A wind speed solving method and system based on pulse laser radar echo signals

Abstract

The application provides a wind speed solving method and system based on pulse laser radar echo signals, and relates to the technical field of laser radar wind measurement. The application precisely acquires atmospheric turbulence data of a predetermined detection path, then performs targeted wavefront modulation on a coded laser pulse sequence through an adjustable optical element, generates a transmission laser pulse sequence that adapts to detection requirements and transmits along the path; receives echo signals returned on the detection path and converts them into initial electric signals, inputs the initial electric signals into a preset constructed neural network, removes distortion and noise interference contained in the signals, and obtains pure echo pulse sequences; decodes the sequences, separates multiple independent echo pulses corresponding to different distance sections, and finally accurately calculates and obtains wind speed data of each distance section according to frequency offsets of the independent echo pulses, so that the laser radar can accurately solve wind speeds of different distance sections.

Description

Technical Field

[0001] This application relates to the field of lidar wind measurement technology, and in particular to a method and system for calculating wind speed based on pulse lidar echo signals. Background Technology

[0002] Radar echo signal wind speed calculation is a key technology that obtains wind speed information by analyzing atmospheric echo signals received by radar. It is widely used in meteorological observation, aerospace, wind power operation and maintenance, and other fields. This technology enables real-time monitoring of wind speeds in different areas, providing crucial data support for safety assurance and precise control in related fields.

[0003] Currently, existing wind speed calculation methods based on pulsed lidar echo signals typically involve emitting a pulsed laser signal with fixed parameters into the detection area, receiving the echo signal generated by atmospheric scattering, and then directly performing simple processing such as filtering and amplification on the echo signal before calculating the wind speed by extracting the frequency characteristics of the echo signal. Some methods incorporate a range segmentation strategy, processing the echo signal separately for different range segments to obtain the wind speed for the corresponding area.

[0004] However, in practical applications, existing technologies suffer from problems such as atmospheric turbulence causing distortion of echo signals and environmental noise interference during laser transmission. Simple signal processing methods are unable to effectively eliminate these effects, resulting in low accuracy of the calculated wind speed data, especially under complex weather conditions where the error is more pronounced. Summary of the Invention

[0005] The purpose of this application is to provide a wind speed calculation method and system based on pulsed lidar echo signals, so as to solve the problem that the accuracy of pulsed lidar wind speed calculation is greatly affected by atmospheric interference in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a wind speed calculation method based on pulsed lidar echo signals, comprising:

[0007] Acquire atmospheric turbulence data along the predetermined detection path;

[0008] Based on the atmospheric turbulence data, the encoded laser pulse sequence is wavefront modulated using tunable optical elements to generate an emitted laser pulse sequence;

[0009] The sequence of emitted laser pulses is emitted along the predetermined detection path;

[0010] Receive the echo signal returned by the emitted laser pulse sequence on the predetermined detection path, and convert the echo signal into an initial electrical signal;

[0011] The initial electrical signal is input into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal.

[0012] The purified echo pulse sequence is decoded to separate multiple independent echo pulses, each corresponding to a different distance segment;

[0013] The wind speed for each distance segment is calculated based on the frequency offset of the multiple independent echo pulses.

[0014] Optionally, based on the atmospheric turbulence data, the encoded laser pulse sequence is wavefront modulated using tunable optical elements to generate an emitted laser pulse sequence, including:

[0015] Turbulence intensity distribution and wind speed gradient information were extracted from the atmospheric turbulence data.

[0016] Based on the turbulence intensity distribution and the wind speed gradient information, calculate the phase compensation amount and amplitude scaling factor required for the tunable optical element;

[0017] Based on the phase compensation amount and the amplitude scaling factor, a control signal sequence for the tunable optical element is generated;

[0018] The control signal sequence is used to drive the tunable optical element to adjust the wavefront shape of each laser pulse in the encoded laser pulse sequence pulse by pulse.

[0019] The encoded laser pulse sequence has a predetermined encoding rule, wherein the width of a single pulse is used to determine the range resolution of the system, and the total energy of the entire pulse sequence is used to maintain long-range detection;

[0020] The adjusted pulse sequences are combined in chronological order to form an emitted laser pulse sequence.

[0021] Optionally, based on the turbulence intensity distribution and the wind speed gradient information, the required phase compensation and amplitude scaling factor for the tunable optical element are calculated, including:

[0022] The turbulence intensity distribution is divided along the beam propagation path according to different height layers to obtain the turbulence intensity value of each height layer;

[0023] Based on the wind speed gradient information, determine the atmospheric flow velocity vector and direction at each altitude level;

[0024] Based on wave optics theory, a mathematical model is established to show the relationship between beam phase change and turbulence intensity at different altitudes.

[0025] The turbulence intensity value and atmospheric flow velocity vector at each altitude level are input into the mathematical relationship model to calculate the cumulative phase distortion of the light beam after passing through the atmospheric channel.

[0026] Based on the accumulated phase distortion, a global phase adjustment amount is determined as the phase compensation amount; simultaneously, based on the scattering and attenuation effect of the turbulence intensity value on the beam energy, the beam intensity attenuation ratio is calculated.

[0027] Based on the beam intensity attenuation ratio, a signal gain adjustment coefficient is determined as an amplitude scaling factor.

[0028] Optionally, the initial electrical signal is input into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal, including:

[0029] The initial electrical signal is divided into multiple overlapping time window segments, and the signal data in each time window segment is analyzed in the time domain and frequency domain to obtain the feature vectors in the time domain and frequency domain.

[0030] The time-domain and frequency-domain feature vectors are input into a pre-constructed neural network.

[0031] Using the hidden layers of the neural network, the absolute difference between different feature dimensions in the feature vectors of the time domain and frequency domain is calculated to generate an interaction matrix representing the relationship between features;

[0032] The elements in the interaction matrix are expanded in a predetermined order to form a transformed feature vector, and a predefined activation function is applied to the transformed feature vector to perform a nonlinear transformation.

[0033] Based on the transformed feature vector, the denoised signal segments of the corresponding time window are reconstructed. All the denoised signal segments of the time window are spliced ​​together in chronological order to generate a purified echo pulse sequence.

[0034] Optionally, the purified echo pulse sequence is decoded to separate multiple independent echo pulses, each corresponding to a different distance segment, including:

[0035] The purified echo pulse sequence is convolved with a predefined decoding matrix to calculate the correlation coefficient sequence for each pulse.

[0036] Based on a preset correlation coefficient threshold, all peak points exceeding the correlation coefficient threshold are selected from the correlation coefficient sequence, and the pulse corresponding to each peak point is marked as a valid pulse.

[0037] Based on the time delay of each effective pulse in the echo pulse sequence, all the effective pulses are divided into different time delay groups, and each time delay group corresponds to a specific distance segment;

[0038] Within each time delay group, the amplitude values ​​of all valid pulses are compared, and the valid pulse with the largest amplitude value is selected as the independent echo pulse corresponding to the distance segment.

[0039] Optionally, using the hidden layers of the neural network, the absolute difference between different feature dimensions in the feature vectors of the time domain and frequency domain is calculated to generate an interaction matrix representing the relationship between features, including:

[0040] Based on the number of dimensions of the feature vectors in the time domain and frequency domain, determine the range of dimension pairs to be calculated, and perform an absolute difference operation on each dimension of the feature vectors in the time domain and frequency domain to obtain the difference data for each pair of dimensions.

[0041] The difference data is filled into the initial matrix according to the row and column organization, where the row index corresponds to the time domain dimension index and the column index corresponds to the frequency domain dimension index.

[0042] Adjust the element layout of the initial matrix to conform to the preset matrix format, and output an interactive matrix.

[0043] Optionally, receiving the echo signal returned by the emitted laser pulse sequence on the predetermined detection path and converting the echo signal into an initial electrical signal includes:

[0044] The echo signal returned along the predetermined detection path is captured by an optical receiving device, and the focused echo signal is converted into an analog electrical signal by a photoelectric converter.

[0045] The analog electrical signal is amplified and filtered, and the processed analog electrical signal is sampled into a digital signal sequence through an analog-to-digital converter;

[0046] The digital signal sequence is timestamped to generate an initial electrical signal.

[0047] Secondly, this application provides a wind speed calculation system based on pulsed lidar echo signals, comprising:

[0048] The acquisition module is used to acquire atmospheric turbulence data along the predetermined detection path;

[0049] The generation module is used to generate an emitted laser pulse sequence by wavefront modulation of the encoded laser pulse sequence using tunable optical elements based on the atmospheric turbulence data.

[0050] The transmitting module is used to transmit the laser pulse sequence along the predetermined detection path;

[0051] A conversion module is used to receive the echo signal returned by the emitted laser pulse sequence on the predetermined detection path and convert the echo signal into an initial electrical signal;

[0052] A purification module is used to input the initial electrical signal into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal.

[0053] The separation module is used to decode the purified echo pulse sequence to separate multiple independent echo pulses, each of which corresponds to a different distance segment.

[0054] The calculation module is used to calculate the wind speed for each distance segment based on the frequency offset of the multiple independent echo pulses.

[0055] Thirdly, this application provides an electronic device, comprising:

[0056] Memory, used to store computer programs;

[0057] A processor is configured to execute the computer program to implement the steps of the wind speed calculation method based on pulsed lidar echo signals as described in the first aspect above.

[0058] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the wind speed calculation method based on pulsed lidar echo signals as described in the first aspect above.

[0059] The wind speed calculation method based on pulsed lidar echo signals provided in this application can provide environmental basis for the precise modulation of subsequent laser pulse sequences by acquiring atmospheric turbulence data along a predetermined detection path. By using adjustable optical elements to perform wavefront modulation on the coded laser pulse sequence based on the atmospheric turbulence data to generate the emitted laser pulse sequence, the emitted laser pulses can be adapted to the atmospheric environment of the detection path, reducing the impact of turbulence on laser transmission. By emitting the emitted laser pulse sequence along the predetermined detection path, precise coverage of the target detection area can be achieved. By receiving the echo signal returned by the emitted laser pulse sequence and converting it into an initial electrical signal, the optical signal can be converted into an electrical signal that can be processed subsequently. By inputting the initial electrical signal into a pre-constructed neural network to obtain a purified echo pulse sequence, distortion and noise in the initial electrical signal can be effectively eliminated, improving signal quality. By decoding the purified echo pulse sequence to separate multiple independent echo pulses corresponding to different distance segments, wind speed information of different distance areas can be distinguished and extracted. By calculating the wind speed of each distance segment based on the frequency offset of multiple independent echo pulses, the wind speed data of each area can be accurately obtained.

[0060] Furthermore, turbulence intensity distribution and wind speed gradient information are extracted from atmospheric turbulence data. Based on this information, the phase compensation and amplitude scaling factors required for the tunable optical element are calculated. A control signal sequence is then generated to drive the tunable optical element to adjust the wavefront shape of each pulse in the coded laser pulse sequence pulse by pulse. Finally, the adjusted pulses are combined in chronological order to form the emitted laser pulse sequence. This step enables precise and adaptive wavefront modulation of the coded laser pulse sequence, making the generated emitted laser pulse sequence more closely match the actual atmospheric turbulence along the detection path. This minimizes turbulence interference with laser transmission while also considering the system's range resolution and long-distance detection requirements, laying a solid foundation for subsequently acquiring high-quality echo signals and improving wind speed calculation accuracy. Attached Figure Description

[0061] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A flowchart illustrating a wind speed calculation method based on pulsed lidar echo signals provided in this application embodiment;

[0063] Figure 2A flowchart illustrating a specific implementation of a wind speed calculation method based on pulsed lidar echo signals provided in this application embodiment;

[0064] Figure 3 This is a schematic diagram of a wind speed calculation system based on pulsed lidar echo signals, provided as an embodiment of this application. Detailed Implementation

[0065] In wind speed calculation techniques based on pulsed lidar echo signals, existing methods generally employ fixed parameters to emit laser signals, performing only simple filtering and amplification of the echo signals before extracting frequency features to calculate wind speed. This approach struggles to address atmospheric interference issues in practical applications. Atmospheric turbulence distorts the echo signals, and combined with environmental noise, the calculated wind speed data suffers from low accuracy, particularly under complex weather conditions, failing to meet the demands for precise wind speed data in meteorological observation, wind power operation and maintenance, and other fields.

[0066] To address the aforementioned issues, this application proposes a wind speed calculation method based on pulsed lidar echo signals. The core of this method lies in first acquiring atmospheric turbulence data along the detection path, adjusting the laser signal parameters accordingly before transmission, and simultaneously using a dedicated signal processing model to purify the echo signal. Specifically, the transmission parameters of the laser pulse sequence are first adjusted using atmospheric turbulence data to make the transmitted laser signal more adaptable to the actual detection environment. Then, the received echo signal is converted into an electrical signal and input into a pre-defined neural network to eliminate distortion and noise, resulting in a clean echo signal. Finally, the clean signal is decoded to separate signals for different distance segments, and the wind speed for each segment is calculated accordingly. This method, through the dual design of environment-adaptive laser transmission and precise signal purification, reduces the impact of atmospheric interference on the signal at the source and effectively eliminates existing interference, fundamentally solving the problem that the accuracy of wind speed calculation is greatly affected by atmospheric interference in existing technologies. This provides more reliable wind speed data support for related fields.

[0067] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0068] The core of this application is to provide a wind speed calculation method based on pulsed lidar echo signals, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes:

[0069] S101. Obtain atmospheric turbulence data along the predetermined detection path.

[0070] The predetermined detection path refers to a specific spatial path that is pre-set for conducting wind speed detection; atmospheric turbulence data includes information such as the intensity distribution of irregular airflow movements in the atmosphere and the gradient of wind speed changes, which are used to reflect the atmospheric environmental interference along the detection path.

[0071] In one specific implementation, an atmospheric environment sensing unit paired with a pulsed lidar can be used to scan and detect a predetermined detection path, capture the irregular movement state of airflow along the path in real time, and then collect and process atmospheric turbulence data containing key information such as turbulence intensity distribution and wind speed gradient.

[0072] S102. Based on the atmospheric turbulence data, the encoded laser pulse sequence is wavefront modulated using tunable optical elements to generate an emitted laser pulse sequence.

[0073] In this step, tunable optical elements refer to elements whose optical characteristics, such as phase and amplitude, can be changed by electrical signals to adjust the propagation pattern of laser pulses; encoded laser pulse sequences refer to a series of laser pulses arranged according to preset rules, which are used to distinguish and enhance signals through encoding; wavefront modulation refers to the process of adjusting the wavefront shape of the laser to make the laser more suitable for the transmission environment; and emitted laser pulse sequences refer to the set of laser pulses that are ultimately used for emission detection after wavefront modulation.

[0074] Optionally, such as Figure 2 As shown, step S102 may specifically include the following steps:

[0075] S1021. Extract the turbulence intensity distribution and wind speed gradient information from the atmospheric turbulence data.

[0076] Among them, turbulence intensity distribution refers to the spatial distribution of the strength of atmospheric turbulence along the detection path, including turbulence intensity data at different locations; wind speed gradient information refers to the rate of change of wind speed at different spatial locations, reflecting the variation law of wind speed with location.

[0077] In one specific implementation, the atmospheric turbulence data obtained along the predetermined detection path is first screened and analyzed: invalid data mainly includes data that exceeds the detection path range, abnormal fluctuation data caused by equipment failure, and data with a signal-to-noise ratio lower than a preset threshold. These invalid data are removed by comparing the detection path coordinate range and setting data fluctuation thresholds and signal-to-noise ratio thresholds.

[0078] Subsequently, information extraction is performed. For turbulence intensity distribution data, a spatial interpolation algorithm is used to complete the filtered effective data points, resulting in a continuous spatial distribution curve of turbulence intensity along the detection path, thus forming the turbulence intensity distribution. For wind speed gradient information, the ratio of the wind speed difference between adjacent effective data points to the corresponding spatial distance is calculated to obtain the wind speed change rate of each segment. After integration, wind speed gradient information reflecting the wind speed change pattern of the entire path is formed.

[0079] For example, in a pulsed lidar atmospheric wind speed detection scenario, assuming the predetermined detection path is a vertical path from the ground to an altitude of 1000 meters, the acquired atmospheric turbulence data includes raw turbulence-related data from 20 altitude points spaced 50 meters apart along this path. During the filtering phase, two data points with coordinates outside the 0-1000 meter altitude range are first removed, then one faulty data point with a value abruptly changing to 20 times the normal range is removed, along with three low-quality data points with a signal-to-noise ratio below 15 dB. Finally, 14 valid data points are retained.

[0080] When extracting turbulence intensity distribution data, Kriging interpolation was used to complete the 14 valid data points, obtaining turbulence intensity values ​​every 10 meters at heights from 0 to 1000 meters, forming a continuous turbulence intensity distribution curve. When extracting wind speed gradient information, the wind speed difference between adjacent 10-meter height points was calculated and then divided by the 10-meter distance to obtain the rate of change of wind speed for each segment. For example, if the wind speed changes from 1.2 m / s to 1.3 m / s at a height of 10 meters below ground, the wind speed gradient is (1.3 - 1.2) / 10 = The wind speed gradient values ​​of all height segments are integrated to form the wind speed gradient information of the path.

[0081] S1022. Based on the turbulence intensity distribution and the wind speed gradient information, calculate the phase compensation amount and amplitude scaling factor required for the adjustable optical element.

[0082] Specifically, step S1022 may include the following processes: dividing the turbulence intensity distribution along the beam transmission path according to different altitude layers to obtain the turbulence intensity value of each altitude layer; determining the atmospheric flow velocity vector and direction of each altitude layer based on the wind speed gradient information; establishing a mathematical relationship model between the beam phase change and the turbulence intensity value of each altitude layer based on wave optics theory; inputting the turbulence intensity value and the atmospheric flow velocity vector of each altitude layer into the mathematical relationship model to calculate the cumulative phase distortion of the beam after passing through the atmospheric channel; determining the global phase adjustment amount as the phase compensation amount based on the cumulative phase distortion; simultaneously calculating the beam intensity attenuation ratio based on the scattering attenuation effect of the turbulence intensity value on the beam energy; and determining the signal gain adjustment coefficient as the amplitude scaling factor based on the beam intensity attenuation ratio.

[0083] In the above steps, the phase compensation amount refers to the phase adjustment amount that needs to be applied by the adjustable optical element to compensate for the laser phase distortion caused by atmospheric turbulence; the amplitude scaling factor refers to the energy gain adjustment coefficient that needs to be applied by the adjustable optical element to compensate for the laser energy attenuation caused by atmospheric turbulence; wave optics theory is a theory that studies the wave nature of light and the interaction between light and matter during propagation; the atmospheric channel refers to the path of laser propagation in the atmosphere; the cumulative phase distortion refers to the degree of total phase distortion caused by turbulence after the laser has propagated through the entire atmospheric channel; the beam intensity attenuation ratio refers to the proportion of energy attenuation caused by turbulent scattering to the initial energy when the laser propagates in the atmospheric channel; and the signal gain adjustment coefficient is a coefficient used to compensate for beam intensity attenuation.

[0084] In this embodiment, firstly, the turbulence intensity distribution is divided along the beam transmission path according to different height layers to obtain the turbulence intensity value of each height layer; then, the atmospheric flow velocity vector and direction of each height layer are determined based on the wind speed gradient information; subsequently, based on wave optics theory and combined with laser transmission characteristics, a mathematical relationship model between beam phase change and turbulence intensity value of each height layer is established.

[0085] The training process of the model is as follows: collect sample data of laser phase change under different turbulence intensities, divide the sample data into training set and test set, take the turbulence intensity value as input and the laser phase change as output, train the model, verify and optimize the model parameters through the test set until the model prediction accuracy meets the preset requirements.

[0086] Next, the turbulence intensity values ​​and atmospheric flow velocity vectors at each altitude layer are input into the trained mathematical relationship model to calculate the cumulative phase distortion of the beam after passing through the atmospheric channel. The global phase adjustment amount is determined based on the cumulative phase distortion amount as the phase compensation amount. At the same time, the beam intensity attenuation ratio is calculated based on the scattering and attenuation effect of the turbulence intensity value on the beam energy. The signal gain adjustment coefficient is determined based on the beam intensity attenuation ratio as the amplitude scaling factor.

[0087] In practical applications, continuing the previous scenario of pulsed lidar atmospheric wind speed detection, the detection path is divided into 10 height levels at 100-meter intervals, and the turbulence intensity value of each height level is obtained, as follows: The atmospheric flow velocity vector for each altitude layer is determined based on the wind speed gradient information. For example, the velocity vector for the first altitude layer is... The second altitude layer is The mathematical relationship model based on wave optics theory is shown in equation (1):

[0088] (1)

[0089] in, For the first The amount of laser phase change caused by each height layer; For wave number, , The wavelength is the laser wavelength; here, the laser wavelength is taken. Calculated ; For the first Turbulence intensity values ​​at each height level, in units of: ; For the first The thickness of each height layer is 100m; For the first The angle between the atmospheric flow velocity vector at each altitude level and the laser propagation direction.

[0090] Assume the turbulence intensity values ​​and included angles at each height layer are as follows: , ; , ;... , Substituting into equation (1), calculate the phase change at each altitude level:

[0091] Level 1: ;

[0092] Level 2: rad.

[0093] After calculating the phase changes at the remaining height levels in sequence, the cumulative phase distortion is calculated. The global phase adjustment is equal to the cumulative phase distortion, i.e., the phase compensation is... .

[0094] The beam intensity attenuation ratio is calculated using equation (2):

[0095] (2)

[0096] in, This represents the beam intensity attenuation ratio. For varying turbulence intensity The varying attenuation coefficient, in units of: Take this place L represents the total length of the detection path.

[0097] Substitute the turbulence intensity values ​​at each height level to calculate the integral: , but That is, the beam intensity attenuation ratio is approximately The amplitude scaling factor is set to 1. .

[0098] The above example is only one example of this application. In practical applications, the model parameters can be adjusted according to parameters such as laser wavelength and detection path length. This application does not limit this.

[0099] S1023. Generate a control signal sequence for the adjustable optical element based on the phase compensation amount and the amplitude scaling factor.

[0100] Among them, the control signal sequence refers to a series of electrical signals used to drive the tunable optical element to achieve phase compensation and amplitude adjustment, and its parameters are matched with the phase compensation amount and amplitude scaling factor.

[0101] Specifically, firstly, based on the operating characteristics of the tunable optical element, a mapping relationship is established between the phase compensation amount, amplitude scaling factor, and control signal parameters. Then, the calculated phase compensation amount and amplitude scaling factor are substituted into the mapping relationship to convert them into electrical signal parameters recognizable by the tunable optical element. Finally, according to the pulse order of the coded laser pulse sequence, a corresponding control signal sequence is generated to ensure that each laser pulse receives the corresponding phase and amplitude adjustment.

[0102] In practical applications, continuing the previous example, assuming the adjustable optical element used has the following mapping relationship between phase compensation and control voltage: , The mapping relationship between the amplitude scaling factor and the control current is as follows: , A is the amplitude scaling factor. Substitute the phase compensation amount. With an amplitude scaling factor of 1.000524, the control voltage is calculated. Control current If the encoded laser pulse sequence contains 100 pulses, generate 100 groups according to the pulse order. The control signal sequence of electrical signals.

[0103] S1024. The control signal sequence is used to drive the adjustable optical element to adjust the wavefront shape of each laser pulse in the encoded laser pulse sequence pulse by pulse.

[0104] The encoded laser pulse sequence has a predetermined encoding rule, wherein the width of a single pulse is used to determine the range resolution of the system, and the total energy of the entire pulse sequence is used to maintain long-range detection.

[0105] In this embodiment, firstly, the generated control signal sequence is synchronized with the coded laser pulse sequence to ensure that each control signal in the control signal sequence corresponds one-to-one with each laser pulse in the coded laser pulse sequence. Then, the synchronized control signal sequence is input into the driving module of the adjustable optical element. The driving module drives the adjustable optical element to work according to the control signal, and performs wavefront morphology adjustment on each laser pulse in the coded laser pulse sequence in sequence to achieve preset phase compensation and amplitude adjustment.

[0106] It should also be noted that the predetermined encoding rules for the encoded laser pulse sequence can be set according to the detection requirements. For example, pseudo-random encoding or orthogonal encoding can be used to give the laser pulses good anti-interference and distinguishability. The narrower the width of a single laser pulse, the smaller the distance between two target points that the system can distinguish, that is, the stronger the distance resolution capability; the greater the total energy of the entire encoded laser pulse sequence, the farther the laser can propagate, thus ensuring the feasibility of long-distance detection.

[0107] For example, continuing the previous example, the control signal sequence containing 100 sets of electrical signals is synchronized with the coded laser pulse sequence containing 100 pulses. When the first laser pulse reaches the adjustable optical element, the drive module outputs the first set of control signals (0.21V, 0.524A) to drive the adjustable optical element to adjust the wavefront of the laser pulse. Subsequently, the subsequent 99 laser pulses are adjusted pulse by pulse.

[0108] S1025. Combine the adjusted pulse sequences in chronological order to form an emitted laser pulse sequence.

[0109] Specifically, after completing the pulse-by-pulse wavefront adjustment of each laser pulse in the coded laser pulse sequence, all the adjusted laser pulses are sequentially combined according to the time order of the original coded laser pulse sequence to form a complete emission laser pulse sequence, which can be directly used for emission along a predetermined detection path.

[0110] In practical applications, continuing the previous example, the 100 laser pulses, after pulse-by-pulse adjustment, are arranged sequentially according to the chronological order of the original encoded laser pulse sequence to form a transmitted laser pulse sequence containing 100 adjusted pulses. The above example is merely one illustration of this application; in practical applications, the pulse combination order can be adjusted according to transmission requirements, and this application does not limit this.

[0111] This application achieves adaptive wavefront modulation of the coded laser pulse sequence through the above steps, which can accurately cancel the laser phase distortion and energy attenuation caused by atmospheric turbulence. It solves the problems of laser transmission being easily affected by atmospheric turbulence and poor detection stability in traditional laser emission schemes. Compared with traditional laser emission schemes with fixed parameters, it significantly improves the transmission quality and detection reliability of laser in complex atmospheric environments.

[0112] S103. The laser pulse sequence is emitted along the predetermined detection path.

[0113] Specifically, the transmitted laser pulse sequence formed after wavefront modulation is transmitted to the transmitting module of the lidar. The transmitting antenna in the transmitting module is aligned with the starting end of the predetermined detection path, and the transmitted laser pulse sequence is accurately emitted along the predetermined detection path according to the preset transmission timing and power, ensuring that the laser pulse can cover the entire detection path range.

[0114] S104. Receive the echo signal returned by the emitted laser pulse sequence on the predetermined detection path, and convert the echo signal into an initial electrical signal.

[0115] Among them, the echo signal refers to the light signal reflected back after the emitted laser pulse sequence interacts with atmospheric particles on the predetermined detection path, and contains wind speed-related information of the detection path; the initial electrical signal refers to the digital electrical signal sequence that can be used for subsequent neural network processing after photoelectric conversion, signal processing and timestamp marking.

[0116] Optionally, step S104 may specifically include the following steps:

[0117] S1041. The echo signal returned on the predetermined detection path is captured by an optical receiving device, and the focused echo signal is converted into an analog electrical signal by a photoelectric converter.

[0118] Among them, the optical receiving device refers to a device composed of components such as lenses and receiving antennas, which is used to capture and focus the returned echo light signal; the photoelectric converter is a device that converts light signals into electrical signals based on the photoelectric effect, and is used to realize the basic conversion from light signals to electrical signals; the analog electrical signal refers to an electrical signal that changes continuously with time, and its amplitude changes correspond to the intensity changes of the echo light signal.

[0119] In one specific implementation, firstly, the angle of the optical receiving device is adjusted to align it with the signal return direction of the predetermined detection path, ensuring effective capture of the returned echo signal. The lens inside the optical receiving device focuses the captured divergent echo signal, enhancing its signal strength. Subsequently, the focused echo signal is transmitted to a photoelectric converter, which, based on the photoelectric effect, converts the intensity change of the light signal into a corresponding voltage or current change, ultimately outputting an analog electrical signal.

[0120] For example, in a pulsed lidar atmospheric wind speed detection scenario, an optical receiving device including a convex lens and a receiving antenna is used. The receiving antenna is adjusted to align with the signal return direction of the detection path. After the captured echo signal is focused by the convex lens, the optical power density is increased to 10 times its original value. Then, the focused echo signal is input to a photodiode-type photoelectric converter with a responsivity of 0.8 A / W. When the input focused echo signal optical power is 10 nW, the output analog voltage signal amplitude is 0.8 A / W × 10 nW = 8 nV. The above example is merely one example of this application. In practical applications, photoelectric converters with different responsivity and lenses with different focal lengths can be selected according to the echo signal intensity. This application does not limit this.

[0121] S1042. The analog electrical signal is amplified and filtered, and the processed analog electrical signal is sampled into a digital signal sequence through an analog-to-digital converter.

[0122] Amplification refers to increasing the amplitude of an analog electrical signal through an amplification circuit to achieve the signal strength required for subsequent processing; filtering refers to removing interference noise from the analog electrical signal through a filtering circuit, retaining the effective signal components; an analog-to-digital converter is a device that converts continuously changing analog electrical signals into discrete digital signals; a digital signal sequence refers to a signal sequence composed of a series of discrete numbers, which can be used for subsequent digital processing.

[0123] Specifically, the weak analog electrical signal output from the photoelectric converter is first input into an amplification circuit composed of operational amplifiers to amplify the analog electrical signal. Then, the amplified analog electrical signal is input into a low-pass filter circuit to remove interference noise with frequencies higher than the effective signal. Finally, the amplified and filtered analog electrical signal is input into an analog-to-digital converter (ADC). The ADC performs discrete sampling and quantization of the analog electrical signal according to a preset sampling frequency and quantization bit depth, converting the continuous analog electrical signal into a discrete digital signal sequence.

[0124] During amplification and filtering, the amplification factor and the filter cutoff frequency can be determined by calculation using simple formulas. The formula for calculating the amplification factor is shown in equation (3):

[0125] (3)

[0126] in, This is the voltage amplification factor. The amplitude of the amplified analog electrical signal. The amplitude of the analog electrical signal before amplification is given. The formula for calculating the cutoff frequency of the low-pass filter circuit is shown in equation (4):

[0127] (4)

[0128] in, R is the low-pass filter cutoff frequency, R is the resistance value of the filter resistor, and C is the capacitance of the filter capacitor.

[0129] In practical applications, continuing the scenario of atmospheric wind speed detection using pulsed lidar mentioned earlier, the amplitude of the analog electrical signal output by the photoelectric converter is 8nV. To meet the requirements of subsequent analog-to-digital conversion, the signal amplitude needs to be amplified to 1V. The required voltage amplification factor is calculated according to equation (3). The operational amplifier is selected to form an amplifier circuit, which amplifies the analog electrical signal by 125,000 times, and then outputs an analog electrical signal with an amplitude of 1V.

[0130] Given that the highest effective echo signal frequency in this scenario is 1MHz, a low-pass filter circuit is designed, and the filter resistor is selected. Filter capacitor The cutoff frequency is calculated according to equation (4). It can effectively remove interference noise with frequencies higher than 1MHz. The filtered analog electrical signal is input to an analog-to-digital converter (ADC), with a sampling frequency of 5MHz (five times the highest frequency of the effective signal, satisfying the Nyquist sampling theorem) and a quantization bit of 16 bits. The ADC then converts the amplitude range... The analog electrical signal is quantized into The digital quantity is used to ultimately output the corresponding digital signal sequence.

[0131] In another specific implementation, an instrumentation amplifier can be used instead of an operational amplifier for amplification. The instrumentation amplifier has higher input impedance and common-mode rejection ratio, enabling better suppression of common-mode interference and making it suitable for detection environments with strong interference. Simultaneously, an active low-pass filter circuit is used instead of a passive low-pass filter circuit, resulting in more stable filtering and more flexible cutoff frequency adjustment.

[0132] S1043. Timestamp the digital signal sequence to generate an initial electrical signal.

[0133] The timestamp refers to adding corresponding time information to each sampling point in the digital signal sequence to identify the time when the signal was generated; the initial electrical signal is a digital signal sequence with timestamp information, containing the intensity and time information of the echo signal, which can be directly input into the subsequent neural network for processing.

[0134] Specifically, a GNSS time synchronization module is first used to provide a high-precision reference time signal. This module provides a whole-second time reference through a PPS second pulse signal and provides UTC time information (year, month, day, hour, minute, second) through GPRMC messages. The reference time signal is synchronized with the sampling clock of the analog-to-digital converter to ensure that the time information corresponds accurately to the sampling points. Then, according to a preset marking rule, corresponding timestamp information is added to each sampling point in the digital signal sequence. The timestamp information includes subdivided time at the second level and millisecond level. Finally, the digital signal sequence with added timestamps is used as the initial electrical signal output.

[0135] The timestamp can be calculated as follows: the time corresponding to the rising edge of the PPS second pulse is taken as the whole second. The sampling period of the analog-to-digital converter is Then the timestamp of the nth sampling point The calculation formula is shown in equation (5):

[0136] (5)

[0137] in, The timestamp of the nth sampling point The reference time is in whole seconds, and n is the sampling point number ( ), This represents the sampling period of the analog-to-digital converter.

[0138] For example, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, a GNSS timing module is used to provide a reference time signal, with the UTC time corresponding to the rising edge of the PPS second pulse. The value is 12:00:00.000. The sampling frequency of the analog-to-digital converter is 5MHz. The sampling period is calculated based on the relationship between the sampling period and the sampling frequency. For the 100th sampling point in the digital signal sequence, n=100, its timestamp is calculated according to equation (5). This method involves sequentially adding a timestamp to each sampling point in the digital signal sequence, ultimately generating an initial electrical signal with complete time information. The above example is merely one illustration of this application; in practical applications, the PTP protocol can also be used for time synchronization and timestamp marking, and this application does not limit this approach.

[0139] This application solves the problems of poor signal quality caused by weak signal, high noise interference, and missing time information in traditional echo signal reception and conversion processes by using precise acquisition and focusing of an optical receiving device, efficient photoelectric conversion, noise suppression and signal enhancement through amplification and filtering, and analog-to-digital conversion with timestamp marking. Compared with traditional solutions, it can significantly improve the reliability and signal quality of echo signal conversion, ensuring that the converted electrical signal accurately reflects the original information of the echo signal.

[0140] S105. The initial electrical signal is input into a pre-constructed neural network to obtain a purified echo pulse sequence.

[0141] The neural network is used to process the distortion and noise contained in the initial electrical signal, and includes an input layer, a hidden layer, and an output layer. The purified echo pulse sequence refers to the echo pulse sequence after processing by the neural network to remove distortion and noise interference, which can truly reflect atmospheric wind speed related information.

[0142] Optionally, step S105 may specifically include the following steps:

[0143] S1051. The initial electrical signal is divided into multiple overlapping time window segments, and the signal data in each time window segment is analyzed in the time domain and frequency domain to obtain the feature vectors in the time domain and frequency domain.

[0144] In this step, the overlapping time window segment refers to the sliding segmentation of the initial electrical signal using a fixed-length time window, with some overlap between adjacent time windows, to avoid feature loss caused by signal segmentation; time domain analysis refers to extracting the signal's features in the time dimension, such as the signal's amplitude, peak value, and mean value; frequency domain analysis refers to converting the signal from the time dimension to the frequency dimension and extracting features such as the signal's frequency distribution and amplitude spectrum; the feature vector refers to the vector data formed by arranging multiple extracted features in a preset order, used to characterize the core features of the signal within the time window segment.

[0145] Specifically, the length and overlap ratio of the time window are first set, and the initial electrical signal is divided into multiple consecutive overlapping time window segments using a sliding window approach. For the signal data within each time window segment, time-domain analysis uses statistical analysis methods to extract time-domain features such as amplitude, peak value, mean, and variance; frequency-domain analysis uses Fast Fourier Transform to convert the time-domain signal into a frequency-domain signal, and then extracts frequency-domain features such as amplitude spectrum peak value, center frequency, and bandwidth. Finally, the extracted time-domain and frequency-domain features are arranged in a preset order to form corresponding time-domain feature vectors and frequency-domain feature vectors.

[0146] For example, continuing the above scenario, the initial electrical signal is a digital signal sequence with a sampling frequency of 5MHz and a duration of 1ms, containing 5000 sampling points. The time window length is set to 200 sampling points, corresponding to a duration of 40μs, with an overlap ratio of 50%. Therefore, the sliding step size for each window is 100 sampling points. Through sliding segmentation, a total of 49 overlapping time window segments are obtained.

[0147] Select a time window containing signal data from 200 sampling points. Time-domain analysis is used to extract the following features: amplitude range of 0-1V, peak value of 0.95V. The mean is calculated using the mean calculation rules. Assuming the sum of the signal amplitudes of the 200 sampling points is 80V, then the mean = 80V / 200 = 0.4V, and the variance is 0.02. These time-domain features are arranged in the order of "amplitude range - peak value - mean - variance" to form a 4-dimensional time-domain feature vector [0-1V, 0.95V, 0.4V, 0.02V]. Frequency domain analysis involves performing a Fast Fourier Transform on the signal within the window segment. After converting it to a frequency domain signal, the amplitude spectrum peak value of 0.8V, center frequency of 500kHz, and bandwidth of 200kHz are extracted and arranged in this order to form a frequency domain feature vector of dimension 3 [0.8V, 500kHz, 200kHz].

[0148] S1052. Input the time-domain and frequency-domain feature vectors into the pre-constructed neural network.

[0149] The dimensions of the pre-built neural network input layer are set to match the total dimensions of the time-domain and frequency-domain feature vectors, and are used to receive and transmit feature vector data to the hidden layer for processing.

[0150] In one specific implementation, the time-domain feature vector and the frequency-domain feature vector are first concatenated to form a combined feature vector, with the concatenation order being a preset "time-domain feature vector first, frequency-domain feature vector last". Then, the combined feature vector is input into the input layer of a pre-constructed neural network. The input layer transmits the combined feature vector to the hidden layer of the neural network, preparing for subsequent feature interactions and nonlinear transformations.

[0151] The training process of the pre-built neural network is as follows: Initial electrical signal samples containing distortion and noise are collected as input data, and the corresponding signal samples after distortion and noise removal are used as label data. The input data and label data are divided into training and test sets in a 7:3 ratio. The parameters of the neural network are initialized, including the number of neurons, weights, and biases of the input, hidden, and output layers. Mean squared error is used as the loss function, and the neural network is trained using a stochastic gradient descent optimization algorithm. The network parameters are iteratively updated using the training set to reduce the loss function value. The denoising effect of the network is verified using the test set. If the verification error meets a preset threshold, training is stopped, and the trained neural network is obtained.

[0152] In practical applications, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, the time-domain feature vector has a dimension of 4, and the frequency-domain feature vector has a dimension of 3. These two are concatenated to form a combined feature vector of dimension 7: [0-1V, 0.95V, 0.4V, 0.02V]. [0.8V, 500kHz, 200kHz]. The pre-built neural network input layer has 7 neurons. The combined feature vector is input into this input layer. After receiving the feature data, the input layer neurons transmit it to the hidden layer containing 32 neurons. The above example is only one example of this application. In practical applications, the number of input layer neurons can be adjusted according to the feature vector dimension. This application does not limit this.

[0153] S1053. Using the hidden layer of the neural network, calculate the absolute difference between different feature dimensions in the feature vectors of the time domain and frequency domain, and generate an interaction matrix representing the relationship between features.

[0154] Specifically, step S1053 may include the following process: determining the range of dimension pairs to be calculated based on the number of dimensions of the feature vectors in the time domain and frequency domain; performing an absolute difference operation on each dimension of the feature vectors in the time domain and frequency domain to obtain the difference data for each pair of dimensions; filling the difference data into an initial matrix according to the row and column organization, wherein the row index corresponds to the time domain dimension number and the column index corresponds to the frequency domain dimension number; adjusting the element layout of the initial matrix to conform to a preset matrix format, and outputting an interactive matrix.

[0155] In the above steps, the hidden layer is a network layer located between the input layer and the output layer in a neural network, used to perform complex feature processing and transformation on the input feature vector; the absolute difference between different feature dimensions refers to the absolute value of the difference between the values ​​of each dimension of the time domain feature vector and each dimension of the frequency domain feature vector, used to characterize the degree of correlation between features in different domains; the interaction matrix is ​​a matrix formed by organizing the absolute differences of all dimension pairs according to preset row and column rules, which can intuitively present the interaction relationship between time domain and frequency domain features.

[0156] In this embodiment, firstly, based on the number of dimensions of the input time-domain and frequency-domain feature vectors, the range of dimension pairs to be calculated is determined. The range of dimension pairs includes all pairwise combinations of time-domain and frequency-domain feature dimensions. An absolute difference operation is performed on each dimension of the time-domain and frequency-domain feature vectors, i.e., the absolute difference between each time-domain feature dimension value and each frequency-domain feature dimension value is calculated sequentially to obtain the difference data for each pair of dimensions. Subsequently, an empty matrix is ​​initialized, with the number of rows equal to the number of dimensions of the time-domain feature vectors and the number of columns equal to the number of dimensions of the frequency-domain feature vectors. The obtained difference data is then filled into the initial matrix according to the rule that the row index corresponds to the time-domain dimension index and the column index corresponds to the frequency-domain dimension index. Finally, the element layout of the initial matrix is ​​adjusted to conform to a preset matrix format, and the output is an interactive matrix.

[0157] For example, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, the time-domain feature vector is dimension 4. Quantifiable feature values ​​are selected, and the frequency domain feature vector is of dimension 3. The dimension has a range of Each dimension pair. Calculate the absolute difference for each dimension pair according to the above absolute difference calculation rules: Because of the different units, it is necessary to unify the units in practical applications. Here, we assume that the calculated value after conversion is 499999.05. After conversion, we get 199999.05; 12 difference data points were calculated in this way.

[0158] Initialize a 4x3 matrix. Fill to row 1, column 1. Fill to row 1, column 2. Fill up to row 1, column 3. Fill the matrix up to the 2nd row and 1st column, and repeat this process for all difference data. Adjust the matrix layout and then output the result. The interaction matrix. The above example is only one example of this application. In practical applications, it is necessary to first unify the units and adjust the selection rules of the dimension pairs according to the feature type. This application does not limit this.

[0159] S1054. Expand the elements in the interaction matrix in a predetermined order to form a transformed feature vector, and apply a predefined activation function to the transformed feature vector to perform a nonlinear transformation.

[0160] Here, the predetermined order refers to the pre-defined rules for expanding matrix elements, such as row-major or column-major order; the transformed feature vector refers to expanding the two-dimensional interaction matrix into one-dimensional vector data, which facilitates processing by subsequent layers of the neural network; the activation function refers to the function used to introduce nonlinear transformations into the neural network, which can enhance the neural network's ability to learn complex features; the nonlinear transformation refers to changing the distribution of the feature vector through the activation function, so that the feature vector can better represent the complex patterns of the signal.

[0161] Specifically, firstly, elements in the interaction matrix are extracted one by one and arranged sequentially according to a preset row-major order, expanding the two-dimensional interaction matrix into a one-dimensional transformed feature vector. Then, the transformed feature vector is input into the next hidden layer of the neural network, and a predefined activation function, such as the ReLU function, is applied to it for a non-linear transformation. This transformation changes the numerical distribution of the feature vector through the mapping relationship of the activation function, preserving effective feature information and suppressing invalid interference information.

[0162] The ReLU activation function operates according to the following rules: for each element in the transformed feature vector, if the value is greater than 0, the value remains unchanged; if the value is less than or equal to 0, it is mapped to 0. This is how nonlinear transformation is achieved.

[0163] In practical applications, continuing the previous scenario, the 4x3 interaction matrix is ​​expanded in row-major order, resulting in a feature vector with a dimension of 4×3=12, namely [0.15V, 499999.05, 199999.05, 0.4V, ...], a total of 12 elements. The ReLU activation function is applied to this feature vector for nonlinear transformation: for the element 0.15V, whose value is greater than 0, 0.15V remains unchanged; for the element 499999.05, the value remains unchanged; for the hypothetical negative element -0.2, it is mapped to 0. This nonlinear transformation is performed on all elements sequentially, yielding the nonlinearly transformed feature vector. The above example is merely one example of this application; in practical applications, other activation functions such as sigmoid and tanh can also be chosen, and this application does not limit this choice.

[0164] In another specific implementation, the interaction matrix can be expanded in a column-first order, while a tanh activation function is selected for nonlinear transformation. The tanh activation function can map the values ​​of the feature vectors to the range of -1 to 1, which helps stabilize the network training process and is suitable for scenarios with large fluctuations in signal features.

[0165] S1055. Based on the transformed feature vector, reconstruct the denoised signal segment of the corresponding time window segment, and splice all the denoised signal segments of the time window segment in time order to generate a purified echo pulse sequence.

[0166] Here, reconstruction refers to restoring the transformed feature vector to the signal data corresponding to the time window segment through the output layer of the neural network; the denoised signal segment refers to the signal of the time window segment after removing distortion and noise; the purified echo pulse sequence refers to the complete pulse sequence formed by splicing all the denoised signal segments in the original time order, which can truly reflect the echo signal characteristics of the detection path.

[0167] In one specific implementation, firstly, the nonlinearly transformed feature vector is input into the output layer of the neural network. The output layer then uses a fully connected layer and a signal reconstruction algorithm to restore the feature vector into a denoised signal segment corresponding to the length of the time window. Since adjacent time windows overlap, the signal data in the overlapping areas are fused using a weighted average to avoid signal abrupt changes after splicing. Finally, all the denoised signal segments that have undergone the fusion process are spliced ​​together sequentially in their original time order to form a complete and purified echo pulse sequence.

[0168] In practical applications, continuing with the atmospheric wind speed detection scenario using pulsed lidar described earlier, the length of the denoised signal segment in each time window is 200 sampling points, with an overlap of 100 sampling points between adjacent signal segments. The signal data in the overlapping area is fused using an equal-weighted average, meaning the value of each sampling point within the overlapping area is the average of the values ​​of corresponding sampling points in two adjacent signal segments. The 49 denoised signal segments are then stitched together chronologically, and after fusing the overlapping area, a purified echo pulse sequence containing 5000 sampling points is finally generated. This sequence no longer contains significant distortion or noise interference.

[0169] This application addresses the problems of traditional signal denoising methods, such as difficulty in simultaneously removing distortion and noise and easy loss of effective signal features, by employing overlapping time window segmentation to avoid feature loss, time-domain and frequency-domain feature fusion extraction, feature interaction matrix construction to strengthen feature correlation, activation function to introduce nonlinear transformation, and overlapping region fusion and splicing. Compared with traditional methods, it can more accurately separate effective signals from interference components, significantly improving the denoising effect and signal integrity of echo signals.

[0170] S106. Decode the purified echo pulse sequence to separate multiple independent echo pulses.

[0171] Among them, multiple independent echo pulses refer to echo pulses that can individually characterize atmospheric information at a specific distance segment, with each independent echo pulse corresponding to a different distance segment on the detection path.

[0172] Optionally, step S106 may specifically include the following steps:

[0173] S1061. Perform convolution operation between the purified echo pulse sequence and the predefined decoding matrix to calculate the correlation coefficient sequence of each pulse.

[0174] The predefined decoding matrix is ​​a matrix pre-constructed according to the encoding rules of the laser pulse sequence encoded at the transmitter. The matrix elements correspond to parameters such as the amplitude and phase of the encoded pulse and are used to perform correlation matching with the echo pulse sequence. Convolution operation is a signal processing algorithm used to calculate the similarity between two signals. The correlation coefficient sequence refers to the sequence data obtained by convolution operation, which characterizes the similarity between the purified echo pulse sequence and each element of the decoding matrix. Each value in the sequence corresponds to the correlation strength at a certain position.

[0175] In this embodiment, firstly, the encoding rule used for the encoded laser pulse sequence at the transmitting end is determined, and a corresponding decoding matrix is ​​constructed according to the rule to ensure that the decoding matrix matches the encoding rule. Subsequently, the purified echo pulse sequence is used as the input signal and convolved with the predefined decoding matrix. By using a point-by-point sliding matching method, the similarity between the input signal and the decoding matrix at each position is calculated, and finally, the correlation coefficient sequence corresponding to each pulse is output.

[0176] The core calculation rule of convolution operation is as follows: For a purified echo pulse sequence X of length N and a decoding matrix W of dimension M×M, take each element of the decoding matrix and multiply it with the corresponding element in the echo pulse sequence, and then sum all the product results to obtain one element of the convolution result. Calculate the result point by point according to this rule to obtain the complete correlation coefficient sequence.

[0177] For example, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, the purified echo pulse sequence is a one-dimensional data sequence with a length of 1000. , The transmitter uses a 4-bit pseudo-random encoding rule, and the corresponding predefined decoding matrix W is... The matrix, the matrix elements are , Perform a convolution operation between sequence X and matrix W. Taking the calculation of the first correlation coefficient as an example, take the first 4 elements of X. With the first row element of W Multiplying corresponding products, we get The result is then used as the first correlation coefficient; subsequently, matrix W is shifted one position to the right to obtain the correlation coefficient of X. The second correlation coefficient is calculated by combining it with the first row of elements of W. This process is repeated for the entire sequence sliding calculation, ultimately resulting in a correlation coefficient sequence of length 997. The above example is merely one illustration of this application. In practical applications, the dimension of the decoding matrix can be adjusted according to the complexity of the encoding rules, and this application does not impose any limitations on this.

[0178] S1062. Based on a preset correlation coefficient threshold, select all peak points exceeding the correlation coefficient threshold from the correlation coefficient sequence, and mark the pulse corresponding to each peak point as a valid pulse.

[0179] In this step, the correlation coefficient threshold is a critical value set in advance based on the detection scenario and signal characteristics, used to distinguish between valid correlation signals and interfering correlation signals; the peak point refers to the position in the correlation coefficient sequence where the value is higher than the adjacent surrounding points, indicating that the echo pulse at that position has the highest matching degree with the decoding matrix; the valid pulse refers to the echo pulse that has been screened by the threshold and determined to truly reflect atmospheric information, excluding the influence of interfering pulses.

[0180] Specifically, firstly, a reasonable correlation coefficient threshold is set based on actual scenario parameters such as the detection range of the pulsed lidar and the complexity of the atmospheric environment. Then, the entire correlation coefficient sequence is traversed to identify all peak points in the sequence, and the value of each peak point is compared with the preset correlation coefficient threshold. Peak points whose values ​​exceed the threshold are selected, and the pulses corresponding to these peak points in the purified echo pulse sequence are marked as valid pulses.

[0181] In practical applications, continuing the previous scenario of pulsed lidar atmospheric wind speed detection, a correlation coefficient threshold of 0.6 was set based on the detection environment. Traversing a correlation coefficient sequence of length 997, eight peak points were identified, with corresponding correlation coefficients of 0.65, 0.58, 0.72, 0.63, 0.55, 0.78, 0.61, and 0.52. These values ​​were compared with the threshold of 0.6, and five peak points exceeding 0.6 were selected: 0.65, 0.72, 0.63, 0.78, and 0.61. The five pulses corresponding to these five peak points in the purified echo pulse sequence were marked as valid pulses, while the remaining three pulses that did not reach the threshold were excluded.

[0182] S1063. Based on the time delay of each effective pulse in the echo pulse sequence, all the effective pulses are divided into different time delay groups, and each time delay group corresponds to a specific distance segment.

[0183] Among them, the time delay refers to the time difference between the effective pulse and the laser emission time, that is, the flight time of the laser pulse from emission to reception; the time delay group refers to grouping effective pulses with the same or similar time delay into a group, and the pulses in the same group correspond to the same distance segment on the detection path; the distance segment refers to the segmented area of ​​the entire predetermined detection path according to the time delay, and each area corresponds to a time delay group.

[0184] Specifically, firstly, the time delay for each valid pulse is determined using timestamp information, calculated by comparing the timestamp of the valid pulse in the echo pulse sequence with the timestamp of the laser emission. Then, based on the total length of the detection path and the preset number of segments, the time delay range for each time delay group is determined, and the time delay range and distance segment are correlated using the speed of light formula. Finally, the time delay for each valid pulse is matched with the time delay range of each group, assigning it to the corresponding time delay group. Each time delay group corresponds to a specific distance segment on the detection path.

[0185] The correspondence between the time delay and the distance segment is calculated using the following formula, as shown in equation (6):

[0186] (6)

[0187] Where d is the distance corresponding to the effective pulse, and c is the speed of light, with a value of [value missing]. , This represents the time delay of the effective pulse. This formula allows us to determine the corresponding distance based on the time delay, and thus divide the distance into segments.

[0188] For example, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, the total length of the predetermined detection path is 1000 meters, pre-divided into 5 range segments, each corresponding to a time delay group. The time delay amounts of the 5 effective pulses are calculated using timestamps. , Calculate the distance corresponding to each time delay according to equation (6): When hour, rice; when hour, The distances corresponding to the other three effective pulses were calculated to be 1050 meters, 1350 meters, and 1650 meters, respectively.

[0189] The time delay ranges for the 5 time delay groups are set as follows: The corresponding distance segments are as follows: rice, rice, rice, rice, Meters. The five valid pulses are divided into corresponding time delay groups to complete the grouping. The above example is merely one example of this application; in practical applications, the number and range of range segments can be adjusted according to the detection resolution requirements, and this application does not limit this.

[0190] S1064. Within each time delay group, compare the amplitude values ​​of all valid pulses and select the valid pulse with the largest amplitude value as the independent echo pulse corresponding to the distance segment.

[0191] In this step, the amplitude value refers to the signal amplitude of the effective pulse, which reflects the signal strength of the echo pulse; the effective pulse with the maximum amplitude value refers to the effective pulse with the largest signal amplitude within the same time delay group. This pulse has the best signal quality and can most accurately reflect the atmospheric information of the corresponding distance segment.

[0192] Specifically, firstly, each time delay group is traversed, and the amplitude values ​​of all valid pulses within the group are extracted. Then, the amplitude values ​​within the same group are compared one by one, and the valid pulse with the largest amplitude value is selected. Finally, the valid pulse with the largest amplitude value selected within each time delay group is determined as the independent echo pulse for the corresponding distance segment of that group, thus completing the separation of multiple independent echo pulses.

[0193] In another specific implementation, the effective pulse amplitude values ​​within the same time delay group can be smoothed and filtered first to remove minor fluctuations in the amplitude values. Then, the amplitude values ​​can be compared and the effective pulse with the largest amplitude value can be selected, which can further improve the selection accuracy of independent echo pulses.

[0194] For example, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, assuming that in the five time delay groups, the first three groups each contain two valid pulses, and the last two groups each contain one valid pulse. In the first group, the amplitude values ​​of the two valid pulses are 0.8V and 0.6V respectively; after comparison, the valid pulse corresponding to 0.8V is selected as the independent echo pulse for the 300-600 meter range. In the second group, the amplitude values ​​of the two valid pulses are 0.75V and 0.9V respectively; the valid pulse corresponding to 0.9V is selected as the independent echo pulse for the 600-900 meter range. In the third group, the amplitude values ​​of the two valid pulses are 0.85V and 0.7V respectively; the valid pulse corresponding to 0.85V is selected as the independent echo pulse for the 900-1200 meter range. The single valid pulse in each of the last two groups is directly used as the independent echo pulse for its corresponding range. This results in five independent echo pulses corresponding to different ranges.

[0195] This application achieves precise correlation matching, threshold filtering to eliminate interfering pulses, time-delay grouping and associating distance segments, and maximum amplitude pulse selection through decoding matrix and convolution operations. This solves the problems of traditional decoding methods, such as susceptibility to interference, inaccurate independent pulse separation, and ambiguous distance segment correspondences. Compared to traditional schemes, it can more accurately separate the independent echo pulses corresponding to each distance segment, improving the reliability and accuracy of pulse separation.

[0196] S107. Calculate the wind speed for each distance segment based on the frequency offset of the multiple independent echo pulses.

[0197] In this step, the frequency offset refers to the difference between the frequency of an independent echo pulse and the original frequency of the laser emitted by the lidar. In the atmospheric wind speed detection scenario of pulse lidar, this offset is generated by the scattering of the laser by aerosol particles or molecules moving in the atmosphere. Its magnitude is fixedly correlated with the wind speed of the corresponding distance segment and can be used as the core basis for wind speed calculation.

[0198] In one specific implementation, firstly, the frequency offset of each independent echo pulse is extracted using signal processing techniques. Specifically, the frequency of each independent echo pulse is compared with the original frequency of the emitted laser, and the difference between the two is calculated to obtain the frequency offset. Then, using the wind speed calculation relationship based on the Doppler effect, the extracted frequency offset is converted into the wind speed for the corresponding distance segment. The core calculation relationship used is shown in equation (7):

[0199] (7)

[0200] in, This represents the wind speed for the corresponding distance segment. The wavelength of the laser emitted by the lidar (system preset parameter). This represents the frequency offset of an independent echo pulse. By substituting the preset laser wavelength parameters and the extracted frequency offset, the wind speed for each distance segment can be calculated.

[0201] For example, continuing with the pulsed lidar atmospheric wind speed detection scenario described earlier, the industry-standard 905nm laser wavelength is selected as the system's preset parameter. This wavelength is moderately priced and suitable for conventional atmospheric detection needs. Assume that the frequency offset of independent echo pulses within a certain range is extracted using signal processing techniques. The frequency is 306.7MHz. Substituting the parameters into equation (7) for calculation: First, clarify the parameter values, , Calculation process: First, calculate the molecule: Divide by 2 to get The result is the wind speed over that distance. The above example is merely one example of this application. In practical applications, the laser wavelength can be selected from other specifications such as 1550nm according to the detection requirements, and the frequency offset is determined by the actual detected echo signal. This application does not limit this.

[0202] The wind speed calculation method based on pulsed lidar echo signals provided in this application effectively solves the practical problems in traditional lidar wind speed detection, such as signal distortion caused by atmospheric turbulence, severe noise interference, and inaccurate wind speed calculation by distance segment. Compared with traditional solutions, it significantly improves the reliability and accuracy of wind speed detection at different distance segments without complex hardware upgrades, enhances the system's adaptability to complex atmospheric environments, and ensures the authenticity and validity of wind speed detection data.

[0203] Figure 3 A schematic diagram illustrating a specific implementation of a wind speed calculation system based on pulsed lidar echo signals provided in this application, referring to... Figure 3 The system may include:

[0204] Acquisition module 31 is used to acquire atmospheric turbulence data along a predetermined detection path;

[0205] The generation module 32 is used to generate an emitted laser pulse sequence by performing wavefront modulation on the encoded laser pulse sequence through tunable optical elements based on the atmospheric turbulence data.

[0206] The transmitting module 33 is used to transmit the transmitted laser pulse sequence along the predetermined detection path;

[0207] Conversion module 34 is used to receive the echo signal returned by the emitted laser pulse sequence on the predetermined detection path and convert the echo signal into an initial electrical signal;

[0208] The purification module 35 is used to input the initial electrical signal into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal.

[0209] The separation module 36 is used to decode the purified echo pulse sequence to separate multiple independent echo pulses, each of which corresponds to a different distance segment.

[0210] The calculation module 37 is used to calculate the wind speed for each distance segment based on the frequency offset of the multiple independent echo pulses.

[0211] The wind speed calculation system based on pulsed lidar echo signals in this application embodiment is used to implement the aforementioned wind speed calculation method based on pulsed lidar echo signals. Therefore, the specific implementation of the wind speed calculation system based on pulsed lidar echo signals can be found in the embodiment section of the wind speed calculation method based on pulsed lidar echo signals above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0212] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described wind speed calculation methods based on pulsed lidar echo signals.

[0213] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described wind speed calculation methods based on pulsed lidar echo signals.

[0214] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0215] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the wind speed calculation method based on pulsed lidar echo signals.

[0216] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0217] The foregoing has provided a detailed description of the wind speed calculation method and system based on pulsed lidar echo signals provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for calculating wind speed based on pulsed lidar echo signals, characterized in that, include: Acquire atmospheric turbulence data along the predetermined detection path; Based on the atmospheric turbulence data, the encoded laser pulse sequence is wavefront modulated using tunable optical elements to generate an emitted laser pulse sequence; The sequence of emitted laser pulses is emitted along the predetermined detection path; Receive the echo signal returned by the emitted laser pulse sequence on the predetermined detection path, and convert the echo signal into an initial electrical signal; The initial electrical signal is input into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal. The purified echo pulse sequence is decoded to separate multiple independent echo pulses, each corresponding to a different distance segment; The wind speed for each distance segment is calculated based on the frequency offset of the multiple independent echo pulses. The process of generating an emitted laser pulse sequence by wavefront modulation of a coded laser pulse sequence using a tunable optical element based on the atmospheric turbulence data includes: resolving turbulence intensity distribution and wind speed gradient information from the atmospheric turbulence data; and calculating the phase compensation amount and amplitude scaling factor required by the tunable optical element based on the turbulence intensity distribution and the wind speed gradient information. Specifically, based on the turbulence intensity distribution and the wind speed gradient information, the required phase compensation and amplitude scaling factor for the tunable optical element are calculated, including: The turbulence intensity distribution is divided along the beam propagation path according to different altitude layers to obtain the turbulence intensity value at each altitude layer. Based on the wind speed gradient information, the atmospheric flow velocity vector and direction at each altitude layer are determined. A mathematical relationship model between the beam phase change and the turbulence intensity value at each altitude layer is established based on wave optics theory. The turbulence intensity value and the atmospheric flow velocity vector at each altitude layer are input into the mathematical relationship model to calculate the cumulative phase distortion of the beam after passing through the atmospheric channel. Based on the cumulative phase distortion, a global phase adjustment amount is determined as the phase compensation amount. Simultaneously, based on the scattering and attenuation effect of the turbulence intensity value on the beam energy, the beam intensity attenuation ratio is calculated. Based on the beam intensity attenuation ratio, a signal gain adjustment coefficient is determined as the amplitude scaling factor.

2. The method according to claim 1, characterized in that, Based on the atmospheric turbulence data, a wavefront modulation of the coded laser pulse sequence is performed using tunable optical elements to generate an emitted laser pulse sequence, including: Based on the phase compensation amount and the amplitude scaling factor, a control signal sequence for the tunable optical element is generated; The control signal sequence is used to drive the tunable optical element to adjust the wavefront shape of each laser pulse in the encoded laser pulse sequence pulse by pulse. The encoded laser pulse sequence has a predetermined encoding rule, wherein the width of a single pulse is used to determine the range resolution of the system, and the total energy of the entire pulse sequence is used to maintain long-range detection; The adjusted pulse sequences are combined in chronological order to form an emitted laser pulse sequence.

3. The method according to claim 1, characterized in that, The initial electrical signal is input into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal, including: The initial electrical signal is divided into multiple overlapping time window segments, and the signal data in each time window segment is analyzed in the time domain and frequency domain to obtain the feature vectors in the time domain and frequency domain. The time-domain and frequency-domain feature vectors are input into a pre-constructed neural network. Using the hidden layers of the neural network, the absolute difference between different feature dimensions in the feature vectors of the time domain and frequency domain is calculated to generate an interaction matrix representing the relationship between features; The elements in the interaction matrix are expanded in a predetermined order to form a transformed feature vector, and a predefined activation function is applied to the transformed feature vector to perform a nonlinear transformation. Based on the transformed feature vector, the denoised signal segments of the corresponding time window are reconstructed. All the denoised signal segments of the time window are spliced ​​together in chronological order to generate a purified echo pulse sequence.

4. The method according to claim 1, characterized in that, The purified echo pulse sequence is decoded to separate multiple independent echo pulses, each corresponding to a different distance segment, including: The purified echo pulse sequence is convolved with a predefined decoding matrix to calculate the correlation coefficient sequence for each pulse. Based on a preset correlation coefficient threshold, all peak points exceeding the correlation coefficient threshold are selected from the correlation coefficient sequence, and the pulse corresponding to each peak point is marked as a valid pulse. Based on the time delay of each effective pulse in the echo pulse sequence, all the effective pulses are divided into different time delay groups, and each time delay group corresponds to a specific distance segment; Within each time delay group, the amplitude values ​​of all valid pulses are compared, and the valid pulse with the largest amplitude value is selected as the independent echo pulse corresponding to the distance segment.

5. The method according to claim 3, characterized in that, Using the hidden layers of the neural network, the absolute differences between different feature dimensions in the feature vectors of the time and frequency domains are calculated to generate an interaction matrix representing the relationship between features, including: Based on the number of dimensions of the feature vectors in the time domain and frequency domain, determine the range of dimension pairs to be calculated, and perform an absolute difference operation on each dimension of the feature vectors in the time domain and frequency domain to obtain the difference data for each pair of dimensions. The difference data is filled into the initial matrix according to the row and column organization, where the row index corresponds to the time domain dimension index and the column index corresponds to the frequency domain dimension index. Adjust the element layout of the initial matrix to conform to the preset matrix format, and output an interactive matrix.

6. The method according to claim 1, characterized in that, Receiving the echo signal returned by the emitted laser pulse sequence on the predetermined detection path, and converting the echo signal into an initial electrical signal, includes: The echo signal returned along the predetermined detection path is captured by an optical receiving device, and the focused echo signal is converted into an analog electrical signal by a photoelectric converter. The analog electrical signal is amplified and filtered, and the processed analog electrical signal is sampled into a digital signal sequence through an analog-to-digital converter; The digital signal sequence is timestamped to generate an initial electrical signal.

7. A wind speed calculation system based on pulsed lidar echo signals, used to execute the wind speed calculation method based on pulsed lidar echo signals as described in any one of claims 1 to 6, characterized in that, include: The acquisition module is used to acquire atmospheric turbulence data along the predetermined detection path; The generation module is used to generate an emitted laser pulse sequence by wavefront modulation of the encoded laser pulse sequence using tunable optical elements based on the atmospheric turbulence data. The transmitting module is used to transmit the laser pulse sequence along the predetermined detection path; A conversion module is used to receive the echo signal returned by the emitted laser pulse sequence on the predetermined detection path and convert the echo signal into an initial electrical signal; A purification module is used to input the initial electrical signal into a pre-constructed neural network to obtain a purified echo pulse sequence; the neural network is used to process the distortion and noise contained in the initial electrical signal. The separation module is used to decode the purified echo pulse sequence to separate multiple independent echo pulses, each of which corresponds to a different distance segment. The calculation module is used to calculate the wind speed for each distance segment based on the frequency offset of the multiple independent echo pulses.

8. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the wind speed calculation method based on pulsed lidar echo signals as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the wind speed calculation method based on pulsed lidar echo signals as described in any one of claims 1 to 6.

Images