Unmanned aerial vehicle control method and device and storage medium
By using the recurrent neural network model to adjust the parameters of the UAV PID controller in real time, the problem of poor control performance of the UAV in complex airflow environments is solved, and more efficient and accurate control effects are achieved.
Patent Information
- Application Number
- CN202411705853.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-02-28
Smart Images
Figure CN119536311A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of automatic control, and in particular to a method, device and storage medium for controlling an unmanned aerial vehicle. Background Art
[0002] With the continuous advancement of science and technology, drones have been widely used in many fields such as agriculture, logistics, photography, and environmental monitoring. In order to ensure the efficient and safe operation of drones in these applications, precise control has become a crucial technical requirement. Among various control methods, the proportional-integral-differential (PID) control method has become a common choice in the field of drone control due to its simplicity, ease of implementation, and strong robustness.
[0003] The existing PID control algorithm requires the pre-setting of three parameters: proportional coefficient, integral time constant and differential time constant. The PID algorithm with preset parameters usually performs well in common environments such as no wind and light wind. However, drones may encounter complex airflow environments such as sudden changes in wind speed. In this case, the fixed-parameter PID control method cannot adjust the parameters to adapt to these changes, which will result in poor control performance such as response lag or reduced control accuracy. Summary of the invention
[0004] The present invention provides a drone control method, device and storage medium, which are used to solve the problem of poor control performance of the drone in a complex airflow environment.
[0005] The present invention adopts the following technical solutions:
[0006] In a first aspect, the present invention provides a method for controlling a drone, comprising: training a preset recurrent neural network model as a parameter adjustment model; collecting flight status data of the drone and control parameters of a PID controller in real time, the flight status data including the attitude angle, position and speed of the drone; converting the flight status data and the control parameters of the PID controller into corresponding time series data, and inputting them into the parameter adjustment model; and controlling the drone through the PID controller using the updated control parameters of the PID controller.
[0007] In a feasible implementation method, a preset recurrent neural network model is trained as a parameter adjustment model, including: obtaining training data, the training data including flight status data of the UAV in various flight environments and control parameters of the PID controller; converting the training data into corresponding time series data; inputting the time series data into a preset recurrent neural network model, training until the output converges, and obtaining a parameter adjustment model.
[0008] In a feasible implementation method, the training data is converted into corresponding time series data, including: performing data cleaning on the training data, the data cleaning includes deleting abnormal data; decomposing the cleaned training data into multiple data segments in chronological order, and the data volume corresponding to each data segment is equal; and serializing each data segment to obtain the corresponding time series data.
[0009] In a feasible implementation method, each data segment is serialized, including: sorting the flight status data and control parameters of the PID controller in each data segment according to a preset order; after the sorting is completed, generating a three-dimensional array for each data segment in time order to complete the serialization process.
[0010] In a feasible implementation method, after the time series data is input into a preset recurrent neural network model, the method also includes: executing a forward propagation process on the recurrent neural network model to calculate the model output result; calculating the loss function value based on the model output result and the pre-acquired data label, the data label is used to indicate the control parameters of the optimal PID controller corresponding to each training data; based on the loss function value, executing a back propagation process by the gradient descent method to update the parameters of the recurrent neural network model; repeating the forward propagation process and the back propagation process until the output converges to obtain a parameter adjustment model.
[0011] In a feasible implementation method, after obtaining the training data, the method also includes: setting the activation function of the recurrent neural network model, including: setting the hidden layer activation function of the recurrent neural network model to a hyperbolic tangent function; and, setting the output layer activation function of the recurrent neural network model to a Sigmoid function.
[0012] In a feasible implementation method, after obtaining the training data, the method further includes: setting the performance indicator function of the recurrent neural network model to: Among them, rout is used to indicate the predicted value of the recurrent neural network model, yout is used to indicate the value of the data label, e is used to indicate the value of the performance indicator function, and k is used to indicate the number of training times.
[0013] In a feasible implementation method, after obtaining the control parameters of the updated PID controller, the method also includes: generating control instructions for the UAV through the PID controller based on the control parameters of the updated PID controller; and sending the control instructions to the UAV so that the actuator of the UAV executes the control instructions.
[0014] In a second aspect, the present invention also provides a drone control device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor so that the at least one processor can execute a drone control method described in any of the above embodiments.
[0015] In a third aspect, the present invention further provides a non-volatile computer storage medium storing computer executable instructions, characterized in that the computer executable instructions are configured to be able to execute a drone control method described in any of the embodiments.
[0016] The present invention provides a drone control method, device and medium, which have the following beneficial effects:
[0017] 1. Based on the recurrent neural network model, the parameter adjustment model can be trained to receive flight status data for a period of time, and determine the control parameters of the optimal PID controller according to the overall flight status for a period of time, which is conducive to improving the accuracy of the control parameters of the PID controller determined by the parameter adjustment model. Using the parameter adjustment model to adjust the control parameters of the PID controller can determine the control parameters of the optimal PID controller suitable for the current environment according to the flight status of the drone and the control parameters of the currently used PID controller, thereby maintaining excellent control effects in complex airflow environments.
[0018] 2. Using the flight status data of the UAV in various flight environments and the control parameters of the PID controller to train the parameter adjustment model can enable the parameter adjustment model to learn the optimal PID controller control parameters corresponding to the flight status data in various flight environments, thereby improving the training effect of the parameter adjustment model.
[0019] 3. Before starting to train the parameter adjustment model, clean the original training data and delete abnormal data to avoid the negative impact of abnormal data on the training effect. At the same time, divide the training data into multiple data segments of equal size and perform serialization processing to ensure the overall integrity of the training data and ensure that the model can learn the temporal dependencies or sequence features in the data. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:
[0021] Figure 1 A flow chart of a drone control method provided by the present invention;
[0022] Figure 2 A schematic structural diagram of a drone control device provided by the present invention. DETAILED DESCRIPTION
[0023] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0024] The present invention provides a UAV control method, which solves the technical problem of poor UAV control effect in a complex airflow environment by using a parameter adjustment model to adjust PID controller parameters in real time.
[0025] The method of the present invention is described in detail below with reference to the accompanying drawings.
[0026] Figure 1 A flow chart of a drone control method provided by the present invention, such as Figure 1 As shown, the method in the present invention at least includes the following execution steps:
[0027] Step 101: train a preset recurrent neural network model as a parameter adjustment model.
[0028] Recurrent Neural Network (RNN) is a type of neural network that specializes in processing sequence data. It can capture the time dependency in the sequence and pass the information of the previous moment to the current moment through the recurrent unit, thereby realizing dynamic modeling of sequence data. The parameter adjustment model is used to adjust the control parameters of the drone's PID controller in real time.
[0029] Specifically, firstly, training data and data labels are obtained. The training data here refers to the flight state data of the drone and the control parameters of the PID controller when the drone flies in various environments. The training data here can be the data collected by the sensor when the drone is actually flying, or it can be the flight state data generated by modeling the drone in a computer and simulating the flight in a simulated environment. The control parameters of the PID controller are randomly generated from a preset range. After collecting the flight state data corresponding to a variety of control parameters in an environment, the PID controller parameters of the drone are adjusted for the above environment, and the optimal PID controller control parameters of the drone in the above environment are determined, which are used as the data labels corresponding to the flight state data collected under the above environment. By collecting the flight state data corresponding to the control parameters of various PID controllers and determining the optimal PID controller control parameters corresponding to the drone in different environments, the recurrent neural network model can learn the difference in the control effects of different PID controller control parameters in different environments. According to the above content and the corresponding data labels, the recurrent neural network model can learn the optimal PID controller control parameters corresponding to different environments.
[0030] After obtaining the training data and data labels, the training data needs to be preprocessed. The preprocessing steps include data cleaning and data segmentation. Data cleaning includes deleting abnormal data and smoothing noise data to ensure the accuracy and integrity of the data, thereby ensuring the effect of model training. Data segmentation is to divide the continuous flight status data into multiple independent data segments, each data segment corresponds to the adjustment process of a PID controller, so as to facilitate subsequent model training. After completing the data segment division, the training data is serialized, that is, the flight status data and the control parameters of the PID controller in each data segment are sorted in a preset order. After the sorting is completed, each data segment is generated into a three-dimensional array in time order to complete the serialization process. Through the above-mentioned serialization process, the overall integrity of the training data can be guaranteed, and it can be ensured that the model can learn the temporal dependencies or sequence features in the data. Finally, the serialized training data is input into the preset recurrent neural network model and trained until the output converges. The specific training steps include performing a forward propagation process on the recurrent neural network model, calculating the model output results, calculating the loss function value based on the model output results and the pre-acquired data labels, and performing a back propagation process based on the loss function value by the gradient descent method to update the parameters of the recurrent neural network model. Repeat the forward propagation process and the back propagation process until the output converges. During the training process, the activation function, performance indicator function, etc. of the model need to be adjusted according to the model output results.
[0031] In one example, the drone is first placed in a variety of airflow environments to perform flight missions and collect flight data. The drone is set to record flight status data and PID controller control parameters every 100ms. Some of the recorded flight data are as follows:
[0032]
[0033] After collecting the above flight status data and PID controller control data, filter out the abnormal values in the data and delete them. Some of the abnormal values filtered out are as follows:
[0034]
[0035] After deleting the outliers, the average of the two values before and after the outliers is used to fill in the gaps to ensure that the interval between the two data is the same. After the outliers are processed, the recorded data is divided into data segments with 30 groups of data per segment. Each data segment contains 30 groups of flight status data and control parameters of the PID controller. One of the data segments is shown below:
[0036]
[0037]
[0038] After dividing the data into segments, it is also necessary to ensure that the data in each row of the data segment are arranged in the same order. In the above data segment example, the order has been completed according to the column name and no further processing is required. Finally, the data segment is converted into a three-dimensional array suitable for input into the RNN. The relevant source code is as follows:
[0039]
[0040]
[0041] After preparing the training data, use the TensorFlow framework to determine the activation function of the recurrent neural network, and use the model.compile() method to specify the performance indicator function. Here, the activation function of the hidden layer is set to the hyperbolic tangent function, the activation function of the output layer is set to the Sigmoid function, and the performance indicator function is:
[0042] After setting the above contents, conduct training and continuously adjust the model's learning rate, training batch, number of network layers and other hyperparameters according to the training results until the training requirements are met.
[0043] In one example, a drone is modeled in a computer, and the flight status data generated by simulating flight in a simulation environment. The drone is a quadcopter drone, weighing 1.6kg, with a diagonal distance of 450mm, each motor is 800g, and the KV value is 980RPM / V. First, Gazebo is selected as the drone simulation software. After selecting the software, a corresponding drone model is created in the simulation software according to the actual physical parameters of the drone. Then the simulation environment is configured, the flight scene is a city, and the environmental parameters are wind speed: 3m / s, wind direction: northeast wind, and temperature: 25℃. Set the flight mode to autonomous flight, and perform a series of actions such as takeoff -> fly to (100,100,50) coordinates -> hover -> return to the take-off point -> land according to the preset route. Execute the above flight process multiple times, and record the flight status data every 100ms. Some data are as follows: Position (x, y, z): 0s: (0, 0, 0), 0.1s: (0.2, 0.3, 0.1), 0.2s: (0.4, 0.5, 0.3); Speed (vx, vy, vz): 0s: (0, 0, 0), 0.1s: (0.5, 0.5, 0.25), 0.2s: (1, 1, 1); Attitude angle (roll, pitch, yaw): 0s: (0°, 0°, 0°), 0.1s: (0.1°, 0.05°, 0.2°), 0.2s: (0.15°, 0.1°, 0.4°); PID controller control parameters are: 1.0, 1.1, 1.05. During the flight, the operation of the PID controller needs to be simulated. Each time the drone completes a round trip, the control parameters of the PID controller are randomly adjusted, and the control parameters and flight status data of different PID controllers are recorded. After multiple round trips, the control parameters of the drone PID controller are adjusted for the above environment, and the optimal PID controller parameters corresponding to the above environment are determined, and the optimal PID controller parameters are used as the data labels corresponding to the above collected training data. Finally, the drone flight status data and PID control parameters collected during the simulated flight are saved as a training data set for training parameter adjustment models.
[0044] Step 102: collect the flight status data of the UAV and the control parameters of the PID controller in real time.
[0045] Specifically, first collect the data sent back by the drone sensor in real time, then record the above data and the control parameters of the PID controller at preset time intervals, and add a timestamp to the data when recording relevant data to determine the time sequence of the data. The time interval for recording data needs to be the same as the time interval for training data during parameter adjustment model training to ensure the adjustment effect of the parameter adjustment model.
[0046] Step 103: Convert the flight status data and the control parameters of the PID controller into corresponding time series data, and input them into a parameter adjustment model.
[0047] Specifically, the collected data is first cleaned, such as deleting abnormal data and backfilling the corresponding data. Then it is determined whether the collected data reaches the number of a data segment, where the data segment size should be equal to the data segment size during parameter adjustment model training. If the collected data reaches the number of a data segment, the data is converted into time series data and input into the parameter adjustment model. Finally, the control parameters of the PID controller are updated to the output results of the parameter adjustment model.
[0048] In a feasible implementation, after the parameter adjustment model outputs the result, it is determined whether the output result is within the preset numerical range. If the input result exceeds the preset numerical range, the parameter adjustment is abandoned. By controlling the input and output results of the model, it can be ensured that when the model outputs abnormal results, the abnormal results are not applied to the PID controller, thereby preventing the PID controller of the drone from being disturbed by the abnormal results and causing oscillation, over-adjustment and other problems.
[0049] Step 104: Use the updated control parameters of the PID controller to control the UAV through the PID controller.
[0050] Specifically, first, the new parameters optimized by the parameter adjustment model are transmitted to the PID controller. These parameters include proportional coefficient, integral coefficient and differential coefficient, which correspond to the three main adjustment links of the PID controller: proportional adjustment, integral adjustment and differential adjustment. Next, the PID controller will collect the current flight status data of the UAV in real time, such as attitude angle, position and speed, which are the basis for the controller to perform precise control. Then, the PID controller uses the updated control parameters to calculate according to the deviation between the collected flight status data and the set target state. The proportional link will immediately generate an adjustment amount according to the size of the deviation; the integral link will consider the cumulative effect of the deviation to eliminate the static error of the system; the differential link predicts the change trend of the deviation and makes adjustments in advance to reduce overshoot and oscillation. Finally, the PID controller combines the adjustment amounts of these three links to generate a control instruction, which will be sent to the actuator of the UAV, such as the motor or servo, so as to adjust the flight state of the UAV so that it gradually approaches and stabilizes on the set target state. In this way, by continuously updating and optimizing the parameters of the PID controller, more accurate and efficient control of the UAV can be achieved.
[0051] Based on the same inventive concept, the present invention also provides a drone control device, the structure of which is as follows: Figure 2 shown.
[0052] Figure 2 This is a schematic diagram of the structure of a drone control device provided by the present invention. Figure 2 As shown, a drone control device 200 in the present invention specifically includes: at least one processor 201; and a memory 203 that is communicatively connected to the at least one processor (connected via a bus 202); wherein the memory 203 stores instructions that can be executed by the at least one processor 201, so that the at least one processor 201 can execute a drone control method as described in the above embodiment.
[0053] In one or more possible implementations of the present invention, the aforementioned processor is used to execute and train a preset recurrent neural network model as a parameter adjustment model; collect the flight status data of the UAV and the control parameters of the PID controller in real time; convert the flight status data and the control parameters of the PID controller into corresponding time series data, and input them into the parameter adjustment model; and use the updated control parameters of the PID controller to control the UAV through the PID controller.
[0054] In addition, the present invention also provides a non-volatile computer storage medium storing computer executable instructions, where the computer executable instructions are configured to be able to execute a drone control method as described in any of the above embodiments.
[0055] In one or more possible implementations of the present invention, the aforementioned computer executable instructions are configured to execute, train a preset recurrent neural network model as a parameter adjustment model; collect the flight status data of the UAV and the control parameters of the PID controller in real time; convert the flight status data and the control parameters of the PID controller into corresponding time series data, and input them into the parameter adjustment model; and use the updated control parameters of the PID controller to control the UAV through the PID controller.
[0056] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the IoT device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.
[0057] The system and medium provided by the present invention correspond one-to-one to the method, and therefore, the system and medium also have similar beneficial technical effects to the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the system and medium will not be repeated here.
[0058] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0059] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0060] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0062] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0063] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0064] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0065] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0066] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A drone control method, characterized in that: The method comprises: Train the preset recurrent neural network model as the parameter adjustment model; Real-time collection of the UAV's flight status data and the control parameters of the PID controller, wherein the flight status data includes the attitude angle, position and speed of the UAV; Converting the flight status data and the control parameters of the PID controller into corresponding time series data, and inputting the data into the parameter adjustment model, so as to update the control parameters of the PID controller according to the output results of the parameter adjustment model; The drone is controlled by the PID controller using the updated control parameters of the PID controller.
2. A drone control method according to claim 1, characterized in that: Train the preset recurrent neural network model as a parameter adjustment model, including: Acquire training data, wherein the training data includes flight status data of the UAV in various flight environments and control parameters of the PID controller; Converting the training data into corresponding time series data; The time series data is input into the preset recurrent neural network model and trained until the output converges to obtain the parameter adjustment model.
3. A drone control method according to claim 2, characterized in that: Converting the training data into corresponding time series data includes: Performing data cleaning on the training data, wherein the data cleaning includes deleting abnormal data; Decompose the cleaned training data into multiple data segments in chronological order, and the amount of data corresponding to each data segment is equal; Each data segment is serialized to obtain the corresponding time series data.
4. A drone control method according to claim 3, characterized in that: Serialize each data segment, including: Sorting the flight status data and the control parameters of the PID controller in each data segment according to a preset order; After the sorting is completed, a three-dimensional array is generated for each data segment in chronological order to complete the serialization process.
5. The method for controlling a drone according to claim 2, characterized in that: After inputting the time series data into the preset recurrent neural network model, the method further includes: Performing a forward propagation process on the recurrent neural network model to calculate the model output result; Calculating a loss function value based on the model output result and pre-acquired data labels, wherein the data labels are used to indicate the optimal control parameters of the PID controller corresponding to each training data; Based on the loss function value, a back propagation process is performed by a gradient descent method to update the parameters of the recurrent neural network model; The forward propagation process and the backward propagation process are repeatedly performed until the output converges to obtain the parameter adjustment model.
6. A drone control method according to claim 2, characterized in that: After obtaining the training data, the method further includes: Setting the activation function of the recurrent neural network model includes: Setting the hidden layer activation function of the recurrent neural network model to the hyperbolic tangent function; And, the output layer activation function of the recurrent neural network model is set to a Sigmoid function.
7. A drone control method according to claim 2, characterized in that: After obtaining the training data, the method further includes: The performance indicator function of the recurrent neural network model is set to: Among them, the r out Used to indicate the predicted value of the recurrent neural network model, y out is used to indicate the value of the data label, and e is used to indicate the value of the performance indicator function.
8. The method for controlling a drone according to claim 1, characterized in that: After obtaining the updated control parameters of the PID controller, the method further includes: Generate control instructions for the drone based on the updated control parameters of the PID controller by means of the PID controller; The control instruction is sent to the drone so that the actuator of the drone executes the control instruction.
9. A drone control device, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute a drone control method according to any one of claims 1-8.
10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions are configured to be able to execute a drone control method according to any one of claims 1-8.