the intelligent patrol dynamic convolutional network model, and calculate the health status value F(X) based on the health status feature vector; Set a judgment threshold T; If F(X)≤T, then the health status is determined to be abnormal; If F(X) > T, then the health status is determined to be normal; One-Class SVM is introduced to detect the health status feature vector and output the health status result; The intelligent patrol dynamic convolutional network model is trained using the training set to optimize the classification task; The performance of the intelligent patrol dynamic convolutional network model was evaluated using a validation set, and hyperparameters were optimized. A comprehensive evaluation of the Intelligent Patrol Dynamic Convolutional Network model and One-Class SVM was conducted using the test set. The specific steps for introducing One-Class SVM to detect health status feature vectors and output health status results are as follows: Based on the generated health status values ​​and the set judgment threshold, normal status feature vector samples are selected. As training data for One-Class SVM; Using the RBF radial basis kernel as the kernel function of the One-Class SVM, the One-Class SVM is trained to learn the distribution of normal state feature vector samples; During the testing phase, the health status feature vector is input into the trained One-Class SVM; One-Class SVM determines whether a test sample belongs to a normal state based on the distribution of normal state feature vector samples and outputs a health status result. The process of utilizing the generated health status results, combined with environmental perception data, and performing dynamic path planning through a recurrent neural network model involves the following steps: Based on the flight mission, environmental perception data and health status values ​​are used as inputs for path planning; The priority of path adjustment is calculated using the following expression: Where A(P) is the route adjustment priority, P is the evaluation target of the current route planning, e is the total number of environmental factors, j is the index of the environmental factor, and E j D is the urgency weight of the j-th environmental factor. j is the distance between the drone and the j-th environmental factor, k is the distance adjustment factor, and m is the result adjustment factor; If the drone's health status is abnormal, it will be directly assigned the priority of returning to base. If the drone is in good health, prioritize avoiding high-risk areas; The time series data of environmental perception data is processed using a recurrent neural network model, and combined with health status values, to generate a dynamically optimized path node sequence as the result of dynamic path planning. The dynamic path planning results are sent as navigation commands to the UAV navigation center; The UAV navigation center receives navigation commands and parses them to generate flight parameters; The drone executes navigation commands and adjusts flight parameters in real time to avoid obstacles; If the environmental conditions in the environmental perception data change, the path is recalculated and the navigation instructions are updated.

2. The intelligent scheduling and management method for unmanned aerial vehicles based on task planning according to claim 1, characterized in that: The flight data includes GPS location data, speed data, acceleration data, battery power data, and environmental perception data.

3. The intelligent scheduling and management method for unmanned aerial vehicles based on task planning according to claim 1, characterized in that: The preprocessing includes data cleaning, removing outliers and noise, and normalizing the data.

4. The intelligent scheduling and management method for unmanned aerial vehicles based on task planning according to claim 1, characterized in that: The health status value is expressed as: Where F(X) is the health status value, d is the dimension of the health status feature vector, and x i is the actual value of the i-th feature, μ is the average of all features, i is the index of the feature, and ∏ is the product sign. It is a factor that adjusts the scale of the results.

5. The intelligent scheduling and management method for unmanned aerial vehicles based on task planning according to claim 1, characterized in that: The process of using a recurrent neural network model to process time-series environmental perception data, and combining it with health status values, to generate a dynamically optimized path node sequence as the dynamic path planning result, involves the following steps: Define the first layer of a recurrent neural network model, input environmental perception data and health status values, and extract time series features; Define the intermediate layer of the recurrent neural network model, and introduce an LSTM (Long Short-Term Memory) network to capture long-term dependencies in the time series and extract global features; Define the last layer of the recurrent neural network model, which combines the extracted time series features and global features to output a dynamically optimized path node sequence; Define a loss function to avoid high-risk areas, optimize path length, and constrain health status; The dynamically optimized path node sequence generated by the recurrent neural network model is used as the result of dynamic path planning.

6. The intelligent scheduling and management method for unmanned aerial vehicles based on task planning according to claim 1, characterized in that: The process of formulating a task and executing a plan based on dynamic path planning and intelligent scheduling algorithm involves the following steps: Input the flight mission, dynamic path planning results, and UAV health status, and initialize the intelligent scheduling algorithm; Through iterative optimization, a task allocation scheme is generated, and a task allocation plan is output. Send the mission allocation plan to the UAV navigation center and begin executing the flight mission; The flight status during flight missions can be monitored in real time through drone communication. After completing its mission, the drone uploads the mission results to the central processing center and updates the mission status.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the task planning-based intelligent scheduling and management method for unmanned aerial vehicles as described in any one of claims 1 to 6.

Images