A Deep Learning-Based Method and System for Wind Profiler Radar Boundary Layer Height Inversion
By combining deep learning convolutional neural networks with multiple atmospheric parameter profiles, the uncertainty problem of boundary layer height inversion under complex atmospheric conditions in traditional algorithms has been solved, achieving high-precision and stable boundary layer height monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WUHAN UNIV
- Filing Date
- 2024-07-22
- Publication Date
- 2026-06-30
AI Technical Summary
The traditional algorithms for measuring boundary layer height using wind profiler radar in existing technologies are cumbersome and have high uncertainty in the results, making it difficult to accurately invert the boundary layer height under complex atmospheric conditions.
By employing a deep learning-based convolutional neural network (CNN) combined with multiple atmospheric parameter profiles, and through preprocessing, training dataset construction, and loss function optimization, a high-precision boundary layer height inversion model is output.
It significantly improves the accuracy and stability of boundary layer height inversion, reduces the sensitivity to local peaks, and has higher robustness and accuracy, enabling precise capture of diurnal variations in the boundary layer.
Smart Images
Figure CN118915006B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of atmospheric remote sensing technology, and in particular to a method and system for inverting boundary layer height using wind profiler radar based on deep learning. Background Technology
[0002] The atmospheric boundary layer is the lowest layer of the troposphere, closest to the Earth's surface. It is in direct contact with the ground and strongly influenced by it. Forced processes such as surface friction drag, evaporation and transpiration, and heat transfer lead to frequent turbulent motion in this layer. Turbulent transport provides important channels for the exchange of heat, momentum, and water vapor. Furthermore, the emission, transport, and transformation of pollutants caused by human activities are also affected by the boundary layer. Therefore, environmental problems in the atmospheric boundary layer directly impact human activities and health. Boundary layer height (BLH) is a crucial parameter characterizing the atmospheric boundary layer and is an important physical parameter in numerical weather prediction models. However, the boundary layer is easily affected by surface thermal and dynamic factors, and its response time is less than one hour, resulting in significant diurnal variations in boundary layer height. After sunrise, solar radiation heating intensifies turbulence, and meteorological elements within the boundary layer mix uniformly, forming a convective boundary layer. After sunset, surface radiation cooling reduces the contribution of buoyancy, forming a neutral boundary layer that maintains a neutral stratification. When nighttime radiative cooling is sufficiently strong, turbulent transport weakens, inversion stratification occurs, and a stable boundary layer is formed. Therefore, continuous and accurate boundary layer height inversion is very challenging.
[0003] Wind profiler radar not only provides continuous boundary layer height estimates but also operates unattended for extended periods under all weather conditions. Therefore, wind profiler radar shows great potential in boundary layer height retrieval. Based on wind profiler radar measurements, a series of boundary layer height estimation algorithms have been developed, including thresholding, peak detection, wavelet transform, and wind profiler methods. However, these methods all use only one atmospheric profile to retrieve boundary layer height, such as the signal-to-noise ratio profile or the wind profile. Other atmospheric profiles are only used as auxiliary parameters to handle situations where multiple possible boundary layer heights exist simultaneously. When multiple local peaks exist in the profile, these methods are prone to introducing significant uncertainties. Although the fuzzy logic method improves the algorithm by simultaneously considering the signal-to-noise ratio, vertical velocity variance, and Doppler spectral width, it requires constructing complex logic functions, making it cumbersome in practical applications.
[0004] Therefore, a more convenient and accurate adaptive boundary layer height inversion algorithm is needed. Summary of the Invention
[0005] This invention provides a method and system for inverting the boundary layer height of wind profiler radar based on deep learning, which solves the defects of traditional algorithms used in wind profiler radar measurement, such as cumbersome steps and high uncertainty of measurement results.
[0006] In a first aspect, the present invention provides a deep learning-based method for inverting the boundary layer height of wind profiler radar, comprising:
[0007] Acquire wind profiler radar product data, preprocess the wind profiler radar product data and perform time matching to obtain a training dataset;
[0008] By filtering multiple parameter profiles from wind profiler radar products, the parameter profile for retrieving the boundary layer height is obtained.
[0009] Based on the training dataset and the parameter profile of the inverted boundary layer height, a convolutional neural network is trained, and the convolutional neural network is converged using a loss function to output a wind profile radar boundary layer height inversion model.
[0010] The boundary layer height inversion results were obtained using the aforementioned wind profiler radar boundary layer height inversion model.
[0011] According to the present invention, a deep learning-based method for inverting the boundary layer height of a wind profiler radar is provided. This method acquires wind profiler radar product data, preprocesses and performs time matching on the wind profiler radar product data to obtain a training dataset, including:
[0012] The wind profiler radar product data is filtered to obtain wind profiler data with preset completeness and preset high quality. The wind profiler data is then organized based on a preset time interval to obtain filtered wind profiler data.
[0013] The boundary layer height obtained from the sounding inversion is time-matched with the selected wind profile data to form the training dataset.
[0014] According to the present invention, a deep learning-based method for inverting boundary layer height from wind profiler radar is provided, which filters multiple parameter profiles in wind profiler radar products to obtain parameter profiles for inverting boundary layer height, including:
[0015] Based on the principle of boundary layer physical characteristics, atmospheric profile parameters in wind profiler radar products are screened out. Then, candidate atmospheric profile parameters are obtained by the correlation between the atmospheric profile parameters and the boundary layer height obtained by sounding inversion at the same time.
[0016] The selected atmospheric profile parameters are input into the inversion model, and the boundary layer height obtained by sounding inversion is used as a reference value to evaluate the model accuracy.
[0017] By randomly adjusting the order of any atmospheric parameter profiles and inputting the resulting new atmospheric parameter profiles into the inversion model, a new model accuracy can be obtained.
[0018] The sequence of all atmospheric parameter profiles was repeatedly adjusted and the model was inverted to obtain multiple model accuracies. All model accuracies were normalized and sorted, and the parameter profiles that participated in the inversion of boundary layer height were selected according to their importance.
[0019] According to the present invention, a deep learning-based method for inverting the boundary layer height of a wind profile radar is provided, which trains a convolutional neural network based on the training dataset and the parameter profile of the inverted boundary layer height, including:
[0020] The convolutional neural network is constructed, which includes several one-dimensional convolutional kernels. The activation function is the ReLU function, and the loss function is constructed from the root mean square error of the boundary layer height calculated by the model and the true boundary height.
[0021] The training dataset is divided into a training set, a validation set, and a test set according to a preset ratio;
[0022] The convolutional neural network is trained using the training set.
[0023] According to the present invention, a deep learning-based method for inverting the boundary layer height of a wind profiler radar is provided. This method utilizes a loss function to converge the convolutional neural network and outputs a wind profiler radar boundary layer height inversion model, comprising:
[0024] The training results are judged based on the changes in the loss function to avoid underfitting and overfitting of the training results;
[0025] The hyperparameters determined during the training process include several kernel sizes and several kernel numbers;
[0026] The model with the minimum loss function is selected as the wind profiler radar boundary layer height inversion model.
[0027] According to the deep learning-based wind profiler radar boundary layer height inversion method provided by the present invention, after converging the convolutional neural network using a loss function and outputting the wind profiler radar boundary layer height inversion model, the method further includes:
[0028] The wind profiler radar boundary layer height inversion model was tested using the test set.
[0029] The accuracy of the model is quantitatively assessed using correlation coefficient, mean absolute error, and root mean square error.
[0030] Secondly, the present invention also provides a deep learning-based wind profiler radar boundary layer height inversion system, comprising:
[0031] The preprocessing module is used to acquire wind profiler radar product data, perform preprocessing and time matching on the wind profiler radar product data, and obtain a training dataset.
[0032] The filtering module is used to filter multiple parameter profiles in wind profiler radar products to obtain the parameter profiles for inverting the boundary layer height.
[0033] The training module is used to train the convolutional neural network based on the training dataset and the parameter profile of the inverted boundary layer height, and to converge the convolutional neural network using a loss function to output the wind profile radar boundary layer height inversion model.
[0034] The inversion module is used to obtain the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
[0035] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the deep learning-based wind profiler radar boundary layer height inversion method as described above.
[0036] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the deep learning-based wind profiler radar boundary layer height inversion method as described above.
[0037] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the deep learning-based wind profiler radar boundary layer height inversion method as described above.
[0038] The present invention provides a deep learning-based wind profiler radar boundary layer height inversion method and system. By combining the advantages of deep learning in discovering complex structures in high-dimensional data, and incorporating multiple atmospheric parameter profiles, it greatly improves the accuracy and stability of boundary layer height inversion, providing strong support for boundary layer research and global climate change. The proposed convolutional neural network inversion algorithm significantly improves the accuracy of boundary layer inversion, overcomes the disadvantages of multiple local peak effects, has higher robustness, and has better performance than traditional algorithms. Attached Figure Description
[0039] To more clearly illustrate the technical solutions in this invention 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0040] Figure 1 This is one of the flowcharts illustrating the deep learning-based wind profiler radar boundary layer height inversion method provided by this invention.
[0041] Figure 2 This is the second flowchart of the deep learning-based wind profiler radar boundary layer height inversion method provided by the present invention.
[0042] Figure 3 This is a structural diagram of the boundary layer height convolutional neural network provided by the present invention;
[0043] Figure 4 This is a schematic diagram of the structure of the wind profiler radar boundary layer height inversion system provided by the present invention.
[0044] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0046] To address the shortcomings of existing technologies for wind profiler radar measurements, such as significant errors in boundary layer height obtained using traditional algorithms under different local peak values, and the strong sensitivity of boundary layer height retrieved using single profile parameters to the number of local peak values under different atmospheric conditions, this invention employs a convolutional neural network to perform inversion calculations using multiple atmospheric parameter profiles. This solves the problem of poor accuracy of traditional algorithms under complex atmospheric conditions, accurately capturing diurnal variations in boundary layer height and providing important support for boundary layer and climate change research.
[0047] Figure 1 This is one of the flowcharts illustrating the deep learning-based wind profiler radar boundary layer height inversion method provided in this embodiment of the invention. Figure 1 As shown, it includes:
[0048] Step 100: Obtain wind profiler radar product data, preprocess the wind profiler radar product data and perform time matching to obtain a training dataset;
[0049] Step 200: Filter multiple parameter profiles in the wind profiler radar product to obtain the parameter profile for retrieving the boundary layer height;
[0050] Step 300: Based on the training dataset and the parameter profile of the inverted boundary layer height, train the convolutional neural network, use the loss function to converge the convolutional neural network, and output the wind profile radar boundary layer height inversion model;
[0051] Step 400: Obtain the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
[0052] Specifically, the algorithm steps of this embodiment of the invention are as follows: Figure 2 As shown, uncertainty analysis is performed on the number of local peaks and different atmospheric conditions, followed by parameter selection, and then a convolutional neural network model is obtained through parameter importance and sensitivity analysis. Finally, the obtained model is used for high-precision boundary layer height monitoring. Specifically, this includes:
[0053] Data from wind profiler radar products were preprocessed to select complete and high-quality profile data, and the dataset was organized into hourly units. Furthermore, the boundary layer height retrieved from sounding inversion was time-matched with the wind profiler radar data to create a complete training dataset.
[0054] By combining physical characteristics and importance, numerous parameter profiles in wind profiler radar products are screened and ranked by importance to determine the parameter profiles used for inverting boundary layer height.
[0055] The selected dataset was used as input to the model, and the boundary layer height obtained from sounding inversion was used as the ground truth to train the convolutional neural network model. The model structure is as follows: Figure 3 As shown, the convolutional kernel is an N×1 one-dimensional convolutional kernel, the activation function is the ReLU function, and the loss function is the root mean square error between the calculated boundary layer height and the true boundary layer height. The training, validation, and test datasets are divided in a 7:1:2 ratio, and the boundary layer height inversion convolutional neural network is trained using the training dataset.
[0056] The quality of the training result is judged by the change in the loss function during training, avoiding underfitting and overfitting. Hyperparameters used in the training process include kernel size (3, 5, 7, 9, 11, 13) and the number of kernels (8, 16, 32, 64, 128, 256). The model with the smallest loss function is selected as the optimal model by setting the kernel size and number of kernels to the above values.
[0057] The optimal model trained in the previous step is tested using a test dataset, and its accuracy is quantitatively evaluated using correlation coefficient, mean absolute error, and root mean square error. Finally, this boundary layer height convolutional neural network model can be used to obtain high-precision boundary layer height inversion results, accurately capturing the diurnal variation of the boundary layer.
[0058] The Convolutional Neural Networks (CNN) model used in this invention has two main advantages compared to traditional algorithms, which suffer from high uncertainty when retrieving boundary layer height. First, the CNN algorithm incorporates more parameters characterizing the boundary layer height, such as spectral width and vertical velocity standard deviation. When traditional algorithms struggle to determine the boundary layer height due to multiple local peaks in the signal-to-noise ratio (SNR) profile, the CNN algorithm can combine other parameter profiles to jointly determine the boundary layer height, thereby reducing the impact of multiple local peaks. Sensitivity analysis of the CNN algorithm and two traditional algorithms to the number of local peaks in the SNR profile shows that the error of the traditional algorithm increases with the number of local peaks. In contrast, the CNN algorithm maintains a lower error even with multiple local peaks, exhibiting better robustness. Specifically, in stable boundary layers, when the number of local peaks is 1, 2, 3, and greater than 3, the error of the CNN algorithm is reduced by 35%, 50%, 69%, and 72% compared to the traditional algorithm, respectively.
[0059] Furthermore, the CNN algorithm learns complex high-order features of all input parameter profiles through the convolution process. Traditional algorithms consider only one type of feature, such as maximum value or gradient, while the CNN algorithm extracts different features through different convolution kernels and uses all of these features to estimate the boundary layer height. Statistical analysis shows that the CNN algorithm has high consistency with the boundary layer height obtained from radiosonde inversion, with correlation coefficients, MAE, and RMSE of 0.81, 0.24 km, and 0.34 km, respectively. Compared with traditional algorithms, the MAE of the CNN algorithm decreases from 0.57±0.60 km to 0.24±0.25 km, and the RMSE decreases from 0.83 km to 0.34 km.
[0060] In some embodiments, wind profiler radar product data is acquired, and the wind profiler radar product data is preprocessed and time-matched to obtain a training dataset, including:
[0061] The wind profiler radar product data is filtered to obtain wind profiler data with preset completeness and preset high quality. The wind profiler data is then organized based on a preset time interval to obtain filtered wind profiler data.
[0062] The boundary layer height obtained from the sounding inversion is time-matched with the selected wind profile data to form the training dataset.
[0063] To analyze the uncertainties in traditional algorithms, the number of local peaks under different signal-to-noise ratio profiles is counted. The mean absolute error (MAE) of the boundary layer height retrieved by the traditional algorithm and the boundary layer height retrieved by sounding is calculated under different numbers of local peaks, and this error is used as an indicator to measure the algorithm's sensitivity to the number of local peaks.
[0064]
[0065] in Indicates the number of samples. This represents the boundary layer height obtained by the traditional method. This represents the boundary layer height obtained through radiosonde inversion. Here, a high MAE indicates that traditional methods are sensitive to the number of local peaks when retrieving boundary layer height, and are easily affected by them.
[0066] This invention employs a method combining physical mechanisms and the importance of arrangement to select numerous atmospheric profile parameters from wind profiler radar products, determine atmospheric profiles that are highly correlated with boundary layer height inversion, and use these as inputs for subsequent inversion models.
[0067] In some embodiments, filtering multiple parameter profiles in a wind profiler radar product to obtain a parameter profile for retrieving the boundary layer height includes:
[0068] Based on the principle of boundary layer physical characteristics, atmospheric profile parameters in wind profiler radar products are screened out. Then, candidate atmospheric profile parameters are obtained by the correlation between the atmospheric profile parameters and the boundary layer height obtained by sounding inversion at the same time.
[0069] The selected atmospheric profile parameters are input into the inversion model, and the boundary layer height obtained by sounding inversion is used as a reference value to evaluate the model accuracy.
[0070] By randomly adjusting the order of any atmospheric parameter profiles and inputting the resulting new atmospheric parameter profiles into the inversion model, a new model accuracy can be obtained.
[0071] The sequence of all atmospheric parameter profiles was repeatedly adjusted and the model was inverted to obtain multiple model accuracies. All model accuracies were normalized and sorted, and the parameter profiles that participated in the inversion of boundary layer height were selected according to their importance.
[0072] Specifically, in this embodiment of the invention, a preliminary analysis and interpretation of the correlation between various atmospheric parameter profiles and boundary layer height in wind profiler radar products is performed from a physical perspective, and the importance of each parameter is quantitatively evaluated using the importance ranking method. This is achieved through the following steps:
[0073] (1) Based on the principle of boundary layer physical characteristics, the atmospheric profile parameters in the wind profile radar products are initially screened, and cases of these atmospheric profile parameters and the boundary layer height obtained by sounding inversion at the same time are drawn to complete the initial screening.
[0074] (2) Input the atmospheric parameter profiles obtained from the initial screening into the inversion model, and use the boundary layer height obtained from the sounding as a reference value to evaluate the accuracy of the model.
[0075] (3) Randomly shuffle the profile of a certain parameter, obtain a new profile and input it into the model, and calculate the change in model accuracy as an evaluation index of importance.
[0076] (4) Repeat the previous step for all candidate atmospheric parameter profiles, normalize and sort the accuracy changes of each parameter profile after shuffling, and select parameter profiles according to their importance to complete the selection of parameter profiles.
[0077] This invention adds more parameter profiles to the signal-to-noise ratio profile, including spectral width, vertical velocity standard deviation, and horizontal wind speed and direction. When traditional algorithms struggle to determine the boundary layer height due to multiple local peaks in the signal-to-noise ratio profile, the CNN algorithm can combine other parameter profiles to jointly determine the boundary layer height, thereby reducing the impact of multiple local peaks.
[0078] In some embodiments, training a convolutional neural network based on the training dataset and the parameter profile of the inverted boundary layer height includes:
[0079] The convolutional neural network is constructed, which includes several one-dimensional convolutional kernels. The activation function is the ReLU function, and the loss function is constructed from the root mean square error of the boundary layer height calculated by the model and the true boundary height.
[0080] The training dataset is divided into a training set, a validation set, and a test set according to a preset ratio;
[0081] The convolutional neural network is trained using the training set.
[0082] The convolutional neural network is converged using a loss function, and the resulting wind profile radar boundary layer height inversion model is output, including:
[0083] The training results are judged based on the changes in the loss function to avoid underfitting and overfitting of the training results;
[0084] The hyperparameters determined during the training process include several kernel sizes and several kernel numbers;
[0085] The model with the minimum loss function is selected as the wind profiler radar boundary layer height inversion model.
[0086] Specifically, this embodiment of the invention studies a boundary layer height inversion algorithm based on convolutional neural networks, incorporating multiple atmospheric parameter profiles into the inversion model to solve the problem of poor accuracy of traditional algorithms under complex atmospheric conditions.
[0087] The input to the network model consists of multiple atmospheric parameter profiles, with the specific structure as follows: Figure 3 As shown, the CNN algorithm consists of two convolutional processes using ReLU activation, followed by a max-pooling operation after each convolutional process. After the two convolutional layers, two fully connected layers output the boundary layer height estimated by the CNN algorithm. Considering the special vertical profile structure of the input variables, one-dimensional convolutional kernels (N×1) are used in the convolutional layers. These kernels can extract complex vertical features from the five input profiles through convolution operations, such as gradients and extrema. Multiple kernels extract a large number of different features, and stacking two convolutional and pooling layers can gradually extract higher-order features. These extracted extensive and complex features are beneficial for the CNN algorithm to invert the boundary layer height. In addition, the network adds two batch normalization layers to accelerate convergence.
[0088] This invention learns complex high-order features of all input parameter profiles through a convolution process. Different complex features are extracted using different convolution kernels, and high-order features are extracted by stacking two convolutions. These complex features are then used to invert the boundary layer height, improving the accuracy and robustness of the inversion.
[0089] In some embodiments, after the convolutional neural network is converged using a loss function and the wind profile radar boundary layer height inversion model is output, the method further includes:
[0090] The wind profiler radar boundary layer height inversion model was tested using the test set.
[0091] The accuracy of the model is quantitatively assessed using correlation coefficient, mean absolute error, and root mean square error.
[0092] Specifically, in this embodiment of the invention, the trained optimal model is tested using a test dataset, and the accuracy of the model is quantitatively evaluated using correlation coefficient, mean absolute error, and root mean square error. Finally, this boundary layer height convolutional neural network model can be used to obtain high-precision boundary layer height inversion results, accurately capturing the diurnal variation of the boundary layer.
[0093] This invention can monitor the diurnal variation of boundary layer height over a long period of time. Through long-term observation, it can analyze the seasonal and interannual variations of boundary layer height at the monitoring location, providing strong data support and analytical tools for climate change research.
[0094] The following describes the deep learning-based wind profiler radar boundary layer height inversion system provided by this invention. The deep learning-based wind profiler radar boundary layer height inversion system described below can be referred to in correspondence with the deep learning-based wind profiler radar boundary layer height inversion method described above.
[0095] Figure 4 This is a schematic diagram of the structure of the deep learning-based wind profiler radar boundary layer height inversion system provided in an embodiment of the present invention, as shown below. Figure 4 As shown, it includes: a preprocessing module 41, a filtering module 42, a training module 43, and an inversion module 44, wherein:
[0096] The preprocessing module 41 is used to acquire wind profiler radar product data, preprocess the wind profiler radar product data and perform time matching to obtain a training dataset; the filtering module 42 is used to filter multiple parameter profiles in the wind profiler radar products to obtain the parameter profiles for inverting the boundary layer height; the training module 43 is used to train a convolutional neural network based on the training dataset and the parameter profiles for inverting the boundary layer height, use a loss function to converge the convolutional neural network, and output a wind profiler radar boundary layer height inversion model; the inversion module 44 is used to obtain the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
[0097] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a deep learning-based wind profiler radar boundary layer height inversion method. This method includes: acquiring wind profiler radar product data; preprocessing and time-matching the wind profiler radar product data to obtain a training dataset; filtering multiple parameter profiles in the wind profiler radar products to obtain parameter profiles for inverting boundary layer height; training a convolutional neural network based on the training dataset and the parameter profiles for inverting boundary layer height; converging the convolutional neural network using a loss function to output a wind profiler radar boundary layer height inversion model; and obtaining the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
[0098] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0099] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the deep learning-based wind profiler radar boundary layer height inversion method provided by the above methods. The method includes: acquiring wind profiler radar product data; preprocessing and time-matching the wind profiler radar product data to obtain a training dataset; filtering multiple parameter profiles in the wind profiler radar products to obtain parameter profiles for inverting boundary layer height; training a convolutional neural network based on the training dataset and the parameter profiles for inverting boundary layer height; converging the convolutional neural network using a loss function to output a wind profiler radar boundary layer height inversion model; and obtaining the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
[0100] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the deep learning-based wind profiler radar boundary layer height inversion method provided by the above methods. The method includes: acquiring wind profiler radar product data; preprocessing and time-matching the wind profiler radar product data to obtain a training dataset; filtering multiple parameter profiles in the wind profiler radar products to obtain parameter profiles for inverting boundary layer height; training a convolutional neural network based on the training dataset and the parameter profiles for inverting boundary layer height; converging the convolutional neural network using a loss function to output a wind profiler radar boundary layer height inversion model; and obtaining the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
[0101] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for retrieving boundary layer height based on deep learning of wind profile radar, characterized in that, include: Acquire wind profiler radar product data, preprocess the wind profiler radar product data and perform time matching to obtain a training dataset; Multiple parameter profiles in wind profiler radar products are screened to obtain the parameter profile for retrieving the boundary layer height. The parameter profile includes the signal-to-noise ratio profile, spectral width, vertical velocity standard deviation, and horizontal wind speed and direction. Based on the training dataset and the parameter profile of the inverted boundary layer height, a convolutional neural network is trained, and the convolutional neural network is converged using a loss function to output a wind profile radar boundary layer height inversion model. The boundary layer height inversion results were obtained using the aforementioned wind profiler radar boundary layer height inversion model.
2. The deep learning based wind profile radar boundary layer height retrieval method according to claim 1, characterized in that, Acquire wind profiler radar product data, preprocess the wind profiler radar product data and perform time matching to obtain a training dataset, including: The wind profiler radar product data is filtered to obtain wind profiler data with preset completeness and preset high quality. The wind profiler data is then organized based on a preset time interval to obtain filtered wind profiler data. The boundary layer height obtained from the sounding inversion is time-matched with the selected wind profile data to form the training dataset.
3. The deep learning based wind profile radar boundary layer height retrieval method according to claim 1, characterized in that, By filtering multiple parameter profiles from wind profiler radar products, the parameter profile for retrieving the boundary layer height is obtained, including: Based on the principle of boundary layer physical characteristics, atmospheric profile parameters in wind profiler radar products are screened out. Then, candidate atmospheric profile parameters are obtained by the correlation between the atmospheric profile parameters and the boundary layer height obtained by sounding inversion at the same time. The selected atmospheric profile parameters are input into the inversion model, and the boundary layer height obtained by sounding inversion is used as a reference value to evaluate the model accuracy. By randomly adjusting the order of any atmospheric parameter profiles and inputting the resulting new atmospheric parameter profiles into the inversion model, a new model accuracy can be obtained. The sequence of all atmospheric parameter profiles was repeatedly adjusted and the model was inverted to obtain multiple model accuracies. All model accuracies were normalized and sorted, and the parameter profiles that participated in the inversion of boundary layer height were selected according to their importance.
4. The deep learning based wind profile radar boundary layer height retrieval method according to claim 1, characterized in that, Based on the training dataset and the parameter profile of the inverted boundary layer height, the convolutional neural network is trained, including: The convolutional neural network is constructed, which includes several one-dimensional convolutional kernels. The activation function is the ReLU function, and the loss function is constructed from the root mean square error of the boundary layer height calculated by the model and the true boundary height. The training dataset is divided into a training set, a validation set, and a test set according to a preset ratio; The convolutional neural network is trained using the training set.
5. The deep learning based wind profile radar boundary layer height retrieval method according to claim 4, characterized in that, The convolutional neural network is converged using a loss function, outputting a wind profile radar boundary layer height inversion model, including: The training results are judged based on the changes in the loss function to avoid underfitting and overfitting of the training results; The hyperparameters determined during the training process include several kernel sizes and several kernel numbers; The model with the minimum loss function is selected as the wind profiler radar boundary layer height inversion model.
6. The deep learning based wind profile radar boundary layer height retrieval method according to claim 5, characterized in that, After the convolutional neural network is converged using a loss function and the wind profile radar boundary layer height inversion model is output, the following steps are also included: The wind profiler radar boundary layer height inversion model was tested using the test set. The accuracy of the model is quantitatively assessed using correlation coefficient, mean absolute error, and root mean square error.
7. A deep learning-based wind profiler radar boundary layer height inversion system, characterized in that, include: The preprocessing module is used to acquire wind profiler radar product data, perform preprocessing and time matching on the wind profiler radar product data, and obtain a training dataset. The filtering module is used to filter multiple parameter profiles in the wind profiler radar product to obtain the parameter profile for inverting the boundary layer height. The parameter profile includes the signal-to-noise ratio profile, spectral width, vertical velocity standard deviation, and horizontal wind speed and direction. The training module is used to train the convolutional neural network based on the training dataset and the parameter profile of the inverted boundary layer height, and to converge the convolutional neural network using a loss function to output the wind profile radar boundary layer height inversion model. The inversion module is used to obtain the boundary layer height inversion result using the wind profiler radar boundary layer height inversion model.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the deep learning-based wind profiler radar boundary layer height inversion method as described in any one of claims 1 to 6.
9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based wind profiler radar boundary layer height inversion method as described in any one of claims 1 to 6.
10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based wind profiler radar boundary layer height inversion method as described in any one of claims 1 to 6.