EDH-SimUNet-based evaporation waveguide diagnosis method

Abstract

The invention discloses an evaporation waveguide diagnosis method based on EDH-SimUNet, and the method comprises the steps: taking collected data in a region as the input of an evaporation waveguide model, and predicting the height of an evaporation waveguide on the sea surface in the region. The structure of the evaporation waveguide model is EDH-SimUNet, the depth direction separable convolution operation is used for down-sampling, transposition convolution is used for up-sampling, and jump connection is used for relieving the problems of gradient disappearance and information loss. The EDH-SimUNet can avoid diagnosis results with large differences caused by determination of a universal function, a stability correction function and scale parameters of a physical model under a non-uniform atmospheric condition and a strong stable atmospheric condition. The diagnosis method effectively solves the problem of high time cost caused by iterative computation when a traditional physical evaporation waveguide model diagnoses the waveguide height, can adapt to constantly changing data and environments, automatically adjusts parameters to optimize performance, and then is easily expanded to a new data set and task.

Description

Technical Field

[0001] The present invention relates to the cross-integration application field of deep learning (DL) and evaporation waveguide modeling, and in particular to a diagnosis method for evaporation waveguide height based on deep learning. Background Art

[0002] Evaporation duct is an atmospheric stratification formed by the sharp decrease of atmospheric humidity with height due to the evaporation of water vapor. This stratification has an abnormal atmospheric refractive index structure. The evaporation duct model can be applied to the study and prediction of various evaporation-induced duct phenomena, such as lake evaporation, reservoir evaporation, soil evaporation, etc.

[0003] At present, based on the Monin-Obukhov similarity theory, scholars have proposed a series of physical models for diagnosing the height of the evaporation duct. Among them, the commonly used evaporation duct models include the PJ model, the Musson-Gauthier-Bruth (MGB) model, the Babin model, the Navy Postgraduate School (NPS) model, the pseudo-refraction model, etc.

[0004] These physical models based on similarity theory require the calculation of the characteristic length L and characteristic parameter θ in the atmospheric boundary layer. * and q * . On the one hand, these parameters use some approximations in the calculation process, which leads to certain uncertainties; on the other hand, the determination of these parameters often requires the use of the CPU for very time-consuming iterative calculations. For example, according to actual tests, the calculation of the evaporation duct height of the South China Sea near the sea surface for one year using the 24 cores of the Intel Xeon Gold 5218 requires continuous calculations for several hours. In the past decade, due to the demand for matrix computing power in fields such as computer graphics and deep learning, the growth rate of GPU computing power has been much faster than that of CPU. For example, the single-precision floating-point performance of the most powerful civilian-grade GPU, NVIDIA RTX 4090, is about 83TFLOPs, while the most powerful civilian-grade CPU, AMD 7995WX, is about 12TFLOPs.

[0005] In order to calculate the evaporation waveguide height of a large sea area more quickly, one of the reasonable research directions is how to make good use of the powerful computing power provided by GPU. Thanks to the computing power support provided by GPU and the massive data brought by the development of sensor technology, the data-driven modeling using deep neural networks has developed rapidly.

[0006] Specifically for the evaporation duct near the sea surface, due to the development of satellite assimilation technology, various databases have accumulated a large amount of weather data, such as the ERA5 data set provided by the European Center for Medium-Range Weather Forecasts (ECMWF) or the reanalysis data provided by the National Center for Atmospheric Research (NCAR), which makes it possible to complete the evaporation duct height diagnosis using deep neural networks. When using this technology, the completion of evaporation duct diagnosis with high temporal and spatial resolution can be shortened to a few seconds.

[0007] In the prior art, the prediction studies on the evaporation duct height mainly include:

[0008] 1. A Chinese patent application with publication number CN112711899A proposes “a fusion prediction method for evaporation duct height”, which includes the following steps: (1) obtaining the true value of evaporation duct height; (2) predicting the evaporation duct model height; (3) model sensitivity analysis; (4) least squares support vector machine regression modeling: according to the sensitivity analysis results, in different environmental ranges, the obtained meteorological and hydrological observation data and the evaporation duct model height prediction value are used as input, and the evaporation duct height true value is used as output, and the nonlinear mapping relationship between the input and output is obtained by regression, and a fusion prediction model of multiple evaporation duct model predicted heights is established; (5) evaporation duct height prediction.

[0009] The main shortcomings of this technology are: the support vector machine algorithm used in this method has shortcomings, such as: (1) Support vector machines are not suitable for processing large-scale data. When there are many data features, the training time will increase significantly. (2) When faced with small data sets, support vector machines are sensitive to noise, which makes the model prone to overfitting.

[0010] 2. The Chinese patent application with publication number CN116559977A proposes “a method for predicting the height of NPS-GRYANIK20 evaporation waveguide based on BP neural network”, which adopts a method of dividing environmental intervals based on the NPS-GRYANIK20 evaporation waveguide model, constructs a BP neural network prediction model based on the predator algorithm in different environmental intervals, and realizes accurate prediction of the height of the evaporation waveguide, specifically including: obtaining the true value of the evaporation waveguide height, dividing the environmental interval based on the NPS-GRYANIK20 evaporation waveguide model, and predicting the height of the evaporation waveguide based on the improved BP neural network prediction model.

[0011] The main drawbacks of this technique are that the predator algorithm optimization process used in this method is divided into multiple stages, involving complex computational steps. This may lead to high computational costs, especially when the neural network is trained on a large-scale dataset. In addition, the predator algorithm is not applicable to all types of problems, and this paper lacks the necessary experiments to verify the applicability of the predator algorithm.

[0012] 3. The Chinese patent application with publication number CN116579225A proposes "a regional prediction method for evaporation duct height based on convolutional neural network". This method combines the numerical forecast product grid data to construct a regional prediction model for evaporation duct height of numerical forecast products based on convolutional neural network (NWPP-EDH).

[0013] The main drawback of this technology is that this method uses 2×2 classic convolution layers in all modules of the network. Smaller convolution kernels have smaller receptive fields, which will lead to poorer model fitting ability. In the experimental part, the RMSE index of NWPP-EDH is greater than 1, which proves this point. In addition, the use of general convolution methods will make the network larger, which not only increases the risk of overfitting, but also reduces the speed of model output results and puts higher requirements on the hardware of the running device. Summary of the invention

[0014] In order to solve the high time cost caused by the iterative calculation required in the current evaporation waveguide model in diagnosing the waveguide height, the present invention provides a diagnosis solution based on a UNet-type deep neural network, which is hereinafter referred to as EDH-SimUNet.

[0015] EDH-SimUNet is a UNet-type network that includes a downsampling part, an upsampling part, and a jump connection part to mitigate the vanishing gradient. EDH-SimUNet uses depth-wise separable convolution operations for downsampling and transposed convolutions for upsampling. According to prior knowledge, weather conditions at long distances are not very helpful for diagnosing the height of the evaporation duct, so there is no need to use a costly attention mechanism.

[0016] According to experimental tests, EDH-SimUNet is an order of magnitude faster than traditional physical models in diagnosis speed. In addition, compared with traditional machine learning methods such as XGBoost and random forest, EDH-SimUNet also has an advantage in diagnosis speed and performs better in terms of error.

[0017] The large-area evaporation waveguide diagnostic method based on EDH-SimUNet of the present invention comprises the following steps:

[0018] Step 1: Filter the temperature, pressure, humidity, wind speed, and sea surface temperature data near the sea surface in the region from the analysis data set as the features of the training data;

[0019] Step 2: Use the evaporation duct model to calculate the evaporation duct height corresponding to the air temperature, air pressure, humidity, wind speed, and sea surface temperature data, that is, to create labels for the training data;

[0020] Step 3: Build the EDH-SimUNet neural network and divide the data set into a training set and a test set. The training set is further randomly divided into a training set and a validation set. Adjust the parameters based on the performance of the validation set;

[0021] Step 4: With the help of the test set, calculate indicators such as RMSE, MAE, MedAE and RS to evaluate and verify the model performance.

[0022] This method EDH-SimUNet uses depth-wise separable convolution operations for downsampling, transposed convolutions for upsampling, and skip connections to alleviate the problems of gradient vanishing and information loss.

[0023] This method effectively solves the high time cost problem caused by iterative calculations in the traditional physical evaporation duct model when diagnosing the duct height. In addition, this method can adapt to changing data and environment, automatically adjust parameters to optimize performance, and easily expand to new data sets and tasks. Therefore, EDH-SimUNet has the potential to avoid the physical model under non-uniform atmospheric conditions and strongly stable atmospheric conditions, which leads to different diagnostic results due to the determination of universal functions, stability correction functions and scale parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Schematic diagram of the large-area evaporation waveguide diagnosis process based on EDH-SimUNet;

[0025] Figure 2 Schematic diagram of the comparison between EDH-SimUNet waveguide diagnosis results and true values ​​(the first row is the true value, the second row is the EDH-SimUNet diagnosis results);

[0026] Figure 3 It is the top-level structure of EDH-SimUNet;

[0027] Figure 4 It is the underlying structure of the input and output processing module in EDH-SimUNet;

[0028] Figure 5 It is the underlying structure of the downsampling module in EDH-SimUNet;

[0029] Figure 6 It is the underlying structure of the upsampling module in EDH-SimUNet; DETAILED DESCRIPTION

[0030] The evaporation duct diagnosis method based on EDH-SimUNet of the present invention uses the collected data of air temperature, air pressure, humidity, wind speed and sea surface temperature in the region as the input of the evaporation duct model to predict the evaporation duct height on the sea surface in the region.

[0031] The structure of the evaporation waveguide model is an EDH-SimUNet neural network; the steps of constructing the EDH-SimUNet neural network model include:

[0032] Step 1: Filter the air temperature, air pressure, humidity, wind speed, and sea surface temperature data near the sea surface in the region from the reanalysis data set as the features of the training data;

[0033] Step 2: Use the evaporation duct model to calculate the evaporation duct height corresponding to the air temperature, air pressure, humidity, wind speed, and sea surface temperature data, that is, to create labels for the training data;

[0034] Step 3: Build the EDH-SimUNet neural network; divide the data set into a training set and a test set; the training set is further randomly divided into a training set and a validation set; adjust the parameters of the EDH-SimUNet neural network according to the performance of the validation set;

[0035] Step 4: Evaluate and verify the performance of the evaporation waveguide model with the help of the test set;

[0036] In the step 3, the framework of the EDH-SimUNet neural network is, in order of data processing, an input processing module, a downsampling module A, a downsampling module B, a downsampling module C, a downsampling module D, an upsampling module D, an upsampling module C, an upsampling module B, an upsampling module A, and an output processing module; each module is jump-connected, that is:

[0037] The output of the input processing module is used as the input of the downsampling module A and the upsampling module A;

[0038] The output of downsampling module A is used as the input of downsampling module B and upsampling module B;

[0039] The output of downsampling module B is used as the input of downsampling module C and upsampling module C;

[0040] The output of the downsampling module C is used as the input of the downsampling module D and the upsampling module D;

[0041] The output of the downsampling module D is used as the input of the upsampling module D;

[0042] The output of upsampling module A is used as the input of output processing module;

[0043] The input to the input processing module is the collected data of air temperature, air pressure, humidity, wind speed and sea surface temperature;

[0044] The output of the output processing module is the prediction result;

[0045] The input processing module maps the input features to the shape required by the downsampling module;

[0046] The output processing module contains only one 1×1 convolution layer, so that the final output of the evaporation waveguide model has a suitable number of channels;

[0047] In the four downsampling modules, each time the data passes through a downsampling module, the number of channels, i.e., the feature dimension, doubles, and the length and width are halved;

[0048] In the four upsampling modules, each time the data passes through an upsampling module, the number of channels, i.e., the feature dimension, is halved, and the length and width are doubled.

[0049] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0050] The process of the large-area evaporation waveguide diagnostic method based on EDH-SimUNet in this example is as follows Figure 1 shown.

[0051] First, a training dataset needs to be prepared, which needs to include near-sea surface air temperature, air pressure, humidity, wind speed, and sea temperature data. Here, this example uses the reanalysis data EAR5 provided by the European Center for Medium-Range Weather Forecasts as an example for explanation and experiment. Using the 10m wind speed, 2m dew point temperature, 2m temperature, average sea surface pressure, and sea surface temperature data in the EAR5 meteorological data, the time resolution is 1 hour and the spatial resolution is 0.25 degrees (about 30km). Since the evaporation duct physical model requires specific humidity, the 2m specific humidity data needs to be calculated based on the average sea surface pressure and 2m dew point temperature. Next, based on the known average sea surface pressure, 10m wind speed, 2m temperature, sea surface temperature, and 2m specific humidity, the evaporation duct physical model is used to calculate the corresponding duct height. In this way, the features of the dataset and the corresponding labels are ready. In principle, the dataset should be as large as possible in order to enable EDH-SimUNet to have excellent generalization ability so that it can make more accurate diagnoses. In the experiment, we prepared a total of 31 years of meteorological data, with the specific time span from 1993 to 2023, and the spatial range from E107° to E122.75° east-west, and N4.25° to N20° north-south.

[0052] Before starting training, the data is preprocessed.

[0053] First, it is necessary to remove abnormal values ​​in the data set. Even if the reanalysis data used in the experiment is small, it is difficult to avoid a small number of NaN values. Secondly, for the efficiency of model training, the data needs to be standardized, that is, the data at each grid point is processed according to the following formula.

[0054]

[0055] Among them, x, μ, and σ are features and their corresponding means and variances, respectively. It should be noted that it is necessary to save μ and σ. When EDH-SimUNet makes predictions, the meteorological elements at the time of prediction should be processed according to the μ and σ of the training samples, rather than according to the mean and variance of the meteorological elements themselves at the time of testing.

[0056] After preparing the data set, you can start building the network structure.

[0057] like Figure 3 As shown in the figure, the overall framework of EDH-SimUNet mainly includes input and output processing modules, downsampling modules and upsampling modules.

[0058] The design of the input processing module is as follows Figure 4 As shown in Figure 1, its main function is to map the input features to the shape required by the downsampling module. The output processing module only contains a 1×1 convolution layer, so that the final output of the model has a suitable number of channels.

[0059] According to experimental tests, the number and structure of downsampling and upsampling modules are as follows Figure 5 and Figure 6 The design will enable EDH-SimUNet to have faster reasoning speed and more accurate diagnosis results.

[0060] The downsampling part contains four downsampling modules, in which each time the data passes through a downsampling module, the number of channels (feature dimension) doubles, and the length and width are halved.

[0061] The upsampling part is the reverse process of downsampling and includes four upsampling modules. Each time the data passes through an upsampling module, the number of channels (feature dimension) is halved and the length and width are doubled.

[0062] Figure 5 The specific design of each downsampling module is shown. In general, the downsampling module needs to include a maximum pooling layer to reduce the resolution of the data, a depth-wise separable convolutional layer to extract features, a batch normalization layer to speed up the model convergence, and a ReLU layer for nonlinear mapping. There is no special arrangement for the data flow in the downsampling part, that is, the data flows through each layer in order from top to bottom. Figure 3 As shown, starting from the input processing module and ending at the down-sampling module C, after each module processes the data, a copy of the output result needs to be temporarily stored so as to be subsequently input into the corresponding up-sampling module.

[0063] The structure of the upsampling module is as follows: Figure 6As shown in the figure, the design of each module is very similar to that of the downsampling part. The difference is that the upsampling part replaces the maximum pooling layer with a transposed convolution layer so that the width and height of the data are gradually restored. The convolution layer of the upsampling part gradually restores the feature dimension of the data. Since the upsampling module has two inputs, the data flow at this time needs special arrangements. Taking the upsampling module D as an example, the output from the downsampling module D is processed by the transposed convolution layer, the number of channels is halved and the width and height are doubled. Then, the input of the transposed convolution layer is spliced ​​with the result temporarily stored in the downsampling module C in the channel dimension and input into the first convolution layer. There is no special arrangement for the subsequent data flow, which can flow through each layer from top to bottom.

[0064] In order to prove the feasibility of EDH-SimUNet in waveguide diagnosis, the present invention trained a SOTA (State of the Art) model according to the above process for evaluation. The training set used in training is the reanalysis data from 1993 to 2022 provided by ERA5. Specifically, during training, the training set is randomly divided into a validation set and a training set at a ratio of 20% as a validation set. In order to measure the generalization ability of the model, the reanalysis data of 2023 was used as a test set. In addition, traditional machine learning algorithms, such as Random Forest (RF) and XGBoost, were selected as control groups for evaluation. The indicators involved in the evaluation are Root Mean Square Error (RMSE), Mean Absolute Error (Meanabsolute Error), Median Absolute Error (MedAE) and Coefficient of Determination (RS). The test platform is: AMD Ryzen 5900X CPU, RTX3080Laptop GPU, 32G DDR4 3200MHz dual-channel memory, Windows 11 23H2 operating system. Table 1 summarizes the performance of the four models on the test set. It is not difficult to find from the table that EDH-SimUNet is significantly better than the traditional machine learning model in all indicators. EDH-SimUNet has extremely small RMSE and RS, which means that EDH-SimUNet has a high accuracy rate in the task of diagnosing waveguide height and can be fully put into actual production life. In addition, the time-consuming column in the table represents the time consumed by each model to diagnose all the data in the test set. Even though EDH-SimUNet is a deep neural network, its inference speed is not inferior to traditional machine learning methods in actual performance, and it is even faster than these models due to the optimization of software and hardware. Compared with the classic UNet using general convolution, EDH-SimUNet has fewer parameters, faster calculation speed, and more accurate diagnosis results. In order to intuitively understand the diagnostic effect of EDH-SimUNet, the present invention draws Figure 2 . Figure 2 The first row shows the waveguide distribution diagram calculated using the evaporation waveguide model, and the second row shows the waveguide distribution diagram diagnosed by EDH-SimUNet. Comparing the two rows, it can be found that the waveguide distribution obtained by EDH-SimUNet is almost exactly the same as the calculation result of the evaporation waveguide model.

[0065]

[0066] Table 1 Model diagnostic performance evaluation

[0067] (↓ means the smaller the better, ↑ means the bigger the better)

Claims

1. An evaporation duct diagnostic method based on EDH-SimUNet, which uses the collected data of air temperature, air pressure, humidity, wind speed and sea surface temperature in the region as the input of the evaporation duct model to predict the evaporation duct height on the sea surface in the region; it is characterized by The structure of the evaporation waveguide model is an EDH-SimUNet neural network; the steps of constructing the EDH-SimUNet neural network model include: Step 1: Filter the air temperature, air pressure, humidity, wind speed, and sea surface temperature data near the sea surface in the region from the reanalysis data set as the features of the training data; Step 2: Use the evaporation duct model to calculate the evaporation duct height corresponding to the air temperature, air pressure, humidity, wind speed, and sea surface temperature data, that is, to create labels for the training data; Step 3: Build the EDH-SimUNet neural network; divide the data set into a training set and a test set; the training set is further randomly divided into a training set and a validation set; adjust the parameters of the EDH-SimUNet neural network according to the performance of the validation set; Step 4: Evaluate and verify the performance of the evaporation waveguide model with the help of the test set; In the step 3, the framework of the EDH-SimUNet neural network is, in order of data processing, an input processing module, a downsampling module A, a downsampling module B, a downsampling module C, a downsampling module D, an upsampling module D, an upsampling module C, an upsampling module B, an upsampling module A, and an output processing module; each module is jump-connected, that is: The output of the input processing module is used as the input of the downsampling module A and the upsampling module A; The output of downsampling module A is used as the input of downsampling module B and upsampling module B; The output of downsampling module B is used as the input of downsampling module C and upsampling module C; The output of the downsampling module C is used as the input of the downsampling module D and the upsampling module D; The output of the downsampling module D is used as the input of the upsampling module D; The output of upsampling module A is used as the input of output processing module; The input to the input processing module is the collected data of air temperature, air pressure, humidity, wind speed and sea surface temperature; The output of the output processing module is the prediction result; The input processing module maps the input features to the shape required by the downsampling module; The output processing module contains only one 1×1 convolution layer, so that the final output of the evaporation waveguide model has a suitable number of channels; In the four downsampling modules, each time the data passes through a downsampling module, the number of channels, i.e., the feature dimension, doubles, and the length and width are halved; In the four upsampling modules, each time the data passes through an upsampling module, the number of channels, i.e., the feature dimension, is halved, and the length and width are doubled.

2. The evaporation waveguide diagnostic method based on EDH-SimUNet according to claim 1 is characterized in that In the step 3, before the training set is used for evaporation waveguide model training, the training set is first standardized and the mean and variance are saved; The test set was normalized by mean and variance before being used for testing.

3. The evaporation waveguide diagnosis method based on EDH-SimUNet according to claim 2 is characterized in that the training data is standardized, that is, the data at each grid point is processed according to the following formula: in, x, μ, and σ are the features and their corresponding means and variances, respectively.

4. The evaporation duct diagnostic method based on EDH-SimUNet according to claim 1 is characterized in that in step 4, the performance of the evaporation duct model is evaluated and verified by calculating the root mean square error RMSE, the mean absolute error MAE, the median absolute error MedAE and / or the relative strength index RS.

5. The large-area evaporation waveguide diagnostic method based on EDH-SimUNet according to claim 1 is characterized by: The order of data processing in the input processing module is: a 3×3 depth-wise separable convolutional layer with 6 input channels, 64 output channels and padding 1; a batch normalization layer; a ReLU layer; a 3×3 depth-wise separable convolutional layer with 64 input channels, 64 output channels and padding 1; a batch normalization layer; a ReLU layer; The order of data processing in the output module is as follows: a 1×1 convolutional layer with 64 input channels and 1 output channel; The data processing order in the downsampling module A is as follows: 2×2 maximum pooling layer; 3×3 depth-wise separable convolution layer with 64 input channels, 128 output channels and padding 1; batch normalization layer; ReLU layer; 3×3 depth-wise separable convolution layer with 128 input channels, 128 output channels and padding 1; batch normalization layer; ReLU layer; The data processing order in the downsampling module B is as follows: 2×2 maximum pooling layer; 3×3 depth-wise separable convolution layer with 128 input channels, 256 output channels and padding 1; batch normalization layer; ReLU layer; 3×3 depth-wise separable convolution layer with 25 input channels, 256 output channels and padding 1; batch normalization layer; ReLU layer; The data processing order in the downsampling module C is as follows: 2×2 maximum pooling layer; input channel is 256; 3×3 depth-wise separable convolution layer with 512 output channels and padding 1; batch normalization layer; ReLU layer; input channel is 512; 3×3 depth-wise separable convolution layer with 512 output channels and padding 1; batch normalization layer; ReLU layer; The data processing order in the downsampling module D is as follows: 2×2 maximum pooling layer; 3×3 depth-wise separable convolution layer with 512 input channels, 1024 output channels and padding 1; batch normalization layer; ReLU layer; 3×3 depth-wise separable convolution layer with 1024 input channels, 1024 output channels and padding 1; batch normalization layer; ReLU layer; The data processing order in the upsampling module A is as follows: 2×2 transposed convolution layer with 128 input channels, 64 output channels, and stride 2; 3×3 depth-wise separable convolution layer with 128 input channels, 64 output channels, and padding 1; batch normalization layer; ReLU layer; 3×3 depth-wise separable convolution layer with 64 input channels, 64 output channels, and padding 1; batch normalization layer; ReLU layer; The data processing order in the upsampling module B is as follows: 2×2 transposed convolution layer with 256 input channels, 128 output channels, and stride 2; 3×3 depth-wise separable convolution layer with 256 input channels, 128 output channels, and padding 1; batch normalization layer; ReLU layer; 3×3 depth-wise separable convolution layer with 128 input channels, 128 output channels, and padding 1; batch normalization layer; ReLU layer; The data processing order in the upsampling module C is as follows: 2×2 transposed convolution layer with an input channel of 512, an output channel of 256, and a stride of 2; 3×3 depth-wise separable convolution layer with an input channel of 512, an output channel of 256, and padding of 1; batch normalization layer; ReLU layer; 3×3 depth-wise separable convolution layer with an input channel of 256, an output channel of 256, and padding of 1; batch normalization layer; ReLU layer; The data processing order in the upsampling module D is as follows: 2×2 transposed convolution layer with an input channel of 1024, an output channel of 512, and a stride of 2; a 3×3 depth-wise separable convolution layer with an input channel of 1024, an output channel of 512, and padding of 1; a batch normalization layer; a ReLU layer; a 3×3 depth-wise separable convolution layer with an input channel of 512, an output channel of 512, and padding of 1; a batch normalization layer; and a ReLU layer.

Images