Laser wind finding radar wind speed extraction method

Abstract

According to the wind speed extraction method of the laser wind finding radar, the end-to-end extraction spectrum peak position of the convolutional neural network is constructed, and the wind speed is calculated, so that multi-step complex calculation processing is avoided; furthermore, deep features of the signals are extracted through deep learning, the applicability and precision of the algorithm are improved, and the weak signal extraction capability under the low signal-to-noise ratio condition is enhanced. According to the invention, the wind speed extraction of the power spectrum data under the condition of low signal-to-noise ratio is realized.

Description

Technical Field

[0001] The invention belongs to the technical field of laser wind radar data processing, and in particular relates to a laser wind radar wind speed extraction method. Background Art

[0002] Laser wind radar has the functions of measuring wind speed, wind direction, wind shear and turbulence intensity. It has high measurement accuracy, strong detection capability, stable operation, convenient use and maintenance. It is widely used in meteorological observation, wind power plant site selection, wind turbine monitoring, aviation safety and other fields. Laser wind radar uses the Doppler effect to measure the Doppler frequency shift between the backscattered light of atmospheric molecules and the emitted laser. The relationship between the Doppler frequency shift and the speed of atmospheric molecule movement (i.e. wind speed) is proportional. The radial wind speed of the detection point can be determined through digital demodulation calculation. The wind field information of the target area to be measured is obtained by processing the radial wind speed information in different directions.

[0003] The wind speed resolution data processing technology of the laser wind radar extracts the wind speed information by demodulating the Doppler frequency shift in the echo signal. The traditional wind speed extraction algorithm first uses fast Fourier transform to obtain the power spectrum, and then extracts the Doppler frequency shift through the peak location algorithm. Commonly used peak search algorithms include derivative method, Gaussian curve fitting method and continuous wavelet transform (CWT), among which the CWT algorithm is more widely used. The basic principle of the CWT peak search method is to use appropriate basis functions and scale parameters to perform continuous wavelet transform on the power spectrum data to obtain the wavelet coefficients at each scale. The local maximum value of each layer of coefficients is the potential peak position, and a longitudinal ridge line will appear. The peak position can be determined by searching the ridge line. The value of the scale parameter of this algorithm is crucial, but because its value range is very wide, it will consume a lot of computing resources in actual measurement, which is not conducive to real-time signal processing. In addition, in the actual measurement environment, the harsh outdoor climate conditions cause a lot of noise in the signal obtained by the system, and the signal-to-noise ratio is very low. In the long-distance wind speed measurement, the noise even drowns the target signal. Traditional wind speed extraction algorithms have been difficult to meet the requirements through complex calculations. In line with the current research trend, deep learning technology is introduced into the processing of wind speed solution data of laser wind radar, and the target power spectrum waveform is fitted end-to-end through the convolutional neural network to solve the wind speed. Summary of the invention

[0004] Based on the above defects of the prior art, the present invention proposes a method for extracting wind speed using a laser wind radar, comprising the following steps:

[0005] (1) Dataset preparation: obtain the simulated dataset DataF and the real dataset DataT;

[0006] The step of acquiring the simulation data set DataF at least comprises: collecting the noise power spectrum P of the laser wind radar under different environmental conditionsNF (f), by comparing with the simulated FFT power spectrum signal P SF0 (f) Synthesize to obtain the analog signal data P with noise SF1 (f) = P SF0 (f)+P NF (f);

[0007] The step of acquiring the real data set DataT at least includes: collecting real signal data, recording the power spectrum signal P of each range gate ST1 (f), and the corresponding true wind speed V, and noise power spectrum P NT (f), the power spectrum of the denoised signal P is calculated SF0 (f) = P SF1 (f)-P NF (f);

[0008] (2) Constructing a power spectrum denoising network: Each layer of the power spectrum denoising network uses a 1*1 convolution kernel, and the convolution kernel consists of an encoder and a decoder; the encoder consists of a convolution operation and a downsampling operation for extracting features; the decoder consists of an upsampling operation, a channel splicing operation, and a convolution operation;

[0009] (3) Constructing a spectrum peak extraction network: The input of the spectrum peak extraction network is an n*1 dimensional vector. First, the input vector is transposed into a 1*n vector, and then processed by a 1*1 convolution layer for dimensionality reduction, and then the spectrum peak position is output;

[0010] (4) Constructing a joint loss to train the network: Joint training is performed based on the power spectrum denoising network and the spectrum peak extraction network, and the total loss function Loss Total It is the weighted sum of Loss1 and Loss2, both Loss1 and Loss2 are SmoothL1 loss functions;

[0011] (5) Wind speed calculation: After the network training is completed, it is directly used for wind speed calculation. First, the collected FFT power spectrum signal P is input. SF0 The frequency value f0 corresponding to the network output spectrum peak is used to calculate the wind speed based on the Doppler frequency shift formula.

[0012] Furthermore, the simulated power spectrum P SF0 (f) Gaussian fitting is adopted, and the Gaussian fitting formula is:

[0013]

[0014] Where λ represents the laser wavelength, which is 1.55*10 -6 m, f0 is the Doppler frequency, which is calculated from the randomly generated wind V, A is a constant representing the amplitude of the power spectrum, and σ is a constant representing the spectral width of the power spectrum.

[0015] Furthermore, the P SF1 (f) and the P SF0 (f) Perform maximum value normalization and map all data to the interval [0, 1],

[0016] The maximum value normalization calculation formula is:

[0017] Furthermore, the convolution operation includes a convolution layer with a convolution kernel of 1*1, a Batch Norm layer and a ReLU activation layer.

[0018] Furthermore, the total loss function calculation formula is:

[0019] Loss Total =0.01Loss1+Loss2

[0020] Among them, Loss1 and Loss2 are both SmoothL1 loss functions; the SmoothL1 loss function is:

[0021]

[0022] Further, the wind speed calculation formula is: wind speed

[0023] The present invention is used to solve the problem of wind speed extraction from power spectrum data under low signal-to-noise ratio conditions. (1) By constructing a convolutional neural network to extract the spectral peak position and solve the wind speed end-to-end, complex calculation processing in multiple steps is avoided; (2) By extracting the deep features of the signal through deep learning, the applicability and accuracy of the algorithm are improved, and the ability to extract weak signals under low signal-to-noise ratio conditions is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0025] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which this application can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by this application, should still fall within the scope of the technical contents disclosed in this application.

[0026] Figure 1 It is a schematic diagram of the power spectrum denoising network structure of the present invention.

[0027] Figure 2 It is a schematic diagram of the spectral peak extraction network structure of the present invention.

[0028] Figure 3 This is a schematic diagram of the joint training of the network loss function of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the embodiments in the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0030] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0031] Example 1

[0032] 1) Dataset preparation: The dataset is divided into two parts: simulated data and real data. Due to the distance limitation of wind speed calibration, it is difficult to obtain low signal-to-noise ratio data at a distance, so simulated data is needed to supplement it.

[0033] Simulated data set DataF: Collect the noise power spectrum P of the laser wind radar under different environmental conditions NF (f), by comparing with the simulated FFT power spectrum signal P SF0 (f) Synthesize to obtain the analog signal data P with noise SF1 (f) = P SF0 (f)+P NF (f), simulated power spectrum P SF0 (f) The Gaussian fitting method is adopted, and the fitting formula is:

[0034]

[0035] Where λ represents the laser wavelength, which is 1.55*10 -6 , f0 is the Doppler frequency, which is calculated from the randomly generated wind V, A is a constant representing the amplitude of the power spectrum, and σ is a constant representing the spectral width of the power spectrum. SF1 (f) and P SF0 (f) Perform extreme value normalization processing and map all data to the interval [0, 1]. The extreme value normalization calculation formula is:

[0036] Real data set DataT: collects real signal data and records the power spectrum signal P of each range gate ST1 (f), and the corresponding true wind speed V, and noise power spectrum P NT (f), the power spectrum of the denoised signal P is calculated SF0 (f) = P SF1 (f)-P NF (f), Doppler frequency calculated based on wind speed P ST1 (F) and P ST0 (f) Perform maximum value normalization and map all data to the interval [0, 1].

[0037] The composition of the entire data set is shown in the following table.

[0038] Table 1 Dataset composition

[0039]

[0040] 2) Construct a power spectrum denoising network: The structure of the power spectrum denoising network is as follows Figure 1 As shown in the figure, each layer of the network uses a 1*1 convolution kernel, and it consists of an encoder and a decoder. The encoder part consists of convolution operations and downsampling operations. Each conv1*1 convolution operation includes a convolution layer with a convolution kernel of 1*1, a Batch Norm layer, and a ReLU activation layer. Downsampling uses maximum pooling (maxpool), and its window size is 2*1. The decoder part consists of upsampling operations, channel splicing operations, and convolution operations. Upsampling uses the nearest neighbor 2x upsampling, channel splicing is the splicing of feature maps in the channel direction, and the convolution operation also includes a convolution layer with a convolution kernel of 1*1, a Batch Norm layer, and a ReLU activation layer.

[0041] 3) Constructing a spectrum peak extraction network: The structure of the spectrum peak extraction network is as follows Figure 2 As shown in the figure, the input of the network is an n*1 dimensional vector. First, the input vector is transposed to a 1*n vector, and then processed by a 1*1 convolution layer for dimensionality reduction, and finally the spectral peak position is output. Each conv1*1 convolution operation includes a convolution layer with a convolution kernel of 1*1, a Batch Norm layer, and a ReLU activation layer.

[0042] 4) Construct a joint loss to train the network: Figure 3 As shown in Figure 2, the power spectrum denoising network and the spectrum peak extraction network are jointly trained, and the total loss function is the weighted sum of Loss1 and Loss2, and the calculation formula is Loss Total =0.01Loss1+Loss2, Loss1 and Loss2 are both Smooth L1 loss functions, the formula is as follows.

[0043]

[0044] 5) Wind speed extraction: After the network training is completed, it is directly used for wind speed extraction. Input the collected FFT power spectrum signal, the network outputs the frequency value f0 corresponding to the spectrum peak, and calculates the wind speed according to the Doppler frequency shift formula.

[0045] In this specification, each embodiment is described in a progressive, parallel, or progressive and parallel manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0046] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that an article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such article or device. In the absence of further restrictions, the elements defined by the sentence "comprising a ..." do not exclude the existence of other identical elements in the article or device including the above elements.

[0047] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting wind speed using a laser wind radar, characterized in that: The steps include: (1) Dataset preparation: obtain the simulated dataset DataF and the real dataset DataT; The step of acquiring the simulation data set DataF at least comprises: collecting the noise power spectrum P of the laser wind radar under different environmental conditions NF (f), by comparing with the simulated FFT power spectrum signal P SF0 (f) Synthesize to obtain the analog signal data P with noise SF1 (f) = P SF0 (f)+P NF (f); The step of acquiring the real data set DataT at least includes: collecting real signal data, recording the power spectrum signal P of each range gate ST1 (F), and the corresponding true wind speed V, and noise power spectrum P NT (F), the power spectrum of the denoised signal P is calculated SF0 (f) = P SF1 (f)-P NF (f); (2) Constructing a power spectrum denoising network: Each layer of the power spectrum denoising network uses a 1*1 convolution kernel, and the convolution kernel consists of an encoder and a decoder; the encoder consists of a convolution operation and a downsampling operation for extracting features; the decoder consists of an upsampling operation, a channel splicing operation, and a convolution operation; (3) Constructing a spectrum peak extraction network: The input of the spectrum peak extraction network is an n*1 dimensional vector. First, the input vector is transposed into a 1*n vector, and then processed by a 1*1 convolution layer for dimensionality reduction, and then the spectrum peak position is output; (4) Constructing a joint loss to train the network: Joint training is performed based on the power spectrum denoising network and the spectrum peak extraction network, and the total loss function Loss Total It is the weighted sum of Loss1 and Loss2, both Loss1 and Loss2 are Smooth L1 loss functions; (5) Wind speed calculation: After the network training is completed, it is directly used for wind speed calculation. First, the collected FFT power spectrum signal P is input. SF0 The frequency value f0 corresponding to the network output spectrum peak is used to calculate the wind speed based on the Doppler frequency shift formula.

2. The method according to claim 1, characterized in that: The simulated power spectrum P SF0 (f) Gaussian fitting is adopted, and the Gaussian fitting formula is: Where λ represents the laser wavelength, which is 1.55*10 -6 m, f0 is the Doppler frequency, which is calculated from the randomly generated wind V, A is a constant representing the amplitude of the power spectrum, and σ is a constant representing the spectral width of the power spectrum.

3. The method according to claim 2, characterized in that: The P SF1 (f) and the P SF0 (f) Perform maximum value normalization and map all data to the interval [0, 1], The maximum value normalization calculation formula is:

4. The method according to claim 3, characterized in that: The convolution operation includes a convolution layer with a convolution kernel of 1*1, a Batch Norm layer and a ReLU activation layer.

5. The method according to claim 1, characterized in that: The total loss function calculation formula is: Loss Total =0.01Loss1+Loss2 Among them, Loss1 and Loss2 are both SmoothL1 loss functions; the SmoothL1 loss function is:

6. The method according to claim 1, characterized in that: The wind speed calculation formula is:

Images