A method for optimizing temperature and humidity lidar monitoring data and identifying inversion layers

Abstract

The present invention relates to inversion layer identification, and specifically to a method for optimizing and processing temperature and humidity lidar monitoring data and identifying an inversion layer. The method comprises the following steps: gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction on the grid data to reduce the influence of noise and improve the accuracy and reliability of the data; filtering the noise-reduced temperature and humidity monitoring data to smooth the data through filtering, and further improve the accuracy and reliability of the data; using a contrast selection algorithm to enhance the contrast of the filtered temperature and humidity monitoring data to further improve the readability and visualization effect of the data; and drawing a radar monitoring map with high color contrast based on the temperature and humidity monitoring data after contrast enhancement. The technical solution provided by the present invention can effectively overcome the defect of the prior art that the inversion layer cannot be accurately identified.

Description

Technical Field

[0001] The present invention relates to inversion layer identification, and in particular to a method for optimizing temperature and humidity laser radar monitoring data and identifying an inversion layer. Background Art

[0002] In the field of meteorological and environmental monitoring, increasing emphasis is being placed on observing high-altitude meteorological conditions. Obtaining vertical monitoring data can provide an important basis for studying high-altitude meteorological characteristics. A temperature inversion occurs when, under certain conditions, the atmospheric temperature gradient reverses to its normal value. Normally, temperatures should be higher closer to the ground, but a temperature inversion causes the atmosphere to cool closer to the ground and warmer higher up in the atmosphere. This inversion has a significant impact on weather, climate, and air quality.

[0003] In the field of monitoring and identifying inversion layers in China, there are devices such as sounding balloons and ground-based lidar. Sounding balloons are a tool commonly used in atmospheric meteorology research, used to measure temperature, humidity, and other meteorological parameters in the atmosphere. By acquiring vertical atmospheric monitoring data, they identify and study inversion layers. In recent years, the rapid development of ground-based lidar has led to the application of temperature and humidity lidar equipment in meteorological environmental monitoring, achieving good results. Temperature and humidity lidar features higher data accuracy, real-time monitoring, and controllable measurement distance. It is particularly effective in identifying inversion layers. By measuring parameters such as temperature and humidity in the atmosphere in the vertical direction, it creates an overall monitoring map that visually displays the inversion layer area, thereby determining information such as the high-altitude temperature, duration, and altitude range of the inversion layer.

[0004] Currently, technologies and products for identifying inversion layers lack a high level of intelligence. Furthermore, these products require extensive manual effort to analyze the overall monitoring map and identify the inversion layer. This consumes significant manpower and material resources, while the identification results are less than ideal, and the real-time nature of early warnings is also limited. Furthermore, the use of traditional mathematical processing methods for signal analysis to calculate inversion layer information only yields good results when monitoring data with ideal, typical characteristics is present. However, in actual applications, due to factors such as the on-site environment, external meteorological factors, and equipment limitations, the overall monitoring map exhibits significant variation and irregularity, making it difficult to accurately analyze the data and capture the changing trends of the map using conventional mathematical processing methods. Summary of the Invention

[0005] (1) Technical problems solved

[0006] In response to the above-mentioned shortcomings of the existing technology, the present invention provides a method for optimizing the processing of temperature and humidity lidar monitoring data and identifying the inversion layer, which can effectively overcome the defect of the existing technology that it cannot accurately identify the inversion layer.

[0007] (2) Technical solution

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0009] A method for optimizing and processing temperature and humidity lidar monitoring data, comprising the following steps:

[0010] S21. Gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction processing on the grid data to reduce the impact of noise and improve the accuracy and reliability of the data;

[0011] S22, filtering the temperature and humidity monitoring data after noise reduction to smooth the data and further improve the accuracy and reliability of the data;

[0012] S23, using a contrast selection algorithm to enhance the contrast of the filtered temperature and humidity monitoring data to further improve the readability and visualization of the data;

[0013] S24. Based on the temperature and humidity monitoring data after contrast enhancement, a radar monitoring map with high color contrast is drawn.

[0014] Preferably, in S21, gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction on the grid data, includes:

[0015] S211, using an interpolation method to draw the temperature data in the temperature and humidity monitoring data into grid data;

[0016] S212, analyze the temperature data characteristics and the location of the temperature and humidity lidar to determine the minimum temperature T min and the maximum temperature T max ;

[0017] S213, select n adjacent temperature data above, below, left and right of a certain temperature data, calculate the one-dimensional derivative of each adjacent temperature data, and compare the one-dimensional derivative with the first determined value. If there is a one-dimensional derivative greater than the first determined value, then the temperature data is regarded as a potential noise value, where the first determined value is

[0018] S214, select n adjacent temperature data above, below, left and right of the potential noise value, calculate the two-dimensional derivative of the adjacent temperature data respectively, and compare the two-dimensional derivative with the second determined value. If the two-dimensional derivative of less than n / 2 adjacent temperature data is greater than the second determined value, then the potential noise value is determined to be a noise value, where the second determined value is

[0019] S215 , selecting the minimum temperature data among the adjacent temperature data to replace the temperature data of the noise value.

[0020] Preferably, filtering the noise-reduced temperature and humidity monitoring data in S22 includes:

[0021] S221, select an n*n grid with a certain temperature data as the center point, the distance between adjacent grids is 1, and all grid points in the grid are filled with temperature data;

[0022] S222. Calculate the distances between the center point temperature data T and the remaining adjacent temperature data in the grid, then square the reciprocals of the distances and sum them. The calculation formula is as follows:

[0023]

[0024] Among them, S represents the sum of the squares of the reciprocals of the distances between the center point temperature data T and the other adjacent temperature data in the grid, d ij Represents the distance between the center point temperature data T and the adjacent temperature data in the i-th row and j-th column in the grid, i, j = 1, 2, ..., n, and i and j cannot be

[0025] S223. Take the reciprocal of the distance between the center point temperature data T and the remaining adjacent temperature data in the grid, divide it by the S value, and then multiply it by the adjacent temperature data and sum them. The calculation formula is as follows:

[0026]

[0027] Among them, T' is the filtered center point temperature data, T ij Represents the data value of the adjacent temperature data in the i-th row and j-th column in the grid, i, j = 1, 2, ..., n, and i and j cannot be

[0028] S224 , replacing the center point temperature data T with the filtered center point temperature data T′.

[0029] Preferably, in S23, contrast enhancement processing is performed on the filtered temperature and humidity monitoring data using a contrast selection algorithm, including:

[0030] S231. All temperature data in the temperature and humidity monitoring data are combined into a temperature data set. Based on the specific conditions and actual experience of temperature data collection by the radar equipment, the 5% position color scale value and the 95% position color scale value are determined using an equal contrast algorithm.

[0031] S232. Randomly select a cutoff value from the temperature data set, concentrate the temperature data less than the cutoff value to the left side of the temperature data set, and concentrate the temperature data not less than the cutoff value to the right side of the temperature data set:

[0032] When the temperature data on the left is 5% of the total data, the dividing value at this time is the 5% position color scale value;

[0033] When the temperature data on the left is less than 5% of the total data, assuming it is 2%, then the equal contrast algorithm needs to be used to calculate the 3% position color scale value from the temperature data on the right, that is, to obtain the 5% position color scale value;

[0034] When the temperature data on the left is greater than 5% of the total data, the equal contrast algorithm is needed to calculate the 5% position color scale value from the temperature data on the left;

[0035] S233, repeat S232 to determine the 95% position color scale value;

[0036] S234 , based on the 5% position color scale value and the 95% position color scale value, a color scale value is taken at every 4° C. interval to determine the color scale range.

[0037] A method for identifying an inversion layer, characterized by comprising the following steps:

[0038] S1. Obtain temperature and humidity monitoring data from the temperature and humidity lidar and draw an overall monitoring map;

[0039] S2. Preprocess and extract features of temperature and humidity monitoring data, and draw radar monitoring maps with high color contrast;

[0040] S3, constructing a training image set and a validation image set for the temperature and humidity lidar inversion layer;

[0041] S4. Constructing an inversion layer recognition model, and using a training image set and a verification image set to train and verify the inversion layer recognition model respectively;

[0042] S5. Deploy the inversion layer identification model, and calculate and display the spatial information of the inversion layer identified by the inversion layer identification model.

[0043] Preferably, in S1, the temperature and humidity monitoring data of the temperature and humidity lidar are obtained, and an overall monitoring map is drawn, including:

[0044] S11. The temperature and humidity lidar performs vertical monitoring and converts the received optical signals into electrical signals, and obtains temperature and humidity monitoring data within the preset altitude layer through the electrical signals;

[0045] S12, using a temperature and humidity lidar monitoring inversion algorithm to perform inversion processing on the temperature and humidity monitoring data to obtain inversion result data, including temperature data, relative humidity data, and temperature, humidity and pressure data;

[0046] S13. Based on the inversion result data, an overall monitoring map is drawn according to the correspondence between colors and color scales.

[0047] Preferably, in S2, the temperature and humidity monitoring data are preprocessed and feature extracted, and a radar monitoring map with high color contrast is drawn, including:

[0048] S21. Gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction processing on the grid data to reduce the impact of noise and improve the accuracy and reliability of the data;

[0049] S22, filtering the temperature and humidity monitoring data after noise reduction to smooth the data and further improve the accuracy and reliability of the data;

[0050] S23, using a contrast selection algorithm to enhance the contrast of the filtered temperature and humidity monitoring data to further improve the readability and visualization of the data;

[0051] S24. Based on the temperature and humidity monitoring data after contrast enhancement, a radar monitoring map with high color contrast is drawn.

[0052] Preferably, the training image set and the verification image set of the temperature and humidity lidar inversion layer are constructed in S3, including:

[0053] S31. Collect normal radar monitoring patterns, abnormal radar monitoring patterns, and defective radar monitoring patterns of various radar equipment under daily use conditions;

[0054] S32. Classify and label the collected radar monitoring images into fully contained inversion areas, inversion layer areas, and semi-contained inversion areas to obtain a training image set of the temperature and humidity lidar inversion layer;

[0055] S33. Collect normal radar monitoring images of various radar equipment under daily use environments as a verification image set of the temperature and humidity lidar inversion layer.

[0056] Preferably, in S4, an inversion layer recognition model is constructed, and the inversion layer recognition model is trained and verified using a training image set and a verification image set, respectively, including:

[0057] S41. Based on the structure of the U-Net model, a deep neural network model including an encoder and a decoder is constructed to obtain an inversion layer recognition model.

[0058] S42. configuring the optimizer and loss function of the inversion layer identification model before training;

[0059] S43, inputting the training image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model for training. During the training process, the inversion layer recognition model adjusts the weights and biases according to the feedback signal of the loss function to gradually optimize the performance of the model;

[0060] S44, inputting the verification image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model to obtain pixel-level inversion layer recognition results, and evaluating the performance and accuracy of the model based on the output inversion layer recognition results;

[0061] In S41, based on the structure of the U-Net model, a deep neural network model including an encoder and a decoder is constructed to obtain an inversion layer recognition model, including:

[0062] S411. Define the encoder part. The encoder consists of a series of convolutional layers and pooling layers. The convolutional layer is used to extract the feature map of the image and is defined using paddle.nn.Conv2D. The pooling layer is used to reduce the size of the feature map and is defined using paddle.nn.MaxPool2D.

[0063] S412. Define the decoder part. The decoder consists of a series of deconvolution layers and upsampling layers. The deconvolution layers and upsampling layers are used to restore the size of the feature map to the original image size. Use paddle.nn.Conv2D to define the deconvolution layer and paddle.nn.Upsample to define the upsampling layer.

[0064] S413, performing jump connections between features in the encoder and features in the decoder, combining low-level features with high-level features, and improving the model's ability to capture detailed information;

[0065] S414, setting the number of channels of the output layer to match the number of categories of the classification label;

[0066] In S42, the optimizer and loss function configuration of the inversion layer identification model before training include:

[0067] S421. Select stochastic gradient descent as the training optimizer, dynamically adjust the learning rate based on the learning rate decay strategy to improve the convergence speed and stability of the model, accelerate the convergence speed of the model by setting the momentum parameter, help the model escape the local minimum, control the complexity of the model by adjusting the weight decay, and select the most appropriate parameter combination by observing the training curve and performance of the model during the training process;

[0068] S422. Select a cross entropy loss function as the training loss function, input the training image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model, obtain the pixel-level inversion layer prediction result, and convert the pixel-level true label corresponding to the input image into a one-hot encoding form;

[0069] The cross entropy loss function calculates the difference between the inversion layer prediction result and the true label by multiplying the logarithmic probability of the inversion layer prediction result and the true label and summing them up; one-hot encoding is a method of representing a category as a binary vector with only one element being 1 and the rest being 0;

[0070] S423. Use the compilation function provided by the deep learning framework to compile the inversion layer identification model, optimizer, and loss function. When compiling the model, you need to specify the optimizer and loss function, and optionally specify evaluation indicators, including accuracy, precision, and recall.

[0071] Preferably, in S5, an inversion layer identification model is deployed, and spatial information of the inversion layer identified by the inversion layer identification model is calculated and displayed, including:

[0072] S51. Perform in-depth calculations and analysis on temperature and humidity monitoring data to determine optimal mapping parameters, including map color, temperature gradient values, and marker size, to generate high-quality, clear, and easy-to-read map files in the business system;

[0073] S52, performing a simple analysis on the inversion layer identified by the inversion layer identification model, and retaining the inversion layer identification results with a confidence level greater than a preset confidence threshold;

[0074] S53, based on the irregular inversion layer area and the height information of the monitoring center point, calculating the high-altitude temperature, duration and height range of the irregular inversion layer area, and performing contour mapping of the inversion layer area;

[0075] S54. Use spatial information technology to clearly map the outline of the inversion layer area and display detailed information about the inversion layer area, including high-altitude temperature, duration, and altitude range. Combined with visualization functions, intuitively display the spatial distribution and changing trends of the inversion layer area.

[0076] (3) Beneficial effects

[0077] Compared with the existing technology, the method for optimizing temperature and humidity lidar monitoring data and identifying inversion layers provided by the present invention has the following beneficial effects:

[0078] 1) When collecting data, the temperature and humidity lidar may experience certain fluctuations and deviations in the temperature and humidity monitoring data due to differences in meteorological conditions and temperature changes at different vertical monitoring heights. The noise reduction, filtering, and contrast selection algorithms used in the present invention can effectively optimize the temperature and humidity monitoring data. These preprocessing techniques can reduce the impact of noise and improve the smoothness, accuracy, and reliability of the data, thereby producing radar monitoring maps with high color contrast, which helps to accurately identify the inversion layer in the subsequent analysis.

[0079] 2) Under certain extreme meteorological conditions, the inversion layer area may present different shapes and contours in the overall monitoring map due to the influence of factors such as cloud cover, wind speed, and wind direction, as well as differences in the temperature and humidity lidar's own parameter settings. The deep learning method used in this invention can automatically and accurately identify the shape types of various inversion layers in real time, helping users to take appropriate measures for timely and effective treatment based on different inversion layer types and actual on-site conditions;

[0080] 3) After accurately identifying the inversion layer, information such as the high-altitude temperature, duration, and altitude range of the inversion layer area can be calculated. At the same time, by matching the outline of the inversion layer area with the spatial coordinate system, this information can be intuitively displayed in the radar monitoring map. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0082] Figure 1 This is a schematic diagram of the process of optimizing the temperature and humidity lidar monitoring data in the present invention;

[0083] Figure 2 Schematic diagram of the process of identifying the temperature inversion layer in the present invention;

[0084] Figure 3 Schematic diagram of the overall monitoring spectrum in the present invention;

[0085] Figure 4 Schematic diagram of the radar monitoring spectrum in the present invention;

[0086] Figure 5 Schematic diagram of the temperature inversion area map in the present invention;

[0087] Figure 6 Schematic diagram of the inversion layer area map in the present invention;

[0088] Figure 7 Schematic diagram of the semi-inclusive inversion region map of the present invention;

[0089] Figures 8 to 13 These are six example images from the training image set of the temperature and humidity lidar inversion layer of the present invention;

[0090] Figure 14 This is a schematic diagram of displaying spatial information of the identified inversion layer in the present invention. DETAILED DESCRIPTION

[0091] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0092] A method for optimizing and processing temperature and humidity lidar monitoring data, such as Figure 1 As shown, S21, gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction processing on the grid data to reduce the impact of noise and improve the accuracy and reliability of the data;

[0093] S22, filtering the temperature and humidity monitoring data after noise reduction to smooth the data and further improve the accuracy and reliability of the data;

[0094] S23, using a contrast selection algorithm to enhance the contrast of the filtered temperature and humidity monitoring data to further improve the readability and visualization of the data;

[0095] S24. Based on the temperature and humidity monitoring data after contrast enhancement, a radar monitoring map with high color contrast is drawn.

[0096] ① In S21, the temperature and humidity monitoring data are gridded to obtain grid data, and the grid data is subjected to noise reduction processing, including:

[0097] S211, using an interpolation method to draw the temperature data in the temperature and humidity monitoring data into grid data;

[0098] S212, analyze the temperature data characteristics and the location of the temperature and humidity lidar to determine the minimum temperature T min and the maximum temperature T max ;

[0099] S213, select n adjacent temperature data above, below, left and right of a certain temperature data, calculate the one-dimensional derivative of each adjacent temperature data, and compare the one-dimensional derivative with the first determined value. If there is a one-dimensional derivative greater than the first determined value, then the temperature data is regarded as a potential noise value, where the first determined value is

[0100] S214, select n adjacent temperature data above, below, left and right of the potential noise value, calculate the two-dimensional derivative of the adjacent temperature data respectively, and compare the two-dimensional derivative with the second determined value. If the two-dimensional derivative of less than n / 2 adjacent temperature data is greater than the second determined value, then the potential noise value is determined to be a noise value, where the second determined value is

[0101] S215 , selecting the minimum temperature data among the adjacent temperature data to replace the temperature data of the noise value.

[0102] In the technical solution of the present application, the one-dimensional derivative represents the rate of change of temperature over time. The larger the one-dimensional derivative, the more unstable the temperature. Therefore, the one-dimensional derivative of adjacent temperature data is used as the first noise point characteristic value of the temperature data; the two-dimensional derivative represents the speed of the temperature change rate over time. The larger the two-dimensional derivative, the more drastic the temperature change. Therefore, the two-dimensional derivative of adjacent temperature data is used as the second noise point characteristic value of the temperature data.

[0103] ② In S22, the temperature and humidity monitoring data after noise reduction is filtered, including:

[0104] S221, select an n*n grid with a certain temperature data as the center point, the distance between adjacent grids is 1, and all grid points in the grid are filled with temperature data;

[0105] S222. Calculate the distances between the center point temperature data T and the remaining adjacent temperature data in the grid, then square the reciprocals of the distances and sum them. The calculation formula is as follows:

[0106]

[0107] Among them, S represents the sum of the squares of the reciprocals of the distances between the center point temperature data T and the other adjacent temperature data in the grid, d ij Represents the distance between the center point temperature data T and the adjacent temperature data in the i-th row and j-th column in the grid, i, j = 1, 2, ..., n, and i and j cannot be

[0108] S223. Take the reciprocal of the distance between the center point temperature data T and the remaining adjacent temperature data in the grid, divide it by the S value, and then multiply it by the adjacent temperature data and sum them. The calculation formula is as follows:

[0109]

[0110] Among them, T' is the filtered center point temperature data, T ij Represents the data value of the adjacent temperature data in the i-th row and j-th column in the grid, i, j = 1, 2, ..., n, and i and j cannot be

[0111] S224 , replacing the center point temperature data T with the filtered center point temperature data T′.

[0112] ③ In S23, a contrast selection algorithm is used to enhance the contrast of the filtered temperature and humidity monitoring data, including:

[0113] S231. All temperature data in the temperature and humidity monitoring data are combined into a temperature data set. Based on the specific conditions and actual experience of temperature data collection by the radar equipment, the 5% position color scale value and the 95% position color scale value are determined using an equal contrast algorithm.

[0114] S232. Randomly select a cutoff value from the temperature data set, concentrate the temperature data less than the cutoff value to the left side of the temperature data set, and concentrate the temperature data not less than the cutoff value to the right side of the temperature data set:

[0115] When the temperature data on the left is 5% of the total data, the dividing value at this time is the 5% position color scale value;

[0116] When the temperature data on the left is less than 5% of the total data, assuming it is 2%, then the equal contrast algorithm needs to be used to calculate the 3% position color scale value from the temperature data on the right, that is, to obtain the 5% position color scale value;

[0117] When the temperature data on the left is greater than 5% of the total data, the equal contrast algorithm is needed to calculate the 5% position color scale value from the temperature data on the left;

[0118] S233, repeat S232 to determine the 95% position color scale value;

[0119] S234 , based on the 5% position color scale value and the 95% position color scale value, a color scale value is taken at every 4° C. interval to determine the color scale range.

[0120] A method for identifying an inversion layer, such as Figure 2 As shown, S1, obtain the temperature and humidity monitoring data of the temperature and humidity lidar, and draw an overall monitoring map;

[0121] S2. Preprocess and extract features of temperature and humidity monitoring data, and draw radar monitoring maps with high color contrast;

[0122] S3, constructing a training image set and a validation image set for the temperature and humidity lidar inversion layer;

[0123] S4. Constructing an inversion layer recognition model, and using a training image set and a verification image set to train and verify the inversion layer recognition model respectively;

[0124] S5. Deploy the inversion layer identification model, and calculate and display the spatial information of the inversion layer identified by the inversion layer identification model.

[0125] ① In S1, the temperature and humidity monitoring data of the temperature and humidity lidar is obtained and an overall monitoring map is drawn, including:

[0126] S11. The temperature and humidity lidar performs vertical monitoring and converts the received optical signals into electrical signals, and obtains temperature and humidity monitoring data within the preset altitude layer through the electrical signals;

[0127] S12, using a temperature and humidity lidar monitoring inversion algorithm to perform inversion processing on the temperature and humidity monitoring data to obtain inversion result data, including temperature data, relative humidity data, and temperature, humidity and pressure data;

[0128] S13. Based on the inversion result data, draw the overall monitoring map according to the corresponding relationship between color and color scale (such as Figure 3 shown).

[0129] ② In S2, the temperature and humidity monitoring data are preprocessed and feature extracted, and a radar monitoring map with high color contrast is drawn, including:

[0130] S21. Gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction processing on the grid data to reduce the impact of noise and improve the accuracy and reliability of the data;

[0131] S22, filtering the temperature and humidity monitoring data after noise reduction to smooth the data and further improve the accuracy and reliability of the data;

[0132] S23, using a contrast selection algorithm to enhance the contrast of the filtered temperature and humidity monitoring data to further improve the readability and visualization of the data;

[0133] S24, based on the temperature and humidity monitoring data after contrast enhancement, draw a radar monitoring map with high color contrast (such as Figure 4 shown).

[0134] ③ In S3, the training and validation image sets of the temperature and humidity lidar inversion layer are constructed, including:

[0135] S31. Collect normal radar monitoring patterns, abnormal radar monitoring patterns, and defective radar monitoring patterns of various radar equipment under daily use conditions;

[0136] S32, the collected radar monitoring maps are analyzed for the fully contained inversion area, the inversion layer area and the semi-contained inversion area (such as Figures 5 to 7 The classification labels are shown in Figure 2, and the training image set of the temperature and humidity lidar inversion layer is obtained (as shown in Figure 2). Figures 8 to 13 shown);

[0137] S33. Collect normal radar monitoring images of various radar equipment under daily use environments as a verification image set of the temperature and humidity lidar inversion layer.

[0138] In the technical solution of this application, when collecting normal radar monitoring maps, abnormal radar monitoring maps, and defect radar monitoring maps under the daily use environment of various radar equipment, temperature and humidity lidars in different time periods, different geographical locations, and different meteorological conditions can be selected to collect map files. By collecting at different time periods, radar monitoring maps in different seasons and weather conditions can be captured to cover the characteristics of inversion layers in different climates; by collecting at different geographical locations, the impact of geographical factors on radar monitoring maps can be examined, such as geographical factors such as terrain and altitude; by collecting under different meteorological conditions, the impact of meteorological factors on radar monitoring maps can be examined, such as meteorological factors such as wind speed and humidity.

[0139] Through the above acquisition strategy, a variety of radar monitoring map files can be obtained, which enables the construction of a comprehensive and representative training image set of temperature and humidity lidar inversion layers. Such a training image set will help improve the accuracy of the machine learning model to better cope with the inversion layer identification in various practical application scenarios.

[0140] ④ In S4, an inversion layer recognition model is constructed, and the training image set and the verification image set are used to train and verify the inversion layer recognition model, including:

[0141] S41. Based on the structure of the U-Net model, a deep neural network model including an encoder and a decoder is constructed to obtain an inversion layer recognition model.

[0142] S42. configuring the optimizer and loss function of the inversion layer identification model before training;

[0143] S43. Input the training image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model for training. During the training process (parameters such as the appropriate number of training rounds, batch size, and learning rate can be set), the inversion layer recognition model adjusts the weights and biases according to the feedback signal of the loss function to gradually optimize the model performance.

[0144] S44. Input the verification image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model to obtain the pixel-level inversion layer recognition result, and evaluate the performance and accuracy of the model based on the output inversion layer recognition result.

[0145] 1) In S41, based on the structure of the U-Net model, a deep neural network model including an encoder and a decoder is constructed to obtain an inversion layer recognition model, including:

[0146] S411. Define the encoder part. The encoder consists of a series of convolutional layers and pooling layers. The convolutional layer is used to extract the feature map of the image and is defined using paddle.nn.Conv2D. The pooling layer is used to reduce the size of the feature map and is defined using paddle.nn.MaxPool2D.

[0147] S412. Define the decoder part. The decoder consists of a series of deconvolution layers and upsampling layers. The deconvolution layers and upsampling layers are used to restore the size of the feature map to the original image size. Use paddle.nn.Conv2D to define the deconvolution layer and paddle.nn.Upsample to define the upsampling layer.

[0148] S413, performing jump connections between features in the encoder and features in the decoder, combining low-level features with high-level features, and improving the model's ability to capture detailed information;

[0149] S414. Set the number of channels of the output layer to match the number of categories of the classification label.

[0150] 2) In S42, the optimizer and loss function are configured for the inversion layer identification model before training, including:

[0151] S421. Select stochastic gradient descent as the training optimizer, dynamically adjust the learning rate based on the learning rate decay strategy to improve the convergence speed and stability of the model, accelerate the convergence speed of the model by setting the momentum parameter, help the model escape the local minimum, control the complexity of the model by adjusting the weight decay, and select the most appropriate parameter combination by observing the training curve and performance of the model during the training process;

[0152] S422. Select a cross entropy loss function as the training loss function, input the training image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model, obtain the pixel-level inversion layer prediction result, and convert the pixel-level true label corresponding to the input image into a one-hot encoding form;

[0153] The cross entropy loss function calculates the difference between the inversion layer prediction result and the true label by multiplying the logarithmic probability of the inversion layer prediction result and the true label and summing them up; one-hot encoding is a method of representing a category as a binary vector with only one element being 1 and the rest being 0;

[0154] S423. Use the compilation function provided by the deep learning framework to compile the inversion layer identification model, optimizer, and loss function. When compiling the model, you need to specify the optimizer and loss function, and optionally specify evaluation indicators, including accuracy, precision, and recall.

[0155] In the technical solution of this application, Baidu's open source deep learning framework, PaddlePaddle, is used, and the U-Net model is used as the basic semantic segmentation algorithm model. U-Net (U-shaped Network) is a fully convolutional neural network mainly used for image segmentation. It consists of two parts: an encoder and a decoder. The encoder part gradually reduces the spatial dimension through a series of downsampling operations, while the decoder part gradually restores the spatial dimension through a series of upsampling operations. This structure makes the U-Net model have better performance when processing image segmentation tasks.

[0156] ⑤ In S5, the inversion layer identification model is deployed, and the spatial information of the inversion layer identified by the inversion layer identification model is calculated and displayed, including:

[0157] S51. Perform in-depth calculations and analysis on temperature and humidity monitoring data to determine optimal mapping parameters, including map color, temperature gradient values, and marker size, to generate high-quality, clear, and easy-to-read map files in the business system;

[0158] S52, performing a simple analysis on the inversion layer identified by the inversion layer identification model, and retaining the inversion layer identification results with a confidence level greater than a preset confidence threshold;

[0159] S53, based on the irregular inversion layer area and the height information of the monitoring center point, calculating the high-altitude temperature, duration and height range of the irregular inversion layer area, and performing contour mapping of the inversion layer area;

[0160] S54. Use spatial information technology to clearly map the contours of the inversion layer and display detailed information about the inversion layer (e.g. Figure 14The data are shown in Figure 2), including upper air temperature, duration and altitude range, and combined with visualization function, it can intuitively show the spatial distribution and change trend of the inversion layer area.

[0161] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for optimizing and processing temperature and humidity lidar monitoring data, characterized by: The following steps are involved: S21, gridding the temperature and humidity monitoring data to obtain grid data, and performing noise reduction processing on the grid data; S22, filtering the temperature and humidity monitoring data after noise reduction to smooth the data through filtering; S23, using a contrast selection algorithm to perform contrast enhancement processing on the filtered temperature and humidity monitoring data; S24. Based on the temperature and humidity monitoring data after contrast enhancement, draw a radar monitoring map with high color contrast; In S21, the temperature and humidity monitoring data are gridded to obtain grid data, and the grid data is subjected to noise reduction processing, including: S211, using an interpolation method to draw the temperature data in the temperature and humidity monitoring data into grid data; S212, analyze the temperature data characteristics and the location of the temperature and humidity lidar to determine the minimum temperature T min and the maximum temperature T max ; S213, select n adjacent temperature data above, below, left and right of a certain temperature data, calculate the one-dimensional derivative of each adjacent temperature data, and compare the one-dimensional derivative with the first determined value. If there is a one-dimensional derivative greater than the first determined value, then the temperature data is regarded as a potential noise value, where the first determined value is S214, select n adjacent temperature data above, below, left and right of the potential noise value, calculate the two-dimensional derivative of the adjacent temperature data respectively, and compare the two-dimensional derivative with the second determined value. If the two-dimensional derivative of less than n / 2 adjacent temperature data is greater than the second determined value, then the potential noise value is determined to be a noise value, where the second determined value is S215, selecting the minimum temperature data among the adjacent temperature data to replace the temperature data of the noise value; S23 uses a contrast selection algorithm to enhance the contrast of the filtered temperature and humidity monitoring data, including: S231. All temperature data in the temperature and humidity monitoring data are combined into a temperature data set. Based on the specific conditions and actual experience of temperature data collection by the radar equipment, the 5% position color scale value and the 95% position color scale value are determined using an equal contrast algorithm. S232. Randomly select a cutoff value from the temperature data set, concentrate the temperature data less than the cutoff value to the left side of the temperature data set, and concentrate the temperature data not less than the cutoff value to the right side of the temperature data set: When the temperature data on the left is 5% of the total data, the dividing value at this time is the 5% position color scale value; When the temperature data on the left is less than 5% of the total data, assuming it is 2%, then the equal contrast algorithm needs to be used to calculate the 3% position color scale value from the temperature data on the right, that is, to obtain the 5% position color scale value; When the temperature data on the left is greater than 5% of the total data, the equal contrast algorithm is needed to calculate the 5% position color scale value from the temperature data on the left; S233, repeat S232 to determine the 95% position color scale value; S234 , based on the 5% position color scale value and the 95% position color scale value, a color scale value is taken at every 4° C. interval to determine the color scale range.

2. The method for optimizing and processing temperature and humidity lidar monitoring data according to claim 1, wherein: In S22, the temperature and humidity monitoring data after noise reduction is filtered, including: S221, select an n*n grid with a certain temperature data as the center point, the distance between adjacent grids is 1, and all grid points in the grid are filled with temperature data; S222. Calculate the distances between the center point temperature data T and the remaining adjacent temperature data in the grid, then square the reciprocals of the distances and sum them. The calculation formula is as follows: Among them, S represents the sum of the squares of the reciprocals of the distances between the center point temperature data T and the other adjacent temperature data in the grid, d ij Represents the distance between the center point temperature data T and the adjacent temperature data in the i-th row and j-th column in the grid, i, j = 1, 2, ..., n, and i and j cannot be S223. Take the reciprocal of the distance between the center point temperature data T and the remaining adjacent temperature data in the grid, divide it by the S value, and then multiply it by the adjacent temperature data and sum them. The calculation formula is as follows: Among them, T' is the filtered center point temperature data, T ij Represents the data value of the adjacent temperature data in the i-th row and j-th column in the grid, i, j = 1, 2, ..., n, and i and j cannot be S224 , replacing the center point temperature data T with the filtered center point temperature data T′.

3. A method for identifying an inversion layer, using the temperature and humidity lidar monitoring data optimization processing method according to claim 1, characterized in that: The following steps are involved: S1. Obtain temperature and humidity monitoring data from the temperature and humidity lidar and draw an overall monitoring map; S2. Use the temperature and humidity lidar monitoring data optimization processing method to preprocess and extract features of the temperature and humidity monitoring data, and draw a radar monitoring map with high color contrast; S3, constructing a training image set and a validation image set for the temperature and humidity lidar inversion layer; S4. Constructing an inversion layer recognition model, and using a training image set and a verification image set to train and verify the inversion layer recognition model respectively; S5. Deploy the inversion layer identification model, and calculate and display the spatial information of the inversion layer identified by the inversion layer identification model.

4. The method for identifying an inversion layer according to claim 3, wherein: S1 obtains temperature and humidity monitoring data from the temperature and humidity lidar and draws an overall monitoring map, including: S11. The temperature and humidity lidar performs vertical monitoring and converts the received optical signals into electrical signals, and obtains temperature and humidity monitoring data within the preset altitude layer through the electrical signals; S12, using a temperature and humidity lidar monitoring inversion algorithm to perform inversion processing on the temperature and humidity monitoring data to obtain inversion result data, including temperature data, relative humidity data, and temperature, humidity and pressure data; S13. Based on the inversion result data, an overall monitoring map is drawn according to the correspondence between colors and color scales.

5. The method for identifying an inversion layer according to claim 3, wherein: The training and validation image sets for the temperature and humidity lidar inversion layer are constructed in S3, including: S31. Collect normal radar monitoring patterns, abnormal radar monitoring patterns, and defective radar monitoring patterns of various radar equipment under daily use conditions; S32. Classify and label the collected radar monitoring images into fully contained inversion areas, inversion layer areas, and semi-contained inversion areas to obtain a training image set of the temperature and humidity lidar inversion layer; S33. Collect normal radar monitoring images of various radar equipment under daily use environments as a verification image set of the temperature and humidity lidar inversion layer.

6. The method for identifying an inversion layer according to claim 3, wherein: In S4, an inversion layer recognition model is constructed, and the training image set and verification image set are used to train and verify the inversion layer recognition model, including: S41. Based on the structure of the U-Net model, a deep neural network model including an encoder and a decoder is constructed to obtain an inversion layer recognition model. S42. configuring the optimizer and loss function of the inversion layer identification model before training; S43, inputting the training image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model for training. During the training process, the inversion layer recognition model adjusts the weights and biases according to the feedback signal of the loss function to gradually optimize the performance of the model; S44, inputting the verification image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model to obtain pixel-level inversion layer recognition results, and evaluating the performance and accuracy of the model based on the output inversion layer recognition results; In S41, a deep neural network model including an encoder and a decoder is constructed based on the structure of the U-Net model to obtain the inversion layer recognition model, including: S411. Define the encoder part. The encoder consists of a series of convolutional layers and pooling layers. The convolutional layer is used to extract the feature map of the image and is defined using paddle.nn.Conv2D. The pooling layer is used to reduce the size of the feature map and is defined using paddle.nn.MaxPool2D. S412. Define the decoder part. The decoder consists of a series of deconvolution layers and upsampling layers. The deconvolution layers and upsampling layers are used to restore the size of the feature map to the original image size. Use paddle.nn.Conv2D to define the deconvolution layer and paddle.nn.Upsample to define the upsampling layer. S413, performing jump connections between features in the encoder and features in the decoder, combining low-level features with high-level features, and improving the model's ability to capture detailed information; S414, setting the number of channels of the output layer to match the number of categories of the classification label; In S42, the optimizer and loss function configuration of the inversion layer identification model before training include: S421. Select stochastic gradient descent as the training optimizer, dynamically adjust the learning rate based on the learning rate decay strategy to improve the convergence speed and stability of the model, accelerate the convergence speed of the model by setting the momentum parameter, help the model escape the local minimum, control the complexity of the model by adjusting the weight decay, and select the most appropriate parameter combination by observing the training curve and performance of the model during the training process; S422. Select a cross entropy loss function as the training loss function, input the training image set of the temperature and humidity lidar inversion layer into the inversion layer recognition model, obtain the pixel-level inversion layer prediction result, and convert the pixel-level true label corresponding to the input image into a one-hot encoding form; The cross entropy loss function calculates the difference between the inversion layer prediction result and the true label by multiplying the logarithmic probability of the inversion layer prediction result and the true label and summing them up; one-hot encoding is a method of representing a category as a binary vector with only one element being 1 and the rest being 0; S423. Use the compilation function provided by the deep learning framework to compile the inversion layer identification model, optimizer, and loss function. When compiling the model, you need to specify the optimizer and loss function, and optionally specify evaluation indicators, including accuracy, precision, and recall.

7. The method for identifying an inversion layer according to claim 3, wherein: In S5, the inversion layer identification model is deployed, and the spatial information of the inversion layer identified by the inversion layer identification model is calculated and displayed, including: S51. Perform in-depth calculations and analysis on temperature and humidity monitoring data to determine optimal mapping parameters, including map color, temperature gradient values, and marker size, to generate high-quality, clear, and easy-to-read map files in the business system; S52, performing a simple analysis on the inversion layer identified by the inversion layer identification model, and retaining the inversion layer identification results with a confidence level greater than a preset confidence threshold; S53, based on the irregular inversion layer area and the height information of the monitoring center point, calculating the high-altitude temperature, duration and height range of the irregular inversion layer area, and performing contour mapping of the inversion layer area; S54. Use spatial information technology to clearly map the outline of the inversion layer area and display detailed information about the inversion layer area, including high-altitude temperature, duration, and altitude range. Combined with visualization functions, intuitively display the spatial distribution and changing trends of the inversion layer area.

Images