Three-dimensional wind vector estimation method and system based on coherent Doppler laser wind finding radar

Abstract

The invention discloses a three-dimensional wind vector estimation method and system based on a coherent Doppler laser wind finding radar, and the method comprises the steps: collecting radar echo signals containing Doppler frequency shift, and splicing one-dimensional power spectrums of the radar echo signals into a two-dimensional power spectrum; extracting two-dimensional power spectrum peak value characteristics of the radar echo signals; carrying out convolution on the column vector and the two-dimensional power spectrum to obtain a one-dimensional second-order finite difference along a frequency axis; normalizing and binarizing the one-dimensional second-order finite difference, and limiting a peak searching range; determining the peak value position of each range gate, judging whether each peak value position is located in a peak searching range or not, giving a weight according to a signal-to-noise ratio, and fitting a three-dimensional wind vector by using an iterative weighted least square method; according to the method, constraint is applied to the peak frequency of the radar echo signal power spectrum, and the measured radial wind speed is subjected to iterative weighted fitting to obtain the three-dimensional wind vector, so that the precision and reliability of wind field inversion are improved.

Description

Technical Field

[0001] The invention relates to the technical field of laser wind radar, and in particular to a three-dimensional wind vector estimation method and system based on a coherent Doppler laser wind radar. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Real-time monitoring of atmospheric wind fields plays an important role in severe weather warning, aviation and aerospace safety, and improving wind energy utilization. Traditional wind measuring devices such as wind towers and sounding balloons have met the basic needs of wind field measurement to a certain extent. However, the construction and maintenance costs of wind towers are high, and the measurement points are relatively fixed, making it difficult to achieve large-area, dynamic wind field monitoring. Although sounding balloons can obtain wind field information within a certain height range, they are greatly affected by meteorological conditions, the temporal and spatial resolution of the measurements is limited, and continuous, real-time monitoring cannot be achieved. Against this background, coherent Doppler laser wind radar has emerged, which provides a high-precision, non-contact, fast and convenient wind measurement method and is widely used in real-time detection of three-dimensional wind fields in space. Laser wind radar uses the principle of laser Doppler effect to infer the wind speed of atmospheric fluids by detecting the frequency changes of the laser beam during backscattering.

[0004] Coherent Doppler lidar faces many challenges during the detection process. On the one hand, its echo signal is easily interfered by various factors such as ground clutter, atmospheric turbulence and internal noise of the system. On the other hand, as the detection altitude continues to increase, its echo signal will gradually weaken, which greatly increases the difficulty of detecting the frequency shift of the echo signal in some areas.

[0005] In the existing technology based on coherent Doppler laser wind radar, most of the work is carried out in three-beam, four-beam and five-beam velocity azimuth display (VAD) scanning mode, and the required atmospheric three-dimensional wind field is obtained by fitting the radial velocity of each beam direction. Although the commonly used wind vector fitting methods such as vector synthesis method and least square fitting can realize the calculation of wind vector to a certain extent, there is a serious problem, that is, they are overly dependent on the accurate detection of radial wind speed of each beam. However, in actual situations, due to the interference of various noises on the echo signal, the measured radial velocity often has errors, and these traditional fitting methods do not fully consider this point, resulting in the wind vector in the area with low signal-to-noise ratio. The final fitted wind vector often has a large error. This error not only reduces the accuracy and reliability of wind field inversion, causing the obtained wind field information to deviate from the actual situation, but also at a certain height, due to the accumulation of errors and the serious decline in signal quality, it may even lead to the appearance of data fault phenomenon, which seriously affects the comprehensive and accurate monitoring of the atmospheric wind field. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a three-dimensional wind vector estimation method and system based on coherent Doppler laser wind measuring radar, which improves the accuracy and reliability of wind field inversion by imposing constraints on the peak frequency of the radar echo signal power spectrum and iteratively weighted fitting of the three-dimensional wind vector of the measured radial wind speed.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a three-dimensional wind vector estimation method based on a coherent Doppler laser wind radar, comprising:

[0009] Collect radar echo signals containing Doppler frequency shift, and splice the one-dimensional power spectrum of the radar echo signal into a two-dimensional power spectrum; extract the peak characteristics of the two-dimensional power spectrum of the radar echo signal; use the column vector to convolve with the two-dimensional power spectrum to obtain a one-dimensional second-order finite difference along the frequency axis; then normalize and binarize the one-dimensional second-order finite difference, and limit the peak search range;

[0010] Determine the peak position of each range gate, judge whether each peak position is within the peak search range, and assign weights according to the signal-to-noise ratio; and use iterative weighted least squares method to fit the three-dimensional wind vector according to the weights.

[0011] As a further technical solution, the one-dimensional power spectrum of the radar echo signal is spliced ​​into a two-dimensional power spectrum. The specific method is: the radar echo signal is converted into a frequency domain signal by using a fast Fourier transform, and the power of each frequency component in the frequency domain signal is calculated to obtain the power spectrum density of the radar echo signal; the one-dimensional power spectrum of the radar echo signal is cut into units of range gates, and the cut one-dimensional power spectra of each range gate are longitudinally spliced ​​into a two-dimensional power spectrum.

[0012] As a further technical solution, the specific method for obtaining the one-dimensional second-order finite difference along the frequency axis is: Among them, H D Represents a column vector, PSD G Represents the two-dimensional power spectrum, PSD D represents one-dimensional second-order finite differences.

[0013] As a further technical solution, the specific method for limiting the peak search range is: first, the one-dimensional second-order finite difference is normalized, that is, PSD N (j,k)=[PSD D (j,k)-m N (k)] / σ N (k); where m N (k) is the average value, σ N (k) is the standard deviation of the kth one-dimensional second-order finite difference; then the threshold is determined by using the triangle rule and the normalized one-dimensional second-order finite difference is globally binarized. In the results of the global binarization, the results greater than or equal to the threshold are limited to the peak search range.

[0014] As a further technical solution, the specific method of assigning weights is as follows: the radial wind speed inverted from the peak within the peak-seeking range has a specific weight value determined by the signal-to-noise ratio of the range gates at the same height in four different radial directions, the highest of which is set to 1, and the rest are set according to the proportion; the peak that is not within the peak-seeking range is discarded and replaced by the peak of the corresponding range gate in the one-dimensional second-order finite difference to invert the radial wind speed, and its weight is set equal to the lowest radial wind speed weight of the other three radial range gates at the same height.

[0015] As a further technical solution, according to the weights, an iterative weighted least squares method is used to fit the three-dimensional wind vector. The specific method is: using the least squares method, a wind vector is determined so that the sum of the projection of the wind vector in each radial direction and the second-order residual of the measured radial velocity is minimized, that is, the following function is minimized: Where p represents the number of beams, Indicates radial wind speed, V k represents the determined wind vector, S j is the unit vector in each radial direction;

[0016] Add the weights to the function, ie: Among them, w j,k Represents weight.

[0017] As a further technical solution, the function is decomposed into:

[0018] in, Represents the three-dimensional wind vector.

[0019] In a second aspect, the present invention provides a three-dimensional wind vector estimation system based on coherent Doppler laser wind radar, comprising the following modules:

[0020] The peak search module is configured to: collect radar echo signals containing Doppler frequency shift, and splice the one-dimensional power spectrum of the radar echo signal into a two-dimensional power spectrum; extract the peak characteristics of the two-dimensional power spectrum of the radar echo signal; use the column vector to convolve with the two-dimensional power spectrum to obtain a one-dimensional second-order finite difference along the frequency axis; then normalize and binarize the one-dimensional second-order finite difference, and limit the peak search range;

[0021] The three-dimensional wind vector estimation module is configured as follows: according to the signal-to-noise ratio of each range gate of the original power spectrum of the radar echo signal, the peak position of each range gate is determined, whether each peak position is within the peak search range, and a weight is assigned according to the signal-to-noise ratio; according to the weight, the three-dimensional wind vector is fitted using the iterative weighted least squares method.

[0022] One or more technical solutions of the present invention have the following beneficial effects:

[0023] (1) The peak-finding accuracy of the present invention is high. The key to the laser wind radar detecting radial wind speed is whether the peak of the echo signal power spectrum can be accurately found. In the low signal-to-noise ratio area, the peak of the echo signal power spectrum is not prominent enough, and the peak-finding effect is not ideal. The present invention is based on the spatial continuity of wind speed, that is, the detection results between adjacent range gates are unlikely to vary. This is reflected in the two-dimensional power spectrum. The peak positions of each range gate are also adjacent and similar. The convolutional neural network is used to extract the peak characteristics of the two-dimensional power spectrum of the radar echo signal, which not only highlights the relatively obvious peaks, but also makes the inconspicuous or even hidden peaks prominent due to the proximity of the peak positions. The subsequent one-dimensional second-order finite difference and normalization further improve the signal-to-noise ratio, so that the success rate of peak-finding is improved.

[0024] (2) The wind vector detection accuracy of the present invention is high. The working mode is a 3-5 beam VAD scanning laser wind measuring radar. When fitting the wind vector, the traditional scheme often uses vector synthesis method, least square fitting method, etc., which often causes errors in the fitted three-dimensional wind vector due to the addition of erroneous radial velocity in the low signal-to-noise ratio area. The present invention sets a peak search range when extracting the power spectrum peak, avoiding the influence of completely erroneous radial wind speed, and uses the least square method during fitting and assigns different weights according to the signal-to-noise ratio, further reducing the error of the three-dimensional wind vector caused by the erroneous radial wind speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0026] Figure 1 It is a peak finding flow chart of the three-dimensional wind vector estimation method of the present invention;

[0027] Figure 2 It is a flow chart of the estimation process in the three-dimensional wind vector estimation method of the present invention; DETAILED DESCRIPTION

[0028] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0029] Embodiment 1

[0030] The present invention provides a three-dimensional wind vector estimation method based on a 1550nm all-fiber coherent Doppler laser wind radar, which fully considers the spatial continuity of wind speed, uses the characteristics of adjacent or similar positions of two-dimensional power spectrum peaks for data analysis and processing, and uses one-dimensional second-order finite differences, normalization and triangle method (Triangle Method) binarization to constrain the peak search range and improve the success rate of power spectrum peak extraction. The specific steps are as follows:

[0031] Step S1: collecting radar echo signals containing Doppler frequency shift received by a radar receiving module, and splicing the one-dimensional power spectrum of the radar echo signal into a two-dimensional power spectrum.

[0032] In step S1, it specifically includes:

[0033] S11, converting the radar echo signal into a frequency domain signal using fast Fourier transform (FFT);

[0034] S12, calculating the power of each frequency component in the frequency domain signal to obtain the one-dimensional power spectrum density of the radar echo signal, and preliminarily reducing the power spectrum noise through pulse accumulation and averaging;

[0035] S13, cutting the one-dimensional power spectrum of the radar echo signal into units of range gates, and vertically splicing the cut one-dimensional power spectra of each range gate into a two-dimensional power spectrum.

[0036] Step S2: using a convolutional neural network to extract the peak features of the two-dimensional power spectrum of the radar echo signal.

[0037] In step S2, it specifically includes:

[0038] S21. Construct a convolutional neural network for feature extraction, which includes an input layer, two convolutional layers, two fully connected layers, four ReLU activation layers and a regression layer.

[0039] S22, selecting the two-dimensional power spectrum with lower noise and clearer peak features in step S1 as training data;

[0040] S23, specify the training options, the maximum number of training iterations is 20, the size of each batch is 64, and the initial learning rate is 0.001;

[0041] S24, training convolutional neural network;

[0042] S25. Use the trained convolutional neural network to reconstruct the input two-dimensional power spectrum to be tested, highlighting the peak characteristics of the two-dimensional power spectrum.

[0043] Step S3: Using column vector H D Convolving with the two-dimensional power spectrum yields a one-dimensional second-order finite difference along the frequency axis.

[0044] In step S3, it specifically includes:

[0045] S31, the selected column vector is the transpose of the horizontal quantity [-1,2,-1], that is, the column vector H D =[-1,2,-1] T ;

[0046] S32, the two-dimensional power spectrum PSD G With H D Perform convolution to get PSD G One-dimensional (1-D) second-order finite-difference PSD along the frequency axis D :

[0047] Among them, H D Represents a column vector, PSD G Represents the two-dimensional power spectrum reconstructed by the neural network, PSDD represents one-dimensional second-order finite differences.

[0048] Step S4: one-dimensional second-order finite difference PSD D Normalization and binarization are performed, and the peak search range is limited.

[0049] In step S4, it specifically includes:

[0050] S41, normalize the one-dimensional second-order finite difference to 1, that is: PSD N (j,k)=[PSD D (j,k)-m N (k)] / σ N (k); where m N (k) is the average value, σ N (k) is the standard deviation of the kth one-dimensional second-order finite difference;

[0051] S42, use the triangle rule to determine the threshold and perform global binarization on the normalized one-dimensional second-order finite difference, and use PSD to calculate the result. B express;

[0052] Among them, the triangle rule is used to deal with the threshold segmentation problem in image processing. The threshold is determined by applying the triangle rule. Specifically, the normalized one-dimensional second-order finite difference PSD N It is regarded as a grayscale image with a grayscale value range of [0,1], and then the threshold is determined using the triangle rule. The ability of the triangle rule to determine the threshold is recognized as effective in the field of image processing.

[0053] S43. Set PSD B The part greater than or equal to the threshold in S42 is 1, and the part less than the value is 0;

[0054] S44. PSD B The part with a median value of 1 is defined as a reasonable peak search range.

[0055] Step S5, according to the signal-to-noise ratio of each range gate of the original power spectrum of the radar echo signal, determine the peak position of each range gate, judge whether each peak position is within the peak search range, and assign weights according to the signal-to-noise ratio; according to the weights, use the iterative weighted least squares method to fit the three-dimensional wind vector.

[0056] In step S5, it specifically includes:

[0057] S51. Calculate the signal-to-noise ratio of each range gate of the original power spectrum of the radar echo signal, and use the conventional peak-finding method to find the peak position of each range gate; where the signal-to-noise ratio formula is: SNR = Ps / Pn; where Ps is the power of the echo signal, and Pn is the noise power. The peak-finding method is implemented using the peak-finding function findpeaks in Matlab.

[0058] S52, determining whether each peak position is within the peak-seeking range, if the peak is within the peak-seeking range, the radial wind speed inverted therefrom is given a weight according to the calculated signal-to-noise ratio, the specific value of the weight is determined by the signal-to-noise ratio of the range gate at the same height in four different radial directions, the highest value is set to 1, and the rest are set according to the proportion;

[0059] S53, if the peak value is not within the peak-finding range, the peak value is discarded and replaced by the peak value of the corresponding range gate in the one-dimensional second-order finite difference to invert the radial wind speed, and its weight is set equal to the lowest radial wind speed weight of the other three radial range gates at the same height;

[0060] S54. Using the least squares method, a wind vector is found such that the sum of the second-order residuals of the wind vector projections in each radial direction and the measured radial velocity is minimized, that is, the following function is minimized:

[0061]

[0062] Wherein, p is the number of beams, and in this embodiment, p=4. is the radial wind speed, V k is the required wind vector, S j are unit vectors in each radial direction:

[0063] S55, the obtained weight w j,k Add to the above function:

[0064]

[0065] S56. By analyzing the above function, a three-dimensional wind vector is obtained, which is specifically:

[0066] in, Represents the three-dimensional wind vector.

[0067] Embodiment 2

[0068] This embodiment discloses a three-dimensional wind vector estimation system based on coherent Doppler laser wind radar, including the following modules:

[0069] The peak search module is configured to: collect radar echo signals containing Doppler frequency shift, and splice the one-dimensional power spectrum of the radar echo signal into a two-dimensional power spectrum; extract the peak characteristics of the two-dimensional power spectrum of the radar echo signal; use the column vector to convolve with the two-dimensional power spectrum to obtain a one-dimensional second-order finite difference along the frequency axis; then normalize and binarize the one-dimensional second-order finite difference, and limit the peak search range;

[0070] The three-dimensional wind vector estimation module is configured to: determine the peak position of each range gate, determine whether each peak position is within the peak search range, and assign weights according to the signal-to-noise ratio; and fit the three-dimensional wind vector using iterative weighted least squares method according to the weights.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A three-dimensional wind vector estimation method based on coherent Doppler laser wind radar, characterized in that: include: Collecting radar echo signals containing Doppler frequency shift, and splicing the one-dimensional power spectrum of the radar echo signal into a two-dimensional power spectrum; Extract the peak characteristics of the two-dimensional power spectrum of the radar echo signal; The column vector is convolved with the two-dimensional power spectrum to obtain a one-dimensional second-order finite difference along the frequency axis; the one-dimensional second-order finite difference is then normalized and binarized, and the peak search range is limited; Determine the peak position of each range gate, judge whether each peak position is within the peak search range, and assign weights according to the signal-to-noise ratio; and use iterative weighted least squares method to fit the three-dimensional wind vector according to the weights.

2. The three-dimensional wind vector estimation method based on coherent Doppler laser wind radar according to claim 1, characterized in that: The one-dimensional power spectrum of the radar echo signal is spliced ​​into a two-dimensional power spectrum. The specific method is: the radar echo signal is converted into a frequency domain signal by using a fast Fourier transform, and the power of each frequency component in the frequency domain signal is calculated to obtain the power spectrum density of the radar echo signal; the one-dimensional power spectrum of the radar echo signal is cut into units of range gates, and the cut one-dimensional power spectra of each range gate are longitudinally spliced ​​into a two-dimensional power spectrum.

3. The three-dimensional wind vector estimation method based on coherent Doppler laser wind radar according to claim 1, characterized in that: The specific method to obtain the one-dimensional second-order finite difference along the frequency axis is: Among them, H D Represents a column vector, PSD G Represents the two-dimensional power spectrum, PSD D represents one-dimensional second-order finite differences.

4. The three-dimensional wind vector estimation method based on coherent Doppler laser wind radar according to claim 1, characterized in that: The specific method to limit the peak search range is: first normalize the one-dimensional second-order finite difference, that is: PSD N (j,k)=[PSD D (j,k)-m N (k)] / σ N (k); where m N (k) is the average value, σ N (k) is the standard deviation of the kth one-dimensional second-order finite difference; then the threshold is determined by using the triangle rule and the normalized one-dimensional second-order finite difference is globally binarized. In the results of the global binarization, the results greater than or equal to the threshold are limited to the peak search range.

5. The three-dimensional wind vector estimation method based on coherent Doppler laser wind radar according to claim 1, characterized in that: The specific method of assigning weights is as follows: for the peak value within the peak-seeking range, the radial wind speed inverted from it has a specific weight value determined by the signal-to-noise ratio of the range gates at the same height in four different radial directions, the highest of which is set to 1, and the rest are set according to the proportion; for the peak value that is not within the peak-seeking range, the peak value is discarded and replaced by the peak value of the corresponding range gate in the one-dimensional second-order finite difference to invert the radial wind speed, and its weight is set equal to the lowest radial wind speed weight of the other three radial range gates at the same height.

6. The three-dimensional wind vector estimation method based on coherent Doppler laser wind radar according to claim 1, characterized in that: According to the weights, the three-dimensional wind vector is fitted using iterative weighted least squares method. The specific method is: using the least squares method, a wind vector is determined so that the sum of the projection of the wind vector in each radial direction and the second-order residual of the measured radial velocity is minimized, that is, the following function is minimized: Where p represents the number of beams, Indicates radial wind speed, V k represents the determined wind vector, S j is the unit vector in each radial direction; Add the weights to the function, ie: Among them, w j,k Represents weight;.

7. The three-dimensional wind vector estimation method based on coherent Doppler laser wind radar according to claim 6, characterized in that: The function is decomposed into: in, Represents the three-dimensional wind vector.

8. A three-dimensional wind vector estimation system based on coherent Doppler laser wind radar, characterized in that: Includes the following modules: The peak-finding module is configured to: collect radar echo signals containing Doppler frequency shift, and splice the one-dimensional power spectrum of the radar echo signal into a two-dimensional power spectrum; extract the peak characteristics of the two-dimensional power spectrum of the radar echo signal; The column vector is convolved with the two-dimensional power spectrum to obtain a one-dimensional second-order finite difference along the frequency axis; the one-dimensional second-order finite difference is then normalized and binarized, and the peak search range is limited; The three-dimensional wind vector estimation module is configured to: determine the peak position of each range gate, determine whether each peak position is within the peak search range, and assign weights according to the signal-to-noise ratio; and fit the three-dimensional wind vector using iterative weighted least squares method according to the weights.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the three-dimensional wind vector estimation method based on coherent Doppler laser wind radar as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the three-dimensional wind vector estimation method based on coherent Doppler laser wind radar as described in any one of claims 1 to 7 are implemented.

Images