High-efficiency data acquisition method adaptive to environment change

Abstract

The invention relates to the technical field of unmanned aerial vehicle path planning, in particular to an efficient data acquisition method adaptive to environmental changes. A data efficient collection method adaptive to environment change comprises the following steps: S1, collecting historical collection path information of an unmanned aerial vehicle, obtaining environment data of a historical path area of the unmanned aerial vehicle based on the historical collection path information, and according to the environment data of the historical path area of the unmanned aerial vehicle, obtaining a data collection path of the unmanned aerial vehicle; acquiring the optimal number of started sensors when the unmanned aerial vehicle executes data acquisition tasks under different environment conditions; s2, according to the collection path information of the historical unmanned aerial vehicles, obtaining an intersection area of the collection paths of the historical unmanned aerial vehicles, analyzing the information collection accuracy, the number of intersection times, the number of sensors and the optimal number of sensors in the intersection area, and obtaining the weight of each intersection area by using a comprehensive evaluation formula. The optimal number of sensors is automatically adjusted in different environments, so that the path planning of the unmanned aerial vehicle is more reasonable.

Description

Technical Field

[0001] The present invention relates to the technical field of UAV path planning, and particularly to a data-efficient acquisition method adaptable to environmental changes. Background Art

[0002] Under the background of the rapid development of modern UAV technology, UAVs have been widely used in many fields such as environmental monitoring, agricultural management, disaster emergency, and urban planning. Existing UAV path planning and data acquisition methods often have the following problems: First, traditional path planning algorithms such as the Dijkstra algorithm and the A* algorithm are static algorithms, and it is difficult to adjust in real time in the face of changing environments and diverse task requirements; second, the adjustment of the acquisition path and the sensor switch fails to fully combine the environmental conditions, and the reasonable path planning leads to power waste and poor quality of the acquired data. Therefore, reasonable path planning for UAVs has become a key problem that must be faced for data-efficient acquisition. Summary of the Invention

[0003] In order to overcome the shortcoming that UAVs cannot combine the environment to plan routes when acquiring data, the present invention provides a data-efficient acquisition method adaptable to environmental changes.

[0004] The technical implementation scheme of the present invention is as follows: A data-efficient acquisition method adaptable to environmental changes, comprising the following steps:

[0005] S1: Collect the historical acquisition path information of the UAV. Based on the historical acquisition path information, obtain the environmental data of the historical path area of the UAV, and according to the environmental data of the historical path area of the UAV, obtain the optimal number of sensors to be turned on when the UAV performs the data acquisition task under different environmental conditions;

[0006] S2: According to the acquisition path information of the historical UAV, obtain the intersection area of each historical UAV acquisition path, analyze the information acquisition accuracy rate, intersection times, number of sensors, and optimal number of sensors in the intersection area, and use the comprehensive evaluation formula to obtain the weight of each intersection area;

[0007] S3: According to the Dijkstra algorithm and the comprehensive weights of the intersection areas, obtain the first optimal planning path;

[0008] S4: Use the genetic algorithm to optimize the first optimal planning path to obtain the second optimal planning path, and select the most suitable planning path according to the first optimal planning path and the second optimal planning path.

[0009] Preferably, collect the historical acquisition path information of the drone, based on the historical acquisition path information, obtain the environmental data of the historical path area of the drone, and according to the environmental data of the historical path area of the drone, obtain the optimal number of sensors to be turned on when the drone performs data acquisition tasks under different environmental conditions, including: obtaining the environmental area range of each historical path area of the drone, and the union of the environmental area ranges should cover all the historical path areas of the drone, input each environmental condition into the sensor number acquisition model, and obtain the optimal number of sensors to be turned on under each environmental condition.

[0010] Preferably, input each environmental condition into the sensor number acquisition model to obtain the optimal number of sensors to be turned on in each environment, including: obtaining each environmental data, performing data preprocessing on the environmental data such as removing duplicate values, removing outliers, and Z-score standardization to obtain preprocessed environmental data, and using the Adam optimizer and binary cross-entropy loss function to optimize the Transformer model, and finally obtaining the trained sensor number acquisition model, input the preprocessed environmental data into the sensor number acquisition model, and obtain the optimal number of sensors to be turned on by the drone in this environment at this time.

[0011] Preferably, according to the historical acquisition path information of the drone, obtain the intersection area of each historical acquisition path of the drone, analyze the accuracy, intersection times, number of sensors, and optimal number of sensors of the collected data in the intersection area, and use the comprehensive evaluation formula to obtain the weight of each intersection area, including: taking the intersection nodes or intersection routes of each historical acquisition path of the drone as the intersection area, obtaining the intersection times within a preset time period in the intersection area, and the accuracy of the collected data corresponding to different numbers of sensors in each intersection area, obtaining the number of sensors currently turned on by the drone and the optimal number of sensors of the drone in each environment, and using the comprehensive evaluation formula to obtain the weight of each intersection area.

[0012] Preferably, use the comprehensive evaluation formula to obtain the weight of each intersection area, including: where the comprehensive evaluation formula is:

[0013]

[0014] where P is the comprehensive weight, α, β, γ are adjustment weights, s is the intersection times of each intersection area, t is the accuracy of the collected data corresponding to different numbers of sensors in each intersection area, b is the optimal number of sensors of the drone in each environment, c is the decline rate when the number of sensors is less than the optimal number of sensors, d is the decline rate when the number of sensors is greater than the optimal number of sensors, and ε, θ are speed adjustment parameters.

[0015] Preferably, obtaining the first optimal planned path according to the Dijkstra algorithm and the comprehensive weights of the intersection regions includes: obtaining the comprehensive weights of the intersection regions, taking the average value of the total weights of adjacent intersection regions as the first edge weight, taking the reciprocal of the first edge weight as the second edge weight, and using the Dijkstra algorithm according to the second edge weight to obtain the first optimal planned path.

[0016] Preferably, optimizing the first optimal planned path using a genetic algorithm to obtain a second optimal planned path, and selecting the most suitable planned path according to the first optimal planned path and the second optimal planned path includes: taking the intersection regions other than the first optimal planned path as the first screening region, taking the intersection regions with comprehensive weights greater than a preset threshold in the first screening region as the second screening region, taking the second screening region as the initial population, randomly selecting intersection regions for combination in the second screening region, and setting the initial parameters of the genetic algorithm, including the crossover probability, mutation probability, and number of iterations; setting a fitness function for evaluating the quality of each path according to the comprehensive weight of the path, and performing crossover and mutation operations on the first optimal planned path and the randomly selected parent paths according to the evaluation results of the fitness function until the preset number of iterations is reached, and selecting the optimal path as the second optimal planned path.

[0017] Preferably, performing crossover and mutation operations on the first optimal planned path and the randomly selected parent paths according to the evaluation results of the fitness function until the preset number of iterations is reached, and selecting the optimal path as the second optimal planned path includes: using the single-point crossover method as the crossover operation and adopting the elitist retention strategy as the selection operation.

[0018] Preferably, optimizing the first optimal planned path using a genetic algorithm to obtain a second optimal planned path, and selecting the most suitable planned path according to the first optimal planned path and the second optimal planned path includes: calculating the power consumption of the first optimal planned path and the second optimal planned path, and making a judgment according to the actual capacity of the current battery, and selecting the path with the lowest power consumption and the actual battery power greater than the power required by the planned path as the most suitable planned path.

[0019] Preferably, calculating the power consumption of the first optimal planned path and the second optimal planned path and making a judgment according to the actual capacity of the current battery includes: obtaining the loss rate of the current UAV battery using the battery loss formula, and obtaining the actual battery power through the loss rate of the battery, where the battery loss formula is:

[0020] C t = 1―exp(―αN―βT―γt);

[0021] Where C t is the battery loss rate, exp(―αN―βT―γt) is the exponential function, N is the number of battery charge and discharge cycles, T is the battery operating temperature, t is the battery usage time, and α, β, and γ are constant coefficients.

[0022] Beneficial effects: By comprehensively considering factors such as the collection accuracy rate, intersection times, and the optimal number of sensors in different environments in the path intersection area, the present invention generates a comprehensive weight. The optimization of the weight makes the path selection more adaptable to different environmental requirements, thereby improving the accuracy of the collected data and the task completion rate;

[0023] Using the Dijkstra algorithm combined with the intersection area weight to calculate the first optimal path, and using the genetic algorithm to optimize the path to form the second optimal path, effectively avoiding the deficiency that a single algorithm is difficult to adapt to the dynamic environment. This method can flexibly handle the intersection of multiple paths in a complex environment, and by selecting the optimal path combination of the number of sensors, it realizes a more intelligent UAV path planning. Brief Description of the Drawings

[0024] Figure 1 is a flowchart of a data-efficient acquisition method for self-adapting to environmental changes according to the present invention;

[0025] Figure 2 is a schematic diagram of the intersection nodes and intersection routes according to the present invention. Detailed Embodiment

[0026] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0027] A data-efficient acquisition method for self-adapting to environmental changes, as Figure 1-2 shown, includes the following steps:

[0028] S1: Collect the historical acquisition path information of the UAV. Based on the historical acquisition path information, obtain the environmental data of the UAV historical path area, and according to the environmental data of the UAV historical path area, obtain the optimal number of sensors turned on when the UAV performs data acquisition tasks under different environmental conditions;

[0029] Obtain the environmental area range of each historical path area of the UAV. The union of the environmental area ranges should include all the historical path areas of the UAV. Input each environmental condition into the sensor number acquisition model to obtain the optimal number of sensors turned on under each environmental condition.

[0030] It should be noted that, first, all the path areas passed by the drone during the execution of the historical data collection task are segmented and divided into different environmental areas. For example, if the drone has flown over forest, river, and urban areas in the past, its path is decomposed into multiple area segments, and relevant environmental data is recorded for each segment; the ranges of all environmental areas in the historical path are merged to obtain an overall environmental area range that includes all historical path areas, and all the experienced environmental conditions can be effectively reflected through this union; the specific environmental conditions of each environmental area are used as input parameters and input into the sensor quantity acquisition model. By analyzing these historical data, the model calculates the optimal number of sensors suitable for activation under different environmental conditions. For example, in a forest area with high humidity, a temperature and humidity sensor needs to be activated, while in a desert area with strong sunlight, a light sensor and a temperature sensor need to be activated; according to the specific requirements under different environmental conditions, the model outputs the corresponding optimal number of sensors for reference in subsequent path planning and data collection tasks.

[0031] Obtain the environmental data of each environment, and perform data preprocessing on the environmental data, including removing duplicate values, removing outliers, and Z-score standardization, to obtain preprocessed environmental data. Then, use the Adam optimizer and the binary cross-entropy loss function to optimize the Transformer model. Finally, obtain the trained sensor quantity acquisition model. Input the preprocessed environmental data into the sensor quantity acquisition model to obtain the optimal number of sensors that should be activated by the drone in this environment at this time.

[0032] It should be noted that removing duplicate values reduces the interference of redundant data in the model's calculation, and removing outliers ensures the accuracy of the data. Performing Z-score standardization on the environmental data converts the data into a standard normal distribution to improve the model's data processing ability and make the differences between different environmental data smoother; input the standardized environmental data into the optimized model, and the model outputs the optimal number of sensors that the drone should activate under different environments according to the learning results of the training data.

[0033] S2: According to the acquisition path information of the historical drones, obtain the intersection areas of the acquisition paths of each historical drone, analyze the information acquisition accuracy, intersection times, number of sensors, and optimal number of sensors in the intersection areas, and use the comprehensive evaluation formula to obtain the weights of each intersection area;

[0034] Take the intersection nodes or intersection routes of the historical collection paths of each UAV as the intersection area, obtain the intersection times within a preset time period in the intersection area, and the accuracy rates of the collected data corresponding to different numbers of sensors in each intersection area, obtain the number of sensors currently enabled on the UAV and the optimal number of sensors for the UAV in each environment, and use the comprehensive evaluation formula to obtain the weights of each intersection area.

[0035] It should be noted that by analyzing the historical collection paths of multiple UAVs, find the nodes or line segments with intersections in the paths, and define these repeatedly collected path segments as the intersection area. For details, reference can be made to Figure 2 , Intersection times: Count the number of times each UAV repeatedly flies in the intersection area within a preset time period, which reflects the collection frequency of this area; Accuracy rate of collected data: Calculate the collection accuracy rates corresponding to different numbers of sensors based on the sensor data in the intersection area; Current number of sensors and optimal number of sensors: Compare the number of sensors actually used by the UAV in the intersection area with the optimal number of sensors recommended in the environment; Use the comprehensive evaluation formula to calculate the weights of each intersection area based on the above parameters. The level of the weight is used as a reference for subsequent path planning. The higher the weight, the more inclined to collect data in the path of this area.

[0036] The comprehensive evaluation formula is as follows:

[0037]

[0038] Where P is the comprehensive weight, α, β, γ are adjustment weights, s is the intersection times of each intersection area, t is the accuracy rate of the collected data corresponding to different numbers of sensors in each intersection area, b is the optimal number of sensors for the UAV in each environment, c is the decline rate when the number of sensors is less than the optimal number, d is the decline rate when the number of sensors is greater than the optimal number, and ε, θ are speed adjustment parameters.

[0039] It should be noted that by reasonably setting the adjustment coefficients and parameters, the comprehensive weight can comprehensively reflect the data collection priority of the intersection area; this formula combines three factors: intersection times, accuracy rate of collected data, and number of sensors. Among them, the intersection times and collection accuracy rate have a positive impact on the weight, while the impact of the number of sensors on the weight depends on whether it is close to the optimal number. If the number of sensors is much higher or lower than the optimal number, the weight will drop significantly.

[0040] S3: Obtain the first optimal planned path according to Dijkstra's algorithm and the comprehensive weights of the intersection areas;

[0041] Obtain the comprehensive weights of the intersection regions, take the average of the total weights of adjacent intersection regions as the first edge weight, take the reciprocal of the first edge weight as the second edge weight, and use the Dijkstra algorithm according to the second edge weight to obtain the first optimal planning path.

[0042] It should be noted that by taking the reciprocal of the first edge weight, it is transformed into the second edge weight suitable for the Dijkstra algorithm, so as to calculate the optimal path more efficiently. For example, a drone needs to pass through forest, desert, and wetland areas in a certain task. By calculating the comprehensive weights of these areas, in the intersection area between the forest and the wetland, the first edge weight is 0.8, while in the intersection area between the forest and the desert, the first edge weight is 0.6. The Dijkstra algorithm preferentially selects the path with a higher comprehensive weight, that is, the path with a smaller second edge weight, to execute the task.

[0043] S4: Optimize the first optimal planning path using a genetic algorithm to obtain the second optimal planning path, and select the most suitable planning path according to the first optimal planning path and the second optimal planning path.

[0044] Define the intersection regions other than the first optimal planning path as the first screening region, define the intersection regions in the first screening region with comprehensive weights greater than the preset threshold as the second screening region, use the second screening region as the initial population, randomly select intersection regions from the second screening region for combination, and set the initial parameters of the genetic algorithm, including the crossover probability, mutation probability, and number of iterations; set the fitness function, which is used to evaluate the quality of each path according to the comprehensive weight of the path. According to the evaluation results of the fitness function, perform crossover operations and mutation operations on the first optimal planning path and the randomly selected parent paths until the preset number of iterations is reached, and select the optimal path as the second optimal planning path.

[0045] It should be noted that the intersection regions not included in the first optimal planning path are defined as the first screening region. This operation can reduce the amount of calculation and ensure that potential high-quality paths are searched outside the existing optimal path. Randomly select a preset number of intersection regions from the second screening region for combination to form the initial population. These combinations form multiple candidate paths, providing initial solutions for the subsequent iterative optimization of the genetic algorithm; set the initial parameters of the genetic algorithm, including the crossover probability (controlling the crossover combination frequency of the paths), mutation probability (controlling the mutation degree of the paths), and number of iterations; according to the fitness function results, randomly select a parent path to perform a crossover operation with the first optimal planning path to generate new offspring paths; the mutation operation randomly changes some of the intersection regions in the path to explore better solutions. This way can avoid falling into local optima and search for the optimal solution among more path combinations; repeat the above crossover and mutation processes until the set number of iterations is reached, and finally select the path with the highest fitness as the second optimal planning path.

[0046] The single-point crossover method is used as the crossover operation, and the elitist retention strategy is adopted as the selection operation.

[0047] It should be noted that single-point crossover means selecting an intersection point at a random position on the parent path and swapping the path segments before and after it to generate a new child path; the elitist retention strategy is a selection operation used to retain the path with the highest fitness in each iteration to ensure that high-quality paths are not eliminated.

[0048] Calculate the power consumption of the first best planned path and the second best planned path, and make a judgment based on the actual capacity of the current battery. Select the path with the lowest power consumption and the actual battery power greater than the power required for the planned path as the most suitable planned path.

[0049] It should be noted that the power consumption of the first best planned path and the second best planned path is compared. In this process, the power consumption of the drone performing tasks on each path is calculated to find the path with the lowest power consumption; according to the remaining capacity of the drone's current battery, it is judged whether the power required for each path exceeds the actual power; ensure that the selected path can be successfully completed with the current power support to improve the reliability of task execution.

[0050] Use the battery loss formula to obtain the loss rate of the current drone battery, and obtain the actual battery power through the loss rate of the battery. The battery loss formula is:

[0051] C t = 1 - exp(―αN―βT―γt);

[0052] In the formula, C t is the battery loss rate, exp(―αN―βT―γt) is the exponential function, N is the number of battery charge and discharge cycles, T is the battery operating temperature, t is the battery usage time, and α, β, and γ are constant coefficients.

[0053] It should be noted that after obtaining the battery loss rate C t the actual capacity of the current battery is the full charge minus the loss value. For example, if the full charge of the battery is 100% and the current loss rate is 20%, the actual power is 80%.

[0054] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that these embodiments can be changed without departing from the principles and spirit of the present invention. The specification and embodiments are only regarded as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. An efficient data collection method that is adaptive to environmental changes, characterized in that: The following steps are involved: S1: Collect historical collection path information of the drone, obtain environmental data of the historical path area of ​​the drone based on the historical collection path information, and obtain the optimal number of sensors to be turned on when the drone performs data collection tasks under different environmental conditions according to the environmental data of the historical path area of ​​the drone; S2: according to the historical drone collection path information, obtain the intersection area of ​​each historical drone collection path, analyze the information collection accuracy, intersection times, number of sensors and optimal number of sensors in the intersection area, and use a comprehensive evaluation formula to obtain the weight of each intersection area; S3: Obtaining a first optimal planning path according to the Dijkstra algorithm and the comprehensive weights of the intersection areas; S4: using a genetic algorithm to optimize the first optimal planned path to obtain a second optimal planned path, and selecting an optimal planned path according to the first optimal planned path and the second optimal planned path.

2. The method for efficiently collecting data that is adaptive to environmental changes according to claim 1, characterized in that: The method collects historical collection path information of the drone, obtains environmental data of the drone's historical path area based on the historical collection path information, and obtains the optimal number of sensors turned on when the drone performs data collection tasks under different environmental conditions based on the environmental data of the drone's historical path area, including: obtaining the environmental area range of each historical path area of ​​the drone, the union of each environmental area range should include each historical path area of ​​all drones, inputting each environmental condition into a sensor quantity acquisition model, and obtaining the optimal number of sensors turned on under each environmental condition.

3. The method for efficiently collecting data that is adaptive to environmental changes according to claim 2 is characterized in that: The method of inputting each environmental condition into the sensor quantity acquisition model to obtain the optimal number of sensors that should be turned on in each environment includes: obtaining each environmental data, and preprocessing the environmental data by removing duplicate values, removing outliers, and Z-score standardization to obtain preprocessed environmental data, and optimizing the Transformer model using an Adam optimizer and a binary cross entropy loss function, and finally obtaining a trained sensor quantity acquisition model, inputting the preprocessed environmental data into the sensor quantity acquisition model, and obtaining the optimal number of sensors that the drone should turn on in this environment at this time.

4. The method for efficiently collecting data that is adaptive to environmental changes according to claim 1 is characterized in that: According to the historical collection path information of the drone, the intersection area of ​​the historical collection paths of each drone is obtained, the accuracy of the data collected in the intersection area, the number of intersections, the number of sensors and the optimal number of sensors are analyzed, and the weight of each intersection area is obtained using a comprehensive evaluation formula, including: taking the intersection nodes or intersection routes of the historical collection paths of each drone as the intersection area, obtaining the number of intersections within a preset time period in the intersection area, and the accuracy of the collection data corresponding to different numbers of sensors in each intersection area, obtaining the number of sensors currently turned on by the drone and the optimal number of sensors for the drone in each environment, and using a comprehensive evaluation formula to obtain the weight of each intersection area.

5. The method for efficiently collecting data that is adaptive to environmental changes according to claim 4 is characterized in that: The comprehensive evaluation formula is used to obtain the weight of each intersection area, including: wherein the comprehensive evaluation formula is: Where P is the comprehensive weight, α, β, and γ are adjustment weights, s is the number of intersections of each intersection area, t is the accuracy of the collected data corresponding to different numbers of sensors in each intersection area, b is the optimal number of sensors for the UAV in each environment, c is the descent speed when the number of sensors is less than the optimal number of sensors, d is the descent speed when the number of sensors is greater than the optimal number of sensors, and ε and θ are speed adjustment parameters.

6. The method for efficiently collecting data that is adaptive to environmental changes according to claim 5, characterized in that: The method of obtaining the first optimal planning path according to the Dijkstra algorithm and the comprehensive weights of the intersection areas includes: obtaining the comprehensive weights of the intersection areas, taking the average of the sum weights of adjacent intersection areas as the first edge weight, taking the inverse of the first edge weight as the second edge weight, and using the Dijkstra algorithm according to the second edge weight to obtain the first optimal planning path.

7. The method for efficiently collecting data that is adaptive to environmental changes according to claim 1, characterized in that: The method uses a genetic algorithm to optimize the first best planned path to obtain a second best planned path, and selects the most suitable planned path based on the first best planned path and the second best planned path, including: taking the intersection area except the first best planned path as the first screening area, taking the intersection area in the first screening area with a comprehensive weight greater than a preset threshold as the second screening area, taking the second screening area as the initial population, randomly selecting intersection area combinations from the second screening area, setting initial parameters of the genetic algorithm, including crossover probability, mutation probability, and number of iterations; setting a fitness function for evaluating the quality of each path based on the comprehensive weight of the path, and performing crossover and mutation operations on the first best planned path and the randomly selected parent path according to the evaluation result of the fitness function until the preset number of iterations is reached, and selecting the optimal path as the second best planned path.

8. The method for efficiently collecting data that is adaptive to environmental changes according to claim 7 is characterized in that: According to the evaluation result of the fitness function, the first best planned path and the randomly selected parent path are crossover operated and mutated until a preset number of iterations is reached, and the best path is selected as the second best planned path, including: using a single-point crossover method as a crossover operation, and adopting an elite retention strategy as a selection operation.

9. The method for efficiently collecting data that is adaptive to environmental changes according to claim 8, characterized in that: The method uses a genetic algorithm to optimize the first optimal planned path to obtain a second optimal planned path, and selects an optimal planned path based on the first optimal planned path and the second optimal planned path, including: calculating the power consumption of the first optimal planned path and the second optimal planned path, and judging based on the current actual battery capacity, selecting a path with the lowest power consumption and an actual battery power greater than the power required for the planned path as the optimal planned path.

10. The method for efficiently collecting data that is adaptive to environmental changes according to claim 9, characterized in that: The calculating of the power consumption of the first optimal planned path and the second optimal planned path and judging according to the current actual battery capacity includes: obtaining the loss rate of the current drone battery using a battery loss formula, and obtaining the actual battery power through the battery loss rate, wherein the battery loss formula is: C t =1―exp(―αN―βT―γt); In the formula, C t is the battery loss rate, exp(―αN―βT―γt) is an exponential function, N is the number of battery charge and discharge times, T is the battery operating temperature, t is the battery usage time, and α, β, and γ are constant coefficients.

Images