Airspeed prediction method, device, multicopter aircraft and readable storage medium

Abstract

The application discloses a kind of air speed prediction method, device, multi-rotor aircraft and readable storage medium, method includes: obtaining the first force information of the multi-rotor aircraft currently, and obtains the first rotor speed of each rotor in the multi-rotor aircraft;Based on the first rotor speed and the first force information, air speed prediction is carried out by neural network model, and predicted air speed is obtained;The predicted air speed is corrected based on the correction coefficient corresponding to the neural network model, to obtain target air speed.The application can predict the air speed of multi-rotor aircraft according to the first force information and the first rotor speed, and correct the predicted air speed by correction coefficient, to accurately obtain the air speed corresponding to multi-rotor aircraft, improve the accuracy of the air speed prediction of multi-rotor aircraft.

Description

Technical Field

[0001] This application relates to the field of aircraft technology, and in particular to an airspeed prediction method, device, multi-rotor aircraft, and readable storage medium. Background Technology

[0002] Currently, airspeed is typically measured by installing a pitot tube on the aircraft. Generally, airspeed is estimated on the aircraft based on the measurement of two pressures: first, the total pressure obtained using a single pitot tube; and second, the static pressure measured using a static pressure probe flush with the aircraft's fuselage skin.

[0003] Pitot tubes and static pressure probes are exposed to the external environment and can be interfered with by substances or objects upstream of the corresponding sensors, potentially causing partial or complete blockage and resulting in erroneous pressure measurements. Furthermore, the application of pitot tubes in multi-rotor aircraft increases costs, requiring the design of telescopic and folding mechanisms to accommodate the pitot tube, which is challenging. The downwash generated by the high-speed rotation of the multi-rotor rotor also interferes with pitot tube measurements, leading to inaccurate airspeed readings.

[0004] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0005] The main objective of this application is to provide an airspeed prediction method, device, multi-rotor aircraft, and readable storage medium, aiming to solve the technical problem of inaccurate airspeed measurement of multi-rotor aircraft.

[0006] To achieve the above objectives, this application provides an airspeed prediction method applied to a multi-rotor aircraft, comprising:

[0007] Obtain the current first force information of the multi-rotor aircraft, and obtain the first rotor speed of each rotor in the multi-rotor aircraft;

[0008] Based on the first rotor speed and the first force information, the airspeed is predicted by a neural network model to obtain the predicted airspeed.

[0009] The predicted airspeed is corrected based on the correction coefficients corresponding to the neural network model to obtain the target airspeed.

[0010] Furthermore, the step of predicting the predicted airspeed by using a neural network model based on the first rotor speed and the first force information includes:

[0011] The first rotor speed and the first force information are input into a principal component analysis network for principal component analysis to obtain a first analysis result, wherein the first analysis result includes multiple first principal components and the first contribution rate corresponding to each first principal component;

[0012] Based on the first contribution rate, determine the first target principal component among multiple first principal components;

[0013] The first target principal component is input into the neural network model for model training to obtain the predicted airspeed.

[0014] Further, the step of determining the first target principal component among multiple first principal components based on the first contribution rate includes:

[0015] The first principal components are sorted in descending order of their contribution rates to obtain the sorting results.

[0016] By iterating through the sorted results, the first principal component of the current iteration is obtained.

[0017] If the sum of the first contribution rates corresponding to the traversed principal components is less than the preset contribution rate, then the first principal component currently being traversed is added to the first target principal component until the sum of the contribution rates corresponding to the traversed principal components is greater than or equal to the preset contribution rate. The traversed principal components include the first principal components traversed before the currently traversed principal component.

[0018] Furthermore, the step of obtaining the current first force information of the multi-rotor aircraft includes:

[0019] Obtain the current first attitude angle and first acceleration of the multi-rotor aircraft;

[0020] Based on the first attitude angle and the first acceleration, the first force information of the multi-rotor aircraft is determined.

[0021] Furthermore, the step of determining the first force information of the multi-rotor aircraft based on the first attitude angle and the first acceleration includes:

[0022] Obtain the first rotation transformation matrix from the geodetic coordinate system to the body coordinate system of the multirotor aircraft for the first attitude angle;

[0023] Based on the mass of the multirotor aircraft, the first rotational transformation matrix, and the first acceleration, the first force information of the multirotor aircraft is determined.

[0024] Furthermore, before the step of obtaining the current first force information of the multi-rotor aircraft, the method further includes:

[0025] The training sample set is input into the initial neural network model to train the model, thereby obtaining the neural network model;

[0026] The test sample set is input into the neural network model for model training to obtain the estimated airspeed at each sampling time. The test sample set includes test samples corresponding to multiple sampling times.

[0027] Obtain the error sequence between the estimated airspeed and the actual airspeed at each sampling time;

[0028] Based on the error sequence, the correction coefficients corresponding to the neural network model are determined.

[0029] Furthermore, before the step of inputting the training sample set into the initial neural network model for model training to obtain the neural network model, the method further includes:

[0030] Obtain the second force information corresponding to the multi-rotor aircraft;

[0031] Obtain the first airspeed corresponding to the multi-rotor aircraft;

[0032] Obtain the second rotor speed of each rotor in the multi-rotor aircraft;

[0033] The training sample set is determined based on the second force information, the second rotor speed, and the first airspeed.

[0034] Furthermore, the step of determining the training sample set based on the second force information, the second rotor speed, and the first airspeed includes:

[0035] Based on the second rotor speed and the second force information, principal component analysis is performed on the principal component analysis network to obtain a second analysis result, wherein the second analysis result includes multiple second principal components and the second contribution rate corresponding to each second principal component;

[0036] The second target principal component is determined from multiple second principal components based on the second contribution rate;

[0037] The training sample set is determined based on the second target principal component and the first airspeed.

[0038] Furthermore, the step of obtaining the current second force information of the multi-rotor aircraft includes:

[0039] Obtain the second attitude angle and second acceleration of the multi-rotor aircraft;

[0040] Obtain the second rotation transformation matrix from the geodetic coordinate system to the body coordinate system of the multirotor aircraft for the second attitude angle;

[0041] Based on the mass of the multirotor aircraft, the second rotational transformation matrix, and the second acceleration, the second force information of the multirotor aircraft is determined.

[0042] Furthermore, the step of obtaining the first airspeed corresponding to the multi-rotor aircraft includes:

[0043] Obtain the velocity of the multirotor aircraft in the corresponding geodetic coordinate system, as well as the ambient wind speed of the environment in which the multirotor aircraft is located;

[0044] The first airspeed is determined based on the speed and the ambient wind speed.

[0045] Furthermore, before the step of inputting the test sample set into the neural network model for model training to obtain the estimated airspeed at each sampling time, the method further includes:

[0046] Obtain the third force information of the multi-rotor aircraft at the sampling time;

[0047] Obtain the second airspeed of the multi-rotor aircraft at the sampling time;

[0048] Obtain the third rotor speed of each rotor in the multi-rotor aircraft at the sampling time;

[0049] The test sample set is determined based on the third force information, the third rotor speed, and the second airspeed.

[0050] Further, the step of determining the correction coefficients corresponding to the neural network model based on the error sequence includes:

[0051] If the error sequence passes the KS test, then a confidence interval corresponding to the preset confidence level is determined in the error sequence;

[0052] The correction coefficient is determined based on the endpoint values ​​of the confidence interval.

[0053] Furthermore, to achieve the above objectives, this application also provides a multi-rotor aircraft, the multi-rotor aircraft comprising:

[0054] The acquisition module is used to acquire the current first force information of the multi-rotor aircraft and acquire the first rotor speed of each rotor in the multi-rotor aircraft;

[0055] The prediction module is used to predict the airspeed based on the first rotor speed and the first force information by using a neural network model.

[0056] The correction module is used to correct the predicted airspeed based on the correction coefficients corresponding to the neural network model in order to obtain the target airspeed.

[0057] In addition, to achieve the above objectives, this application also provides an airspeed prediction device, characterized in that the airspeed prediction device includes: a memory, a processor, and an airspeed prediction program stored in the memory and executable on the processor, wherein when the airspeed prediction program is executed by the processor, it implements the steps of the airspeed prediction method as described above.

[0058] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing an airspeed prediction program, which, when executed by a processor, implements the steps of the airspeed prediction method as described above.

[0059] This application obtains the first force information of the multi-rotor aircraft and the first rotor speed of each rotor in the multi-rotor aircraft; then, based on the first rotor speed and the first force information, it uses a neural network model to predict the airspeed and obtain the predicted airspeed; subsequently, it corrects the predicted airspeed based on the correction coefficients corresponding to the neural network model to obtain the target airspeed. This method can predict the airspeed of the multi-rotor aircraft based on the first force information and the first rotor speed, and correct the predicted airspeed using correction coefficients to accurately obtain the corresponding airspeed of the multi-rotor aircraft, thus improving the accuracy of airspeed prediction. Compared with measuring airspeed using an airspeed tube, this method significantly reduces the cost of the multi-rotor aircraft and eliminates the interference caused by the downwash flow generated by the high-speed rotor rotation on the airspeed tube measurement, thereby improving the accuracy of airspeed prediction. Compared with using a voltage sensor to measure the equivalent voltage of the multi-rotor drive motor for prediction, directly acquiring the rotational speed of the flight electric drive (first rotor speed) avoids the parameter conversion errors caused by equivalent voltage and equivalent current, thus improving the accuracy of airspeed prediction. Attached Figure Description

[0060] Figure 1 This is a schematic diagram of the airspeed prediction device in the hardware operating environment involved in the embodiments of this application;

[0061] Figure 2 This is a flowchart illustrating the first embodiment of the airspeed prediction method of this application;

[0062] Figure 3 This is a schematic diagram of the functional modules of the multi-rotor aircraft of this application.

[0063] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0064] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0065] like Figure 1 As shown, Figure 1 This is a schematic diagram of the airspeed prediction device in the hardware operating environment involved in the embodiments of this application.

[0066] The airspeed prediction device in this application embodiment can be a multi-rotor aircraft, which can be a 4-rotor, 6-rotor, 8-rotor, or other aircraft, and can be a manned aircraft or an unmanned aircraft.

[0067] like Figure 1 As shown, the airspeed prediction device may include: a processor 1001, such as a CPU; a network interface 1004; a user interface 1003; a memory 1005; and a communication bus 1002. The communication bus 1002 is used to establish communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0068] Optionally, the airspeed prediction device may also include a camera, RF (Radio Frequency) circuitry, sensors, audio circuitry, a WiFi module, and so on. These sensors may include, for example, light sensors, motion sensors, and other sensors. Of course, the airspeed prediction device may also be equipped with other sensors such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, which will not be elaborated upon here.

[0069] Those skilled in the art will understand that Figure 1 The terminal structure shown does not constitute a limitation on the airspeed prediction device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0070] like Figure 1 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an airspeed prediction program.

[0071] exist Figure 1In the airspeed prediction device shown, the network interface 1004 is mainly used to connect to the backend server and communicate data with the backend server; the user interface 1003 is mainly used to connect to the client (user terminal) and communicate data with the client; and the processor 1001 can be used to call the airspeed prediction program stored in the memory 1005.

[0072] In this embodiment, the airspeed prediction device includes: a memory 1005, a processor 1001, and an airspeed prediction program stored in the memory 1005 and executable on the processor 1001. When the processor 1001 calls the airspeed prediction program stored in the memory 1005, it executes the steps of the airspeed prediction methods in the following embodiments.

[0073] This application also provides an airspeed prediction method, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the airspeed prediction method of this application.

[0074] This airspeed prediction method is applied to multi-rotor aircraft, which can be 4-rotor, 6-rotor, 8-rotor, etc., and can be manned or unmanned aircraft.

[0075] Specifically, the airspeed prediction method includes:

[0076] Step S101: Obtain the current first force information of the multi-rotor aircraft and obtain the first rotor speed of each rotor in the multi-rotor aircraft;

[0077] When predicting the airspeed of a multi-rotor aircraft, the multi-rotor aircraft obtains the current first force information, which is the force on the multi-rotor aircraft in its body coordinate system. The multi-rotor aircraft can first obtain the current attitude angle, acceleration and mass of the multi-rotor aircraft, and calculate the first force information based on the attitude angle, acceleration and mass using Newton's second law.

[0078] Meanwhile, the multi-rotor aircraft obtains the first rotor speed of each rotor in the multi-rotor aircraft. Specifically, a speed sensor can be installed in the flight electric drive of the multi-rotor aircraft to collect the current speed of the flight electric drive in order to obtain the first rotor speed of each rotor.

[0079] Step S102: Based on the first rotor speed and the first force information, airspeed is predicted using a neural network model to obtain the predicted airspeed.

[0080] Before predicting the airspeed of a multirotor aircraft, an initial neural network model is trained to obtain the neural network model. This neural network model can be a backpropagation neural network model (BP neural network model) or an Elman neural network module, etc.

[0081] After obtaining the first rotor speed and the first force information, the airspeed is predicted by a neural network model based on the first rotor speed and the first force information. Specifically, principal component analysis is first performed on the first rotor speed and the first force information. Training samples (target principal components) are obtained according to the analysis results. The training samples are input into the neural network model for model training. The output of the neural network model is used as the predicted airspeed.

[0082] Step S103: Correct the predicted airspeed based on the correction coefficients corresponding to the neural network model to obtain the target airspeed.

[0083] Before predicting the airspeed of a multirotor aircraft, the neural network model is trained and tested using test data (e.g., a test sample set), and correction coefficients are obtained from the test information.

[0084] When the predicted airspeed is obtained through a neural network model, the corresponding correction coefficient is acquired, and the predicted airspeed is corrected using this correction coefficient to obtain the target airspeed. For example, if the predicted airspeed is... If the correction factor is β, then the target airspeed This allows for the accurate acquisition of the airspeed of the multirotor aircraft in its current environment.

[0085] It should be noted that when the target airspeed is obtained, it can be transmitted to the flight control system so that the flight control system can adjust the flight of the multi-rotor aircraft according to the target airspeed.

[0086] By acquiring the current first force information of the multirotor aircraft and obtaining the first rotor speed of each rotor in the multirotor aircraft; then, based on the first rotor speed and the first force information, airspeed prediction is performed through a neural network model to obtain the predicted airspeed; subsequently, the predicted airspeed is corrected based on the correction coefficients corresponding to the neural network model to obtain the target airspeed. This method can predict the airspeed of the multirotor aircraft based on the first force information and the first rotor speed, and correct the predicted airspeed through correction coefficients to accurately obtain the corresponding airspeed of the multirotor aircraft, thus improving the accuracy of airspeed prediction. Compared with measuring airspeed through an airspeed tube, this method significantly reduces the cost of the multirotor aircraft and eliminates the interference caused by the downwash flow generated by the high-speed rotor rotation on the airspeed tube measurement, thereby improving the accuracy of airspeed prediction. Compared with using a voltage sensor to measure the equivalent voltage of the multirotor drive motor for prediction, directly acquiring the rotational speed of the flight electric drive (first rotor speed) avoids the parameter conversion errors caused by equivalent voltage and equivalent current, thus improving the accuracy of airspeed prediction.

[0087] Based on the first embodiment, a second embodiment of the airspeed prediction method of this application is proposed, which includes all the contents of the first embodiment, wherein step S102 includes:

[0088] Step S201: Input the first rotor speed and the first force information into the principal component analysis network to perform principal component analysis and obtain the first analysis result, wherein the first analysis result includes multiple first principal components and the first contribution rate corresponding to each first principal component;

[0089] Step S202: Determine the first target principal component among multiple first principal components based on the first contribution rate;

[0090] Step S203: Input the first target principal component into the neural network model for model training to obtain the predicted airspeed.

[0091] After obtaining the first rotor speed and the first force information, the first rotor speed and the first force information are input into the principal component analysis network for principal component analysis to obtain the first analysis result. The principal component analysis network first normalizes the data, then solves the correlation matrix between variables, and finally uses the Jacobian method to calculate the eigenvalues ​​(first principal components) and their contribution rates of the principal components. The principal component analysis network can be an existing principal component analysis network. The first analysis result includes multiple first principal components and the first contribution rate corresponding to each first principal component.

[0092] After obtaining the first contribution rate and the first principal component, the first target principal component is determined from multiple first principal components based on the first contribution rate. Specifically, in one possible implementation, step S202 includes:

[0093] Step a: Sort the first principal components in descending order of their contribution rates to obtain a sorting result;

[0094] Step b: Iterate through the sorting results sequentially to obtain the first principal component of the current iteration;

[0095] Step c: If the sum of the contribution rates of the traversed principal components is less than the preset contribution rate, then the first principal component currently being traversed is added to the first target principal component until the sum of the contribution rates of the traversed principal components is greater than or equal to the preset contribution rate. The traversed principal components include the principal components traversed before the first principal component currently being traversed.

[0096] After obtaining the first contribution rate and the first principal component, the multi-rotor aircraft sorts each first principal component in descending order of the first contribution rate to obtain the sorting result, and then iterates through the sorting result in turn, that is, it iterates through the sorting result starting from the first position of the sorting result to obtain the first principal component of the current iteration.

[0097] Next, the multi-rotor aircraft calculates the sum of contribution rates corresponding to the traversed principal components, that is, adds up the first contribution rates corresponding to each traversed principal component. The traversed principal components include the first principal components traversed before the currently traversed first principal component. If the currently traversed first principal component is not the first principal component, then the sum of contribution rates corresponding to the traversed principal components is zero, and it is determined whether the sum of contribution rates is less than the preset contribution rate. If the sum of contribution rates is less than the preset contribution rate, the currently traversed first principal component is added to the first target principal component, and the sorting result of the traversal is returned to obtain the currently traversed first principal component, until the sum of contribution rates corresponding to the traversed principal components is greater than or equal to the preset contribution rate.

[0098] By selecting the most correlated components as the first target principal components, most of the original data features are preserved, while avoiding the problems of model complexity, high computational requirements, and non-convergence caused by too many input parameters in the neural network model. This can effectively reduce the amount of data for model training, improve the efficiency of neural network model training, and thus improve the efficiency of airspeed prediction.

[0099] After obtaining the first target principal component, the first target principal component is input into the neural network model for model training, and the output of the neural network model is used as the predicted airspeed.

[0100] By inputting the first rotor speed and the first force information into a principal component analysis network for principal component analysis, a first analysis result is obtained, wherein the first analysis result includes multiple first principal components and a first contribution rate corresponding to each first principal component; then, a first target principal component is determined among the multiple first principal components based on the first contribution rate; then, the first target principal component is input into the neural network model for model training to obtain the predicted airspeed. By selecting the first target principal component based on the contribution rate among the first principal components, the amount of data for model training can be effectively reduced, the efficiency of neural network model training can be improved, and thus the efficiency of airspeed prediction can be improved.

[0101] Based on the first embodiment, a third embodiment of the airspeed prediction method of this application is proposed, which includes all the contents of the first embodiment, wherein step S101 includes:

[0102] Step S301: Obtain the current first attitude angle and first acceleration of the multi-rotor aircraft;

[0103] Step S302: Based on the first attitude angle and the first acceleration, determine the first force information of the multi-rotor aircraft.

[0104] When predicting the airspeed of a multirotor aircraft, the multirotor aircraft can first obtain its current first attitude angle and first acceleration. Specifically, the first attitude angle [Ф, θ, ψ] of the aircraft in the body coordinate system is measured by the gyroscope of the multirotor aircraft. T The first acceleration a = [a] of the multi-rotor aircraft in the body coordinate system is measured using the accelerometer. bx a by a bz ] T .

[0105] Next, based on the first attitude angle and the first acceleration, the multirotor determines the first force information of the multirotor. Specifically, in one possible implementation, step S302 includes:

[0106] Step d: Obtain the first rotation transformation matrix of the first attitude angle from the geodetic coordinate system to the body coordinate system of the multi-rotor aircraft;

[0107] Step e: Based on the mass of the multirotor aircraft, the first rotational transformation matrix, and the first acceleration, determine the first force information of the multirotor aircraft.

[0108] After obtaining the first attitude angle and the first acceleration, the multirotor aircraft obtains the first rotation transformation matrix from the geodetic coordinate system to the multirotor aircraft's body coordinate system for the first attitude angle. This first rotation transformation matrix R ebThe formula is:

[0109]

[0110] Among them, R eb Let be the first rotation transformation matrix, and Ф, θ, ψ be the first attitude angles [Ф, θ, ψ]. T The parameters.

[0111] After obtaining the first rotational transformation matrix, the multirotor aircraft obtains its mass m. This mass m can be obtained by weighing and stored in the multirotor aircraft, or the mass of the aircraft can be obtained by collecting the rotor speed in a windless environment and estimating it through an existing rotor dynamics model.

[0112] Then, based on the mass of the multirotor aircraft, the first rotational transformation matrix, and the first acceleration, the first force information of the multirotor aircraft, i.e., the current force vector of the multirotor aircraft, is determined. Specifically, the formula for the first force information is:

[0113] F b =m(aR) eb g);

[0114] Among them, F b As the first force information, R eb Let m be the first rotation transformation matrix, m be the mass of the multirotor aircraft, and g be the gravitational acceleration at the current position of the multirotor aircraft.

[0115] By acquiring the current first attitude angle and first acceleration of the multirotor aircraft, and then determining the first force information of the multirotor aircraft based on the first attitude angle and first acceleration, the first force information can be accurately obtained through the first attitude angle and first acceleration, thereby improving the accuracy of predicted airspeed and further improving the accuracy of airspeed estimation.

[0116] Based on the above embodiments, a fourth embodiment of the airspeed prediction method of this application is proposed, wherein, before step S101, the airspeed prediction method further includes:

[0117] Step S401: Input the training sample set into the initial neural network model to train the model, so as to obtain the neural network model;

[0118] Step S402: Input the test sample set into the neural network model for model training to obtain the estimated airspeed at each sampling time. The test sample set includes test samples corresponding to multiple sampling times.

[0119] Step S403: Obtain the error sequence between the estimated airspeed and the actual airspeed corresponding to each sampling time.

[0120] Step S404: Based on the error sequence, determine the correction coefficients corresponding to the neural network model.

[0121] Before predicting the airspeed of a multi-rotor aircraft, the multi-rotor aircraft first trains an initial neural network model to obtain the neural network model. Specifically, the multi-rotor aircraft acquires a training sample set, which includes multiple sets of training samples. Each set of training samples includes the second target principal component and the first airspeed at the same time. The second target principal component is obtained by collecting the second force information and the second rotor speed at the same time.

[0122] Next, the multirotor aircraft inputs the training sample set into the initial neural network model for model training to obtain the neural network model. Specifically, the multirotor aircraft sequentially inputs the training samples from the training sample set into the initial neural network model for model training. Each training iteration yields a trained neural network model. If the trained neural network model converges upon completion of a particular training iteration, then that trained neural network model is used as the neural network model. Existing algorithms, such as the loss function method, can be used to determine whether the trained neural network model has converged. For example, the loss function of the trained neural network model can be calculated using the output of the trained neural network model and the first airspeed in the corresponding training sample. If the loss function is less than a preset value, the trained neural network model is considered to have converged.

[0123] After obtaining the neural network model, the multi-rotor aircraft acquires a test sample set, which includes test samples corresponding to multiple sampling times. Specifically, the test sample set includes test samples at multiple sampling times. Each set of test samples includes the third target principal component and the second airspeed at the same time. The third target principal component is obtained by collecting the third force information and the third rotor speed at the same time. The test samples in the test sample set are arranged in the order of sampling times.

[0124] The test sample set is input into the neural network model for model training to obtain the estimated airspeed at each sampling time. Specifically, each test sample in the test sample set can be sequentially input into the neural network model for model training, and the output of the neural network model can be used as the estimated airspeed at each sampling time.

[0125] Then, the error sequence between the estimated airspeed and the actual airspeed at each sampling time is obtained. For example, the estimated airspeeds are respectively The actual airspeeds are U b =[U b1 U b2 ...U bt ] TThen the error sequence δ=[δ1,δ2,...δ t ] T ,in,

[0126] Finally, based on the error sequence, the correction coefficients corresponding to the neural network model are determined. Specifically, in one implementation, step S404 includes:

[0127] Step f: If the error sequence passes the KS test, then determine the confidence interval corresponding to the preset confidence level in the error sequence;

[0128] Step g: Determine the correction coefficient based on the endpoint values ​​of the confidence interval.

[0129] After obtaining the error sequence, the Kolmogorov-Smirnov test is performed on the error sequence to determine whether the error sequence conforms to a normal distribution. Specifically, the asymptotic significance of the error sequence is obtained through the KS test algorithm. If the asymptotic significance is greater than or equal to the preset value α, it is determined that the error sequence conforms to a normal distribution and that the error sequence passes the KS test.

[0130] When the error sequence passes the KS test, a confidence interval corresponding to a pre-set confidence level is determined in the error sequence. This pre-set confidence level can be 90%, 95%, etc. Then, the correction coefficient is determined based on the endpoint values ​​of the confidence interval. For example, if the confidence interval is [a, b], the correction coefficient β is determined to be β = (a + b) / 2. This can correct the redundancy error of the airspeed estimate (predicted airspeed) through the confidence interval, improve the rationality of the airspeed estimate result through the KS test, and further improve the accuracy of the target airspeed.

[0131] It should be noted that if the error sequence fails the KS test, the test sample set is re-acquired, and the process returns to step S402 until the error sequence passes the KS test.

[0132] Understandably, to ensure experimental safety and wind field uniformity, the multirotor aircraft should be placed in an outdoor environment with a wide field of vision when collecting test and training sample sets. The three-dimensional ultrasonic anemometer of the multirotor aircraft should be installed above the multirotor aircraft or at a distance of no more than 6 meters from the multirotor aircraft at the same altitude plane. This is to ensure that the measured wind speed is approximately equal to the wind speed of the multirotor aircraft and is not affected by the turbulence generated by the high-speed rotating rotor. The ground clearance of the multirotor aircraft should be above 8 meters to avoid the influence of the ground effect on the rotor thrust.

[0133] The neural network model is trained by inputting a training sample set into an initial neural network model. Then, a test sample set is input into the neural network model for further training to obtain estimated airspeeds at various sampling times. The test sample set includes test samples corresponding to multiple sampling times. Next, an error sequence is obtained between the estimated airspeed and the actual airspeed at each sampling time. Finally, based on the error sequence, correction coefficients are determined for the neural network model. By pre-training the neural network model using the training sample set and obtaining the correction coefficients using the test sample set, the efficiency of multirotor airspeed prediction is improved.

[0134] Based on the fourth embodiment, a fifth embodiment of the airspeed prediction method of this application is proposed, wherein, before step S401, the airspeed prediction method further includes:

[0135] Step S501: Obtain the second force information corresponding to the multi-rotor aircraft;

[0136] Step S502: Obtain the first airspeed corresponding to the multi-rotor aircraft;

[0137] Step S503: Obtain the second rotor speed of each rotor in the multi-rotor aircraft;

[0138] Step S504: Determine the training sample set based on the second force information, the second rotor speed, and the first airspeed.

[0139] The multi-rotor aircraft acquires corresponding second force information, which refers to the forces acting on the multi-rotor aircraft in its body coordinate system. The multi-rotor aircraft can first acquire its attitude angles, acceleration, and mass, and then calculate the second force information using Newton's second law based on these parameters. The training sample set may include training samples collected at multiple acquisition times, and the second force information includes force information from multiple acquisition times during the multi-rotor aircraft's flight.

[0140] The first airspeed corresponding to the multi-rotor aircraft is obtained. The first airspeed includes the airspeed at multiple acquisition moments during the flight of the multi-rotor aircraft. Specifically, for each first airspeed, the first airspeed at that acquisition moment can be calculated by the speed of the multi-rotor aircraft in the corresponding geodetic coordinate system and the ambient wind speed at the corresponding acquisition moment.

[0141] Simultaneously, the second rotor speed of each rotor in the multi-rotor aircraft is obtained. The second rotor speed includes the rotor speed of each rotor at each acquisition moment. For each acquisition moment, the rotational speed of the flight electric drive at the current moment can be collected by the rotational speed sensor to obtain the second rotor speed of each rotor.

[0142] Finally, based on the second force information, the second rotor speed, and the first airspeed, a training sample set is determined. Specifically, in one possible implementation, step S504 includes:

[0143] Step S5041: Based on the second rotor speed and the second force information, input the principal component analysis network to perform principal component analysis and obtain a second analysis result. The second analysis result includes multiple second principal components and the second contribution rate corresponding to each second principal component.

[0144] Step S5042: Determine the second target principal component among multiple second principal components based on the second contribution rate;

[0145] Step S5043: Determine the training sample set based on the second target principal component and the first airspeed.

[0146] After obtaining the second rotor speed and the second force information, the second rotor speed and the second force information are input into a principal component analysis network for principal component analysis to obtain the second analysis result. The principal component analysis network first normalizes the data, then solves the correlation matrix between variables, and finally uses the Jacobian method to calculate the eigenvalues ​​(second principal components) and their contribution rates of the principal components. The principal component analysis network can be an existing principal component analysis network. The second analysis result includes multiple second principal components and the second contribution rate corresponding to each second principal component.

[0147] After obtaining the second contribution rate and the second principal component, the multi-rotor aircraft determines the second target principal component from multiple second principal components based on the second contribution rate. Specifically, the multi-rotor aircraft first sorts each second principal component in descending order of the second contribution rate to obtain a sorting result, and then iterates through the sorting result sequentially, starting from the second position of the sorting result to obtain the currently traversed second principal component. The multi-rotor aircraft calculates the sum of the contribution rates corresponding to the traversed principal components, that is, it adds up the second contribution rates corresponding to the traversed principal components. The traversed principal components include the second principal components traversed before the currently traversed second principal component. If the currently traversed first principal component is a first principal component, then the sum of the contribution rates corresponding to the traversed principal components is zero, and it is determined whether the sum of the contribution rates is less than the preset contribution rate. If the sum of the contribution rates is less than the preset contribution rate, the currently traversed second principal component is added to the second target principal component, and the process of traversing the sorting result is returned to obtain the currently traversed second principal component, until the sum of the contribution rates corresponding to the traversed principal components is greater than or equal to the preset contribution rate.

[0148] By selecting the most correlated components as the second principal components, most of the original data features are preserved, while avoiding the problems of model complexity, high computational requirements, and non-convergence caused by too many input parameters in the neural network model. This can effectively reduce the amount of data for model training and improve the efficiency of initial neural network model training.

[0149] Next, after obtaining the second target principal component, the multi-rotor aircraft determines the training sample set based on the second target principal component and the first airspeed. That is, each training sample in the training sample set includes the second target principal component and the first airspeed corresponding to each acquisition time. In other words, the second target principal component and the first airspeed corresponding to each acquisition time constitute a training sample, and the training samples from multiple acquisition times form the training sample set.

[0150] In one possible implementation, step S501 includes:

[0151] Step S5011: Obtain the second attitude angle and second acceleration of the multi-rotor aircraft;

[0152] Step S5012: Obtain the second rotation transformation matrix of the second attitude angle from the geodetic coordinate system to the body coordinate system of the multi-rotor aircraft;

[0153] Step S5013: Based on the mass of the multi-rotor aircraft, the second rotation transformation matrix, and the second acceleration, determine the second force information of the multi-rotor aircraft.

[0154] The multi-rotor aircraft can first acquire the second attitude angle and second acceleration corresponding to each acquisition moment. Specifically, for each acquisition moment, the second attitude angle of the aircraft in the body coordinate system is measured by the gyroscope of the multi-rotor aircraft, and the second acceleration of the aircraft in the body coordinate system is measured by the accelerometer of the multi-rotor aircraft.

[0155] Next, the multi-rotor aircraft acquires the second attitude angle at each acquisition time, and the second rotation transformation matrix from the geodetic coordinate system to the multi-rotor aircraft's body coordinate system. The formula for the second rotation transformation matrix is ​​similar to that for the first rotation transformation matrix, and will not be repeated here.

[0156] After obtaining the first rotational transformation matrix, the multirotor aircraft acquires its mass. Based on the multirotor aircraft's mass, the second rotational transformation matrix, and the second acceleration, the second force information of the multirotor aircraft is determined. That is, for each acquisition moment, the second force information at that acquisition moment is calculated based on the multirotor aircraft's mass, the second rotational transformation matrix at that acquisition moment, and the second acceleration. The formula for the second force information is similar to the formula for the first force information and will not be repeated here. The second force information at each acquisition moment can be accurately obtained through the second attitude angle and the second acceleration, thereby improving the efficiency of model training of the initial neural network model.

[0157] In one possible implementation, step S502 includes:

[0158] Step S5021: Obtain the velocity in the geodetic coordinate system corresponding to the multi-rotor aircraft, and the ambient wind speed of the environment in which the multi-rotor aircraft is located.

[0159] Step S5022: Determine the first airspeed based on the speed and the ambient wind speed.

[0160] At each data collection moment, the multi-rotor aircraft uses GPS to collect its velocity in the geodetic coordinate system and a three-dimensional ultrasonic anemometer to collect the ambient wind speed of the environment in which the multi-rotor aircraft is located.

[0161] Next, the multi-rotor aircraft determines the first airspeed based on the speed and ambient wind speed. Specifically, for each data acquisition moment, the multi-rotor aircraft calculates the first airspeed at that time based on the speed and ambient wind speed. Furthermore, the multi-rotor aircraft calculates the first airspeed at that time based on the speed, ambient wind speed, and the second rotation transformation matrix. Thus, the first airspeed can be accurately obtained through speed and ambient wind speed, thereby improving the efficiency of the initial neural network model for model training.

[0162] For example, the velocity V at a certain acquisition moment e =[Vex V ey V ez ] T Ambient wind speed W e =[W ex W ey W ez ] T The second rotation transformation matrix is ​​R. eb1 Then the first air velocity U at that sampling moment b =R eb1 (V e +W e ).

[0163] By acquiring the second force information corresponding to the multi-rotor aircraft; then acquiring the first airspeed corresponding to the multi-rotor aircraft; and then acquiring the second rotor speed of each rotor in the multi-rotor aircraft; and then determining the training sample set based on the second force information, the second rotor speed, and the first airspeed, the training sample set can be accurately obtained through the second force information, the second rotor speed, and the first airspeed, thereby training a neural network model and further improving the accuracy of the neural network model in predicting airspeed.

[0164] Based on the fourth embodiment, a sixth embodiment of the airspeed prediction method of this application is proposed, wherein, before step S402, the airspeed prediction method further includes:

[0165] Step S601: Obtain the third force information of the multi-rotor aircraft at the sampling time;

[0166] Step S602: Obtain the second airspeed of the multi-rotor aircraft at the sampling time;

[0167] Step S603: Obtain the third rotor speed of each rotor in the multi-rotor aircraft at the sampling time;

[0168] Step S604: Determine the test sample set based on the third force information, the third rotor speed, and the second airspeed.

[0169] After obtaining the neural network model, it is necessary to test the neural network model. The multi-rotor aircraft first obtains a test sample set, which includes test samples corresponding to multiple sampling times.

[0170] For each sampling moment, the multi-rotor aircraft acquires the third force information, the third rotor speed, and the second airspeed corresponding to that sampling moment.

[0171] Specifically, for each sampling moment, the third attitude angle and the third acceleration of the multi-rotor aircraft are obtained, the third rotation transformation matrix of the third attitude angle from the geodetic coordinate system to the body coordinate system of the multi-rotor aircraft is obtained, and the third force information of the multi-rotor aircraft is determined based on the mass of the multi-rotor aircraft, the third rotation transformation matrix, and the third acceleration.

[0172] In other words, for each sampling moment, the third attitude angle of the multirotor aircraft in the body coordinate system is measured by the gyroscope, and the third acceleration of the multirotor aircraft in the body coordinate system is measured by the accelerometer. The third rotational transformation matrix from the geodetic coordinate system to the body coordinate system of the multirotor aircraft is obtained for the third attitude angle at that sampling moment. The formula for this third rotational transformation matrix is ​​similar to that of the first rotational transformation matrix and will not be repeated here. Based on the mass of the multirotor aircraft, the third rotational transformation matrix, and the third acceleration, the third force information of the multirotor aircraft is determined. The formula for this third force information is similar to that of the first force information and will not be repeated here.

[0173] Next, the second airspeed is obtained in the same manner as the first airspeed, and the third rotor speed is obtained in the same manner as the second rotor speed.

[0174] Then, a test sample set is determined based on the third force information, the third rotor speed, and the second airspeed. Specifically, for each sampling moment, principal component analysis (PCA) is performed on the PCA network based on the third rotor speed and the third force information to obtain a third analysis result. This third analysis result includes multiple third principal components and the corresponding third contribution rate for each third principal component. A third target principal component is determined from the multiple third principal components based on the third contribution rate. The process of PCA is the same as that described in the previous embodiment, and the process of determining the third target principal component is the same as that of determining the second target principal component; therefore, it will not be repeated here. Finally, the test sample set is determined based on the third target principal component and the second airspeed. Specifically, the third target principal component and the second airspeed at each sampling moment are used as a test sample to form the test sample set.

[0175] By acquiring the third force information of the multi-rotor aircraft at the sampling time; then acquiring the second airspeed of the multi-rotor aircraft at the sampling time; and then acquiring the third rotor speed of each rotor in the multi-rotor aircraft at the sampling time; and then determining the test sample set based on the third force information, the third rotor speed, and the second airspeed, the test sample set can be accurately obtained, improving the accuracy of determining the correction coefficient by testing the magic network model through the test sample set, and further improving the accuracy of airspeed prediction through the magic network model.

[0176] In addition, this application also proposes a multi-rotor aircraft, referring to... Figure 3 The multi-rotor aircraft includes:

[0177] The acquisition module 10 is used to acquire the current first force information of the multi-rotor aircraft and acquire the first rotor speed of each rotor in the multi-rotor aircraft.

[0178] Prediction module 20 is used to predict airspeed based on the first rotor speed and the first force information by using a neural network model to obtain the predicted airspeed.

[0179] The correction module 30 is used to correct the predicted airspeed based on the correction coefficients corresponding to the neural network model in order to obtain the target airspeed.

[0180] The methods executed by the above-mentioned program units can be referred to in the various embodiments of the airspeed prediction method of this application, and will not be repeated here.

[0181] Furthermore, this application also proposes a computer-readable storage medium storing an airspeed prediction program, which, when executed by a processor, implements the steps of the airspeed prediction method as described above.

[0182] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0183] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0184] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0185] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An airspeed prediction method characterized by, Applied to a multi-rotor aircraft, comprising: obtaining the current first force information of the multi-rotor aircraft, and obtaining the first rotor speed of each rotor in the multi-rotor aircraft; based on the first rotor speed and the first force information, the airspeed is predicted by a neural network model to obtain a predicted airspeed; based on the correction coefficient corresponding to the neural network model, the predicted airspeed is corrected to obtain a target airspeed; Before the step of obtaining the current first force information of the multi-rotor aircraft, further comprising: input the training sample set into the initial neural network model to obtain the neural network model; input the test sample set into the neural network model to obtain the estimated airspeed at each sampling time, wherein the test sample set includes a plurality of test samples corresponding to the sampling time; obtain the error sequence between the estimated airspeed and the actual airspeed corresponding to each sampling time; if the error sequence passes the K-S test, determine the confidence interval corresponding to the pre-set confidence in the error sequence, and determine the correction coefficient based on the endpoint value of the confidence interval.

2. The airspeed prediction method of claim 1, wherein, The step of obtaining the predicted airspeed by the neural network model based on the first rotor speed and the first force information comprises: input the first rotor speed and the first force information into the principal component analysis network to obtain the first analysis result, wherein the first analysis result includes a plurality of first principal components and a first contribution rate corresponding to each first principal component; determine the first target principal component in the plurality of first principal components based on the first contribution rate; input the first target principal component into the neural network model to obtain the predicted airspeed.

3. The airspeed prediction method of claim 2, wherein, The step of determining the first target principal component in the plurality of first principal components based on the first contribution rate comprises: sort the first principal components in descending order of the first contribution rate to obtain a sorting result; iterate through the sorting result to obtain the currently iterated first principal component; if the sum of the first contribution rates of the iterated principal components is less than the preset contribution rate, add the currently iterated first principal component to the first target principal component, until the sum of the contribution rates of the iterated principal components is greater than or equal to the preset contribution rate, wherein the iterated principal components include the first principal components iterated before the currently iterated principal component.

4. The airspeed prediction method of claim 1, wherein, The step of obtaining the current first force information of the multi-rotor aircraft comprises: obtaining the current first attitude angle and the first acceleration of the multi-rotor aircraft; determine the first force information of the multi-rotor aircraft based on the first attitude angle and the first acceleration.

5. The airspeed prediction method of claim 4, wherein, The step of determining the first force information of the multi-rotor aircraft based on the first attitude angle and the first acceleration comprises: obtaining the first rotation transformation matrix of the first attitude angle from the geodetic coordinate system to the body coordinate system of the multi-rotor aircraft; determine the first force information of the multi-rotor aircraft based on the mass of the multi-rotor aircraft, the first rotation transformation matrix and the first acceleration.

6. The airspeed prediction method of any one of claims 1 to 5, wherein, Before the step of inputting the training sample set into the initial neural network model for model training to obtain the neural network model, the method further comprises the following steps of: obtaining second force information corresponding to the multi-rotor aircraft; obtaining first airspeed corresponding to the multi-rotor aircraft; obtaining second rotor speed of each rotor in the multi-rotor aircraft; determining the training sample set based on the second force information, the second rotor speed and the first airspeed.

7. The airspeed prediction method of claim 6, wherein, The step of determining the training sample set based on the second force information, the second rotor speed and the first airspeed comprises the following steps of: inputting the second rotor speed and the second force information into a principal component analysis network for principal component analysis to obtain a second analysis result, wherein the second analysis result comprises a plurality of second principal components and a second contribution rate corresponding to each second principal component; determining a second target principal component from the plurality of second principal components based on the second contribution rate; determining the training sample set based on the second target principal component and the first airspeed.

8. The airspeed prediction method of claim 6, wherein, The step of obtaining the current second force information of the multi-rotor aircraft comprises the following steps of: obtaining a second attitude angle and a second acceleration of the multi-rotor aircraft; obtaining a second rotation transformation matrix of the second attitude angle from a geodetic coordinate system to a body coordinate system of the multi-rotor aircraft; determining the second force information of the multi-rotor aircraft based on the mass of the multi-rotor aircraft, the second rotation transformation matrix and the second acceleration.

9. The airspeed prediction method of claim 6, wherein, The step of obtaining the first airspeed corresponding to the multi-rotor aircraft comprises the following steps of: obtaining a speed in a geodetic coordinate system corresponding to the multi-rotor aircraft, and an environmental wind speed of an environment in which the multi-rotor aircraft is located; determining the first airspeed based on the speed and the environmental wind speed.

10. The airspeed prediction method of any one of claims 1 to 5, wherein, Before the step of inputting the test sample set into the neural network model for model training to obtain the estimated airspeed at each sampling time, the method further comprises the following steps of: obtaining third force information of the multi-rotor aircraft at a sampling time; obtaining second airspeed of the multi-rotor aircraft at the sampling time; obtaining third rotor speed of each rotor in the multi-rotor aircraft at the sampling time; determining the test sample set based on the third force information, the third rotor speed and the second airspeed.

11. A multicopter characterized by The multi-rotor aircraft comprises: an acquisition module configured to acquire current first force information of the multi-rotor aircraft and first rotor speed of each rotor in the multi-rotor aircraft; a prediction module configured to perform airspeed prediction through a neural network model based on the first rotor speed and the first force information to obtain a predicted airspeed; a correction module configured to perform correction processing on the predicted airspeed based on a correction coefficient corresponding to the neural network model to obtain a target airspeed; The multi-rotor aircraft is further configured to: The training sample set is input into an initial neural network model for model training to obtain the neural network model; the test sample set is input into the neural network model for model training to obtain estimated airspeeds at respective sampling instants, wherein the test sample set comprises test samples corresponding to a plurality of sampling instants; an error sequence between the estimated airspeeds and actual airspeeds corresponding to the respective sampling instants is obtained; if the error sequence passes the K-S test, a confidence interval corresponding to a preset confidence level is determined in the error sequence, and the correction coefficient is determined based on end point values of the confidence interval.

12. An airspeed prediction device, characterized by, The airspeed prediction device comprises a memory, a processor, and an airspeed prediction program stored on the memory and executable on the processor, and the airspeed prediction program, when executed by the processor, implements the steps of the airspeed prediction method according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, The computer readable storage medium stores an airspeed prediction program, and the airspeed prediction program, when executed by the processor, implements the steps of the airspeed prediction method according to any one of claims 1 to 10.

Images