Unmanned aerial vehicle complex vector wind field sensing method and system

Abstract

The invention relates to the technical field of unmanned aerial vehicle sensing, and discloses an unmanned aerial vehicle complex vector wind field sensing method and system, and the method comprises the steps: respectively recording the original data of a wind field sensing device in a wind tunnel simulation wind field test and the original data of an unmanned aerial vehicle carrying the wind field sensing device for a flight test, and constructing a feature data set; using the constructed feature data set to train an LSTM model, so that the LSTM model can predict wind field information at the current moment in real time according to historical wind field data, installing a wind field sensing device array on the unmanned aerial vehicle, collecting wind field information data in real time, and using the trained LSTM model to predict the wind field information in real time; and performing data fusion on prediction results of the plurality of sensors by adopting a Kalman filtering algorithm, and calculating wind field information such as three-dimensional wind speed, wind-induced air pressure change and turbulence. According to the method, real-time efficient prediction of the complex wind field environment in the air of the unmanned aerial vehicle is realized, and an effective means is provided for improving wind field detection and reliable operation of various unmanned aerial vehicles in the complex wind field environment.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned aerial vehicle sensing technology, and in particular relates to a method and system for sensing complex vector wind fields of unmanned aerial vehicles. Background Art

[0002] With the rapid development of UAV technology and its wide application in military reconnaissance, agricultural monitoring, logistics distribution, environmental monitoring and other fields, UAVs need to perform tasks in various complex and changing environments. Especially under complex wind field conditions such as strong winds and turbulence, how to accurately perceive the surrounding environment and make corresponding adjustments has become one of the key issues to improve the performance of UAVs. At present, although the traditional wind speed sensors commonly used by UAVs (such as pitot tubes or hot wire anemometers) can provide basic wind speed information, their response speed is slow and the accuracy is limited, which makes it difficult to meet the precise measurement requirements under real-time dynamically changing wind field conditions. In addition, these sensors can usually only provide wind speed data at local points and cannot fully reflect the complex changes of three-dimensional wind fields, which greatly limits the autonomous navigation capability and mission execution efficiency of UAVs in complex wind fields.

[0003] In recent years, with the advancement of artificial intelligence and machine learning technologies, researchers have begun to explore the use of advanced algorithms to improve the perception of complex wind field environments by UAVs. For example, by combining data from multiple sensors and using deep learning models to predict the changing trend of wind fields. Although these methods have certain potential, they still face many challenges in practical applications, such as the difficulty in obtaining high-quality data sets required for model training and the lack of real-time performance due to high consumption of computing resources. In addition, how to effectively fuse information from sensors at different locations to obtain a more accurate description of the wind field is also a hot topic in current research. Therefore, it is particularly important to develop a complex vector wind field perception equipment and method for UAVs that can ensure high accuracy and achieve rapid response, and can effectively handle multi-source information fusion. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a complex vector wind field perception equipment and method for UAV based on the combination of LSTM model and Kalman filter algorithm. By constructing a feature data set, training a prediction model and performing real-time data analysis, it realizes the effective prediction of the three-dimensional wind speed, wind-induced pressure changes and turbulence around the UAV, thereby improving the operational safety and efficiency of the UAV in complex wind fields.

[0005] A method for sensing complex vector wind fields of unmanned aerial vehicles, comprising the following steps:

[0006] S1. Dataset construction: Record the original data of the wind field sensor device in the wind tunnel simulated wind field test and the UAV equipped with the wind field sensor device for flight test, and construct a feature data set;

[0007] S2. Model training: Use the constructed feature data set to train the LSTM model so that it can predict the wind field information at the current moment in real time based on the historical wind field data;

[0008] S3. Real-time prediction: Install a wind field sensor array on the drone to collect wind field information data in real time, and use the trained LSTM model to make real-time predictions of wind field information;

[0009] S4. Post-processing and data fusion: The Kalman filter algorithm is used to fuse the prediction results of multiple sensors to calculate wind field information such as three-dimensional wind speed, wind-induced pressure changes and turbulence.

[0010] As a preferred embodiment of the present invention, in step S1, the wind tunnel simulates a variety of changing wind fields including constant wind fields, step-changing wind fields, and sinusoidal changing wind fields under various wind speed and wind direction combinations, and the actual flight of the UAV includes a variety of flight conditions including UAV take-off and landing, hovering, straight flight, and circling flight.

[0011] As a preferred embodiment of the present invention, the original data of the wind field sensing device is time series data, wherein each moment includes a sampling moment, an n-dimensional air pressure value vector, an n-dimensional temperature value vector, a 4-dimensional real value vector, a 3-dimensional wind speed vector and a wind-induced air pressure change value.

[0012] As a preferred implementation of the present invention, in step S1, data preprocessing includes sample segmentation, outlier detection, data smoothing and feature extraction, and the specific steps are as follows:

[0013] (1) The original data measured in different scenarios are summarized into different sub-samples. For sub-samples with too large data volume, they are further divided into several smaller sub-samples;

[0014] (2) Based on the 3δ method, the overall mean μ and standard deviation δ of each subsample are calculated, and all data points falling outside the range (μ-3δ, μ+3δ) are marked as outliers;

[0015] (3) Using interpolation method, replace the outliers marked in step (1);

[0016] (4) Use the sliding window method to convert time series data into feature data sets.

[0017] As a preferred embodiment of the present invention, in step S2, a back propagation algorithm is used in the training process of the LSTM model, wherein the loss function L is calculated as follows:

[0018]

[0019] Where: N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample.

[0020] As a preferred implementation of the present invention, in step S3, the input of the LSTM model is a [W, 2n] matrix, which is generated by a sliding window method with a size of W, and each time step contains an n-dimensional pressure value vector and an n-dimensional temperature value vector; the LSTM model outputs a 4-dimensional vector, which is further decomposed into a three-dimensional wind speed component z and a wind-induced pressure change value d.

[0021] As a preferred implementation of the present invention, in step S4, the updating steps of the Kalman filter algorithm are as follows:

[0022] predict:

[0023]

[0024] renew:

[0025]

[0026] in, and P k|k-1 are the prior prediction values ​​of the state estimate and covariance matrix respectively; and P k|k are the correction values; F k is the state transfer matrix; Q k is the process noise covariance matrix; H k is the observation matrix; R k is the observation noise covariance matrix; K k is the Kalman gain; z k is the three-dimensional wind speed observation value, and T represents the matrix transpose.

[0027] As a preferred embodiment of the present invention, in step S4, the wind field information includes the smooth wind speed U, the pulsating wind speed u', the wind-induced pressure change value d and the turbulence degree I u , the specific calculation steps are as follows:

[0028] Calculate the smoothed wind speed U of each wind field sensor device separately i , pulsating wind speed u' i and weight w i , where: i = 1, 2, ..., m represents the i-th sensor device:

[0029]

[0030]

[0031] Calculate the fused smooth wind speed U, fused fluctuating wind speed u' and wind-induced pressure change d: where d i Represents the wind-induced pressure change value output by the i-th sensor device:

[0032]

[0033] Calculation of turbulence I u :

[0034]

[0035] Among them U min is the preset threshold.

[0036] The present invention also discloses a complex vector wind field perception system for unmanned aerial vehicles, comprising:

[0037] Drone: as the carrier of the entire system;

[0038] Wind field sensor array: It consists of m wind field sensor devices, which are arranged in a centrally symmetrical form around the yaw axis of the aircraft body and installed on the UAV. They are arranged in the area outside the UAV where the total interference of the rotor wind field is the smallest. This area is determined by fluid dynamics simulation; it is used to collect wind field data in the environment in real time;

[0039] Wind field information calculation unit: installed on the UAV and connected to the wind field sensor array, it is used to process and calculate the data from the wind speed sensor to achieve real-time monitoring and prediction of complex vector wind fields.

[0040] As a preferred implementation of the present invention, the wind field information solving unit includes:

[0041] A communication module, used to read raw data from the wind farm sensor array in real time;

[0042] Computation module: used to realize the complex vector wind field perception method of the UAV and obtain wind field information data;

[0043] Storage module: used to store the original data collected from the wind field sensor array, any algorithm program required by the UAV complex vector wind field perception method, and the wind field information data obtained by the method;

[0044] Power supply module: used to provide a stable operating voltage for the communication module, computing module, storage module and other related module circuits, and ensure that the wind farm sensor array continues to obtain the necessary power support.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] (1) The present invention combines LSTM and Kalman filtering methods to more accurately calculate the wind field information at the current moment based on historical wind field data, thereby improving the adaptability of drones to complex wind field environments and the accuracy of complex wind field measurements;

[0047] (2) The present invention can also provide additional key wind field parameters such as wind-induced pressure changes and turbulence, providing a new means for the analysis and detection of dynamic wind fields and the precise control of UAVs.

[0048] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In the attached picture:

[0050] Figure 1 This is a flow chart of the complex vector wind field perception method of the UAV of the present invention.

[0051] Figure 2 This is a structural diagram of the LSTM model in Example 1 of the present invention.

[0052] Figure 3 This is a structural diagram of the wind field information calculation method in Example 1 of the present invention.

[0053] Figure 4 This is a structural diagram of the complex vector wind field perception system of the UAV of the present invention.

[0054] Figure 5 This is a schematic diagram of the installation of the wind field sensor array in Example 2 of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The following embodiments are used to illustrate the present invention.

[0056] Example 1

[0057] like Figure 1 As shown, the complex vector wind field perception method of the UAV of the present invention comprises the following steps:

[0058] S1. Dataset construction: Record the original data of the wind field sensor device in the wind tunnel simulated wind field test and the UAV equipped with the wind field sensor device for flight test, and construct a feature data set;

[0059] S2. Model training: Use the constructed feature data set to train the LSTM model so that it can predict the wind field information at the current moment in real time based on the historical wind field data;

[0060] S3. Real-time prediction: Install a wind field sensor array on the drone to collect wind field information data in real time, and use the trained LSTM model to make real-time predictions of wind field information;

[0061] S4. Post-processing and data fusion: The Kalman filter algorithm is used to fuse the prediction results of multiple sensors to calculate wind field information such as three-dimensional wind speed, wind-induced pressure changes and turbulence.

[0062] The wind tunnel simulated wind field in step S1 includes various wind fields such as constant wind field, step change wind field and sinusoidal change wind field under various wind speed and wind direction combinations, and the actual flight of the UAV includes various flight conditions such as UAV take-off and landing, hovering, straight flight and circling flight; in this embodiment, the wind speed range of the wind tunnel simulated wind field is 0-10m / s, the spatial wind direction resolution is 15°, and the wind field frequency range is 0-5Hz. The actual flight adopts ZD550 quad-rotor UAV, and the horizontal speed range is 0-5m / s;

[0063] The original data of the wind field sensor device is time series data, in which each moment contains: sampling moment; an n-dimensional pressure value vector; an n-dimensional temperature value vector; a 4-dimensional real value vector, including a 3-dimensional wind speed vector and a wind-induced pressure change value. In this embodiment, the wind field sensor device includes 8 MEMS pressure sensors, n = 8;

[0064] In this embodiment, the data preprocessing in step S1 includes sample segmentation, outlier detection, data smoothing and feature extraction. The specific steps are as follows:

[0065] (1) The original data measured in different scenarios are summarized into [1000, 2000] sub-samples. For samples with data volume K < 1000, they are discarded; for sub-samples with data volume between 1000 ≤ K ≤ 2000, they are retained; for sub-samples with data volume between K > 2000, they are separated into two sub-samples with data volumes of 1000 and K-1000 respectively, and the above operation is repeated for the latter;

[0066] (2) Based on the 3δ method, the overall mean μ and standard deviation δ of each subsample are calculated, and all data points falling outside the range (μ-3δ, μ+3δ) are marked as outliers;

[0067] (3) Using the linear interpolation method, replace the outlier value y at the sampling time t marked in step (1) with:

[0068]

[0069] Where: y k-1 ,y k+1 are the data values ​​of the previous moment and the next moment respectively, tk-1 ,t k+1 are the sampling moments of the previous moment and the next moment respectively

[0070] (4) Using the sliding window method, with a window size of W = 10, the time series data is converted into a feature data set;

[0071] The back propagation algorithm is used in the training process of the LSTM model in step S2, and the loss function L is defined as the mean square error (MSE):

[0072] Where: N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample;

[0073] In this embodiment, the input of the LSTM model in step S3 is a [W, 2n] matrix, which is generated by a sliding window method with a size of W, and each time step contains an n-dimensional pressure value vector and an n-dimensional temperature value vector; the LSTM model outputs a 4-dimensional vector, which is further decomposed into a three-dimensional wind speed component z and a wind-induced pressure change value d.

[0074] In this embodiment, the updating steps of the Kalman filter algorithm in step S4 are as follows:

[0075] (1) Prediction:

[0076]

[0077] (2) Update:

[0078]

[0079] in, and P k|k-1 are the prior prediction values ​​of the state estimate and covariance matrix respectively; and P k|k are the correction values; F k is the state transfer matrix; Q k is the process noise covariance matrix; H k is the observation matrix; R k is the observation noise covariance matrix; K k is the Kalman gain; z k is the three-dimensional wind speed observation value.

[0080] In this embodiment, Q k and R k The steps for determining the value of are as follows: record the wind field sensor device remaining stationary for 30 seconds in a windless environment, and record the output time series of the LSTM model; for each component of the three-dimensional wind speed in the time series, calculate the variance σi ,i=1,2,3; take R k is a 3rd-order diagonal matrix, and each diagonal element is σ i ; Take Q k =I3*0.01, where I3 is the third-order unit matrix.

[0081] In this embodiment, the wind field information in step S4 includes the smooth wind speed U, the pulsating wind speed u', the wind-induced pressure change value d and the turbulence degree I u , the specific steps are as follows:

[0082] (1) Calculate the smoothed wind speed U of each wind field sensor device separately i , pulsating wind speed u' i and weight w i , where: i = 1, 2, ..., m represents the i-th sensor device:

[0083]

[0084] (2) Calculate the fused smooth wind speed U, fused fluctuating wind speed u' and wind-induced pressure change d: where d i Represents the wind-induced pressure change value output by the i-th sensor device:

[0085]

[0086] (3) Calculation of turbulence intensity I u In order to avoid abnormal turbulence calculation results when the smooth wind speed is too small U, a threshold U is introduced min :

[0087] In this embodiment, take U min =0.5, indicating that turbulence is not calculated when the spatial wind speed is less than 0.5m / s;

[0088] Example 2

[0089] like Figure 4 As shown, the complex vector wind field perception system of the UAV of the present invention can realize the efficient perception and analysis of the complex vector wind field of the UAV, including:

[0090] (1) UAV: ​​as the carrier of the entire system;

[0091] (2) Wind field sensor array: installed on the drone to collect wind field data in the environment in real time;

[0092] (3) Wind field information calculation unit: installed on the UAV and connected to the wind field sensor array, used to process and calculate the data from the wind speed sensor device to achieve real-time monitoring and prediction of complex vector wind fields.

[0093] like Figure 5 As shown, in this embodiment, the UAV adopts a ZD550 four-rotor UAV, and the wind field sensor array consists of m wind field sensor devices, which are arranged above the UAV, arranged around the yaw axis of the fuselage in a centrally symmetrical form, and arranged in the area outside the UAV where the total interference from the rotor wind field is the smallest, and the area is determined by fluid dynamics simulation; all wind field sensor devices are electrically connected to the UAV wind field information solution unit through the SPI bus.

[0094] The wind field information solving unit comprises:

[0095] (1) a communication module, used to read raw data from the wind farm sensor array in real time;

[0096] (2) Operation module: used to implement the UAV complex vector wind field perception method and obtain wind field information data;

[0097] (3) Storage module: used to store the raw data collected from the wind field sensor array, any algorithm program required by the UAV complex vector wind field perception method, and the wind field information data obtained by the method;

[0098] (4) Power module: used to provide a stable operating voltage for the communication module, computing module, storage module and other related module circuits, and ensure that the wind farm sensor array continues to obtain the necessary power support;

[0099] The above is only a preferred implementation case of the present disclosure. Although the present disclosure is described in conjunction with the accompanying drawings, the purpose is not to limit the present disclosure. For those skilled in the art, the present disclosure may have various changes and modifications. Any modification, replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A complex vector wind field perception method for unmanned aerial vehicles, characterized in that: The following steps are involved: S1. Dataset construction: Record the original data of the wind field sensor device in the wind tunnel simulated wind field test and the UAV equipped with the wind field sensor device for flight test, and construct a feature data set; S2. Model training: Use the constructed feature data set to train the LSTM model so that it can predict the wind field information at the current moment in real time based on the historical wind field data; S3. Real-time prediction: Install a wind field sensor array on the drone to collect wind field information data in real time, and use the trained LSTM model to make real-time predictions of wind field information; S4. Post-processing and data fusion: The Kalman filter algorithm is used to fuse the prediction results of multiple sensors to calculate wind field information such as three-dimensional wind speed, wind-induced pressure changes and turbulence.

2. The complex vector wind field sensing method for unmanned aerial vehicles according to claim 1 is characterized in that: In step S1, the wind tunnel simulates a variety of wind fields including constant wind fields, step-change wind fields, and sinusoidal change wind fields under various wind speed and wind direction combinations, and the actual flight of the UAV includes a variety of flight conditions including take-off and landing, hovering, straight flight, and circling flight.

3. The method for sensing complex vector wind fields of unmanned aerial vehicles according to claim 1, characterized in that: The original data of the wind field sensor device is time series data, wherein each moment includes a sampling moment, an n-dimensional air pressure value vector, an n-dimensional temperature value vector, a 4-dimensional true value vector, a 3-dimensional wind speed vector and a wind-induced air pressure change value.

4. The method for sensing complex vector wind fields of unmanned aerial vehicles according to claim 1, characterized in that: In step S1, data preprocessing includes sample segmentation, outlier detection, data smoothing and feature extraction. The specific steps are as follows: (1) The original data measured in different scenarios are summarized into different sub-samples. For sub-samples with too large data volume, they are further divided into several smaller sub-samples; (2) Based on the 3δ method, the overall mean μ and standard deviation δ of each subsample are calculated, and all data points falling outside the range (μ-3δ, μ+3δ) are marked as outliers; (3) Using interpolation method, replace the outliers marked in step (1); (4) Use the sliding window method to convert time series data into feature data sets.

5. The method for sensing complex vector wind fields of unmanned aerial vehicles according to claim 1, characterized in that: In step S2, the back propagation algorithm is used in the training process of the LSTM model, wherein the loss function L is calculated as follows: Where: N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the i-th sample.

6. The complex vector wind field sensing method for unmanned aerial vehicles according to claim 1 is characterized in that: In step S3, the input of the LSTM model is a [W, 2n] matrix, which is generated by a sliding window method with a size of W, and each time step contains an n-dimensional pressure value vector and an n-dimensional temperature value vector; the LSTM model outputs a 4-dimensional vector, which is further decomposed into a three-dimensional wind speed component z and a wind-induced pressure change value d.

7. The method for sensing complex vector wind fields of unmanned aerial vehicles according to claim 1, characterized in that: In step S4, the updating steps of the Kalman filter algorithm are as follows: predict: renew: P k|k =(I-K k H k )P k|k-1 in, and P k|k-1 are the prior prediction values ​​of the state estimate and covariance matrix respectively; and P k|k are the correction values; F k is the state transfer matrix; Q k is the process noise covariance matrix; H k is the observation matrix; R k is the observation noise covariance matrix; K k is the Kalman gain; z k is the three-dimensional wind speed observation value, and T represents the matrix transpose.

8. The method for sensing complex vector wind fields of unmanned aerial vehicles according to claim 1, characterized in that: In step S4, the wind field information includes the smooth wind speed U, the fluctuating wind speed u', the wind-induced pressure change value d and the turbulence degree I u , the specific calculation steps are as follows: Calculate the smoothed wind speed U of each wind field sensor device separately i , pulsating wind speed u' i and weight w i , where: i = 1, 2, ..., m represents the i-th sensor device: Calculate the fused smooth wind speed U, fused fluctuating wind speed u' and wind-induced pressure change d: where d i Represents the wind-induced pressure change value output by the i-th sensor device: Calculation of turbulence I u : Among them U min is the preset threshold.

9. A complex vector wind field perception system for unmanned aerial vehicles, applied to a complex vector wind field perception method for unmanned aerial vehicles as claimed in claim 1, characterized in that: include: Drone: as the carrier of the entire system; Wind field sensor array: It consists of m wind field sensor devices, which are arranged in a centrally symmetrical form around the yaw axis of the aircraft body and installed on the UAV. They are arranged in the area outside the UAV where the total interference of the rotor wind field is the smallest. This area is determined by fluid dynamics simulation; it is used to collect wind field data in the environment in real time; Wind field information calculation unit: installed on the UAV and connected to the wind field sensor array, it is used to process and calculate the data from the wind speed sensor to achieve real-time monitoring and prediction of complex vector wind fields.

10. The complex vector wind field sensing system for unmanned aerial vehicles according to claim 9, characterized in that: The wind field information solving unit comprises: A communication module, used to read raw data from the wind farm sensor array in real time; Computation module: used to realize the complex vector wind field perception method of the UAV and obtain wind field information data; Storage module: used to store the original data collected from the wind field sensor array, any algorithm program required by the UAV complex vector wind field perception method, and the wind field information data obtained by the method; Power supply module: used to provide a stable operating voltage for the communication module, computing module, storage module and other related module circuits, and ensure that the wind farm sensor array continues to obtain the necessary power support.

Images