Method for rapidly measuring melting rate of accumulated snow
By meshing and data preprocessing of snow-covered target areas, combining distributed energy balance model and multi-layer perceptrons, ablation rate prediction model is established, which solves the complexity of snow ablation rate measurement in high-altitude mountainous areas, and achieves rapid and accurate snow ablation rate prediction.
Patent Information
- Application Number
- CN202410835199.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-26
- Publication Date
- 2024-11-29
AI Technical Summary
In high-altitude mountainous areas, the terrain in the mountainous areas has a large fluctuation, and there is great uncertainty in the range and state changes of snow accumulation during the melting period. It is difficult to accurately identify the distribution range and ablation state based on optical satellite remote sensing data, and the snow accumulation melts. Rate is affected by a variety of environmental factors, resulting in complex accurate measurement and prediction.
By meshing the snow-covered target area, we can obtain meteorological data, surface data and snow-covered data for preprocessing, and combine the distributed energy balance model and improved multi-layer perceptron for training, establish an ablation rate prediction model and calculate snow-covered data. Ablation rate.
The rapid and accurate determination of the snow ablation rate is achieved. The distributed energy balance model describes the energy exchange process in detail. The multi-layer perceptron learns complex nonlinear relationships through machine learning, improving the flexibility and adaptability of prediction.
Smart Images

Figure CN119046744A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of measurement technology, and in particular relates to a method for quickly measuring snow melting rate. Background Art
[0002] The snow cover decay process in alpine mountainous areas is complex, and accurate identification of snow cover information is a prerequisite for simulating the decay process. The snow cover information extraction method based on visible light remote sensing data mainly uses the snow cover index method (NDSI), which extracts the snow cover area by setting a threshold. In addition to snow cover information identification, snow cover state parameters reflect the characteristics of the snow itself, and snow cover parameters have corresponding sensitive bands for the spectrum. Currently, commonly used snow cover state parameters include snow density, snow particle size index, snow depth, snow water equivalent, etc. Based on the spectral characteristics of seasonal snow, Negi studied the changes in the normalized snow cover index (NDSI), indicating that with the increase in the degree of snow aging and the increase in snow particle size and snow layer water content, the NDSI value has a tendency to gradually increase; Hao Xiaohua et al. divided the snow into five layers based on the stratified characteristics of the snow, measured the snow density, snow particle size and snow layer water content of each layer, evaluated the ART optimization inversion band and snow shape factor optimization method, and determined the best band for snow particle size inversion and the best snow shape factor value. However, mountainous terrain is undulating, and the range and state of snow during the snowmelt period are highly uncertain. It is difficult to accurately identify the distribution range and melting state of snow based on optical satellite remote sensing data. In addition, the melting rate of snow is affected by a variety of environmental factors, including temperature, humidity, wind speed, solar radiation intensity, etc. These factors can vary greatly in different locations and times, making it complicated to accurately measure and predict snow melting. Summary of the invention
[0003] In order to solve the above technical problems, the present invention proposes a method for quickly measuring the snow melting rate to solve the problems existing in the above prior art.
[0004] To achieve the above object, the present invention provides a method for quickly measuring snow melting rate, comprising:
[0005] Divide the target snow accumulation area into grids, obtain meteorological data, surface data and snow accumulation data in each grid area, and pre-process the meteorological data, surface data and snow accumulation data to obtain a pre-processed data set;
[0006] The energy budget balance data of the snow surface in each grid area is obtained through the distributed energy balance model;
[0007] Constructing an improved multilayer perceptron, and training the improved multilayer perceptron using the preprocessed data set and the energy balance data to obtain an ablation rate prediction model;
[0008] The snow ablation rate is calculated based on the ablation rate prediction model to obtain a rate calculation result.
[0009] Preferably, the process of gridding the snow target area includes:
[0010] Acquire digital elevation model data of a target snow accumulation area, perform smoothing processing and missing value filling on the digital elevation model data, and obtain processed model data;
[0011] The processing model data is gridded based on an adaptive grid method, wherein the grid is an orthogonal grid.
[0012] Preferably, the process of obtaining meteorological data, surface data and snow data in each grid area includes:
[0013] Obtain meteorological parameter data for each grid area based on real-time data from weather stations;
[0014] Obtain terrain information within each grid area, and simultaneously obtain soil type and vegetation coverage data within each grid area;
[0015] Snow cover data is obtained, and a time series of the snow cover data is obtained.
[0016] Preferably, the meteorological data, surface data and snow data are preprocessed to obtain a preprocessed data set, comprising:
[0017] Checking the meteorological data, surface data and snow data, cleaning damaged data, and deleting data associated with the damaged data;
[0018] The cleaned data is subjected to z-score normalization processing to obtain the preprocessed data set.
[0019] Preferably, the process of obtaining energy balance data of the snow surface in each grid area through the distributed energy balance model includes:
[0020] Construct a distributed energy balance model based on energy balance theory;
[0021] The energy balance data of the snow surface in each grid area is obtained through the meteorological data, surface data and snow data in each grid area and the distributed energy balance model.
[0022] Preferably, the process of constructing an improved multilayer perceptron comprises:
[0023] Constructing a Transformer encoder, training the Transformer encoder, and obtaining a parameter output vector of the Transformer encoder;
[0024] A multilayer perceptron is constructed, the parameter output vector is used as the initial input of the multilayer perceptron, and an attention mechanism module is introduced into the hidden layer of the multilayer perceptron to obtain the improved multilayer perceptron.
[0025] Preferably, the attention mechanism module is introduced into the hidden layer of the multilayer perceptron, and the process of obtaining the improved multilayer perceptron includes: constructing a multilayer perceptron based on a feedforward neural network, using the parameter output vector as the starting input parameter of the input layer of the multilayer perceptron, and replacing the first three layers of the hidden layer of the multilayer perceptron with the attention mechanism module to obtain the improved multilayer perceptron.
[0026] Preferably, the process of obtaining the ablation rate prediction model includes:
[0027] Pre-training the improved multi-layer perceptron through a BERT pre-training model to obtain a pre-training model;
[0028] The improved multi-layer perceptron is iteratively trained by using the preprocessed data set as a sample set and the energy balance data as input features until the pre-trained model reaches the point where the relationship between the input parameters and the snow accumulation rate is automatically learned, and the iteration is terminated to obtain the ablation rate prediction model.
[0029] Preferably, the snow ablation rate is calculated based on the ablation rate prediction model, and the process of obtaining the rate calculation result includes: acquiring meteorological data, surface data and snow data in the real-time area, inputting the meteorological data, surface data and snow data in the real-time area into the ablation rate prediction model for calculation, and obtaining the snow ablation rate prediction curve.
[0030] Compared with the prior art, the present invention has the following advantages and technical effects:
[0031] The present invention combines a distributed energy balance model and a multi-layer perceptron to predict the snow melting rate. The distributed energy balance model is based on physical laws and processes, and considers in detail various energy exchange processes such as solar radiation, atmospheric radiation, thermal convection, and thermal conduction. It can more accurately describe the energy balance on the snow surface and the main energy sources that cause snow melting. As a powerful machine learning model, MLP can learn complex nonlinear relationships and establish a mapping relationship between input and output through a large amount of data training. The present invention combines a distributed energy balance model and a multi-layer perceptron to quickly and accurately determine the snow melting rate. The distributed energy balance model provides a detailed description and basis for the physical process, and the MLP learns the complex relationship between the snow melting rate and environmental factors from the data through machine learning methods, making the prediction more flexible and adaptable to changes in different conditions. This combination takes advantage of the advantages of both methods and helps to improve the comprehensiveness and practicality of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0033] Figure 1 This is a flow chart of a method for rapidly determining snow melting rate according to an embodiment of the present invention. DETAILED DESCRIPTION
[0034] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.
[0035] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0036] Embodiment 1
[0037] like Figure 1 As shown, this embodiment provides a method for quickly determining the snow melting rate, including:
[0038] Divide the target snow accumulation area into grids, obtain meteorological data, surface data and snow accumulation data in each grid area, and pre-process the meteorological data, surface data and snow accumulation data to obtain a pre-processed data set;
[0039] The energy budget balance data of the snow surface in each grid area is obtained through the distributed energy balance model;
[0040] Constructing an improved multilayer perceptron, and training the improved multilayer perceptron using the preprocessed data set and the energy balance data to obtain an ablation rate prediction model;
[0041] The snow ablation rate is calculated based on the ablation rate prediction model to obtain a rate calculation result.
[0042] To further optimize the scheme, the process of meshing the snow target area includes:
[0043] Acquire digital elevation model data of a target snow accumulation area, perform smoothing processing and missing value filling on the digital elevation model data, and obtain processed model data;
[0044] Get the digital elevation model (DEM) data of the target area. DEM data can be obtained through satellite remote sensing, lidar, etc., which provides elevation information of the surface. Smooth or filter the DEM data to remove possible noise and ensure the accuracy and smoothness of the elevation data. Process the missing values or blank areas that may exist in the DEM to ensure that the entire target area has complete elevation data.
[0045] The processed DEM data is used to perform grid division according to the set grid size and resolution. This embodiment adopts an orthogonal grid form, and then processes the grid boundary to ensure that the grid division is complete and continuous within the target area.
[0046] The processing model data is gridded based on an adaptive grid method, wherein the grid is an orthogonal grid.
[0047] To further optimize the scheme, the process of obtaining meteorological data, surface data, and snow data in each grid area includes:
[0048] Obtain meteorological parameter data for each grid area based on real-time data from weather stations;
[0049] Obtain terrain information within each grid area, and simultaneously obtain soil type and vegetation coverage data within each grid area;
[0050] Snow cover data is obtained, and a time series of the snow cover data is obtained.
[0051] The scheme is further optimized, and the meteorological data, surface data and snow data are preprocessed. The process of obtaining the preprocessed data set includes:
[0052] Checking the meteorological data, surface data and snow data, cleaning damaged data, and deleting data associated with the damaged data;
[0053] The cleaned data is subjected to z-score normalization processing to obtain the preprocessed data set.
[0054] The meteorological data, surface data and snow accumulation data are cleaned to remove abnormal values, missing values and duplicate data, etc., so as to ensure the quality and reliability of the data. After data cleaning, in this embodiment, the data is normalized to eliminate the unit differences and dimensional effects between different physical quantities, so as to make the data more comparable and analyzable. Through normalization, in this embodiment, the data of different physical quantities can be converted into a common numerical range, so as to better reflect their influence on the snow accumulation process.
[0055] Topographic factors such as altitude, slope, aspect and terrain roughness are obtained from DEM data. The slope and aspect are extracted using ENVI's TopographicModeling tool, and the terrain roughness is obtained by calculating the slope.
[0056] The surface temperature is obtained by inverting the modified universal single-channel algorithm. The key input parameter is the water vapor content, which is inverted using the MODIS1B product MOD02HKM with a resolution of 500m. After obtaining the water vapor content, it is uniformly cropped and resampled and then input into the model to obtain the surface temperature data results.
[0057] The precipitation data is selected from the China regional ground meteorological element driven data set, and the random forest method is used to downscale the precipitation data. Considering that the influence of daily precipitation on the snow melting process has a time lag, the present invention uses the accumulated precipitation between each typical snow melting day as the precipitation influencing factor of snow decay on the typical date.
[0058] To further optimize the scheme, the process of obtaining the energy balance data of the snow surface in each grid area through the distributed energy balance model includes:
[0059] Construct a distributed energy balance model based on energy balance theory;
[0060] The energy balance data of the snow surface in each grid area is obtained through the meteorological data, surface data and snow data in each grid area and the distributed energy balance model. Based on the collected meteorological data, snow characteristic data and terrain information, a distributed energy balance model is established. This includes considering the radiation balance (solar radiation and surface radiation), convection transmission (heat transfer between the atmosphere and the snow layer), heat conduction and other processes to calculate the energy balance of the snow surface.
[0061] The distributed energy balance model is based on the following basic energy balance equation, taking into account the absorption, radiation, conduction and convection processes of the surface:
[0062] [Q_{net}=Q_{SW}+Q_{LW}+Q_{H}+Q_{G}];
[0063] Where: (Q_{net}): net radiation energy (W / m 2 ), (Q_{SW}): Net solar radiation (W / m 2 ), (Q_{LW}): Net radiation of atmospheric longwave radiation (W / m 2 ), (Q_{H}): heat flux between the surface and the atmosphere (W / m 2 ), (Q_{G}): heat flux between the surface and the underlying soil (W / m 2 ), are obtained based on meteorological data, snow characteristics data and terrain information, and the Stefan-Boltzmann constant.
[0064] Further optimizing the scheme, the process of constructing the improved multilayer perceptron includes:
[0065] Constructing a Transformer encoder, training the Transformer encoder, and obtaining a parameter output vector of the Transformer encoder;
[0066] A multilayer perceptron is constructed, the parameter output vector is used as the initial input of the multilayer perceptron, and an attention mechanism module is introduced into the hidden layer of the multilayer perceptron to obtain the improved multilayer perceptron.
[0067] The scheme is further optimized, and the attention mechanism module is introduced into the hidden layer of the multilayer perceptron. The process of obtaining the improved multilayer perceptron includes: constructing a multilayer perceptron based on a feedforward neural network, using the parameter output vector as the starting input parameter of the input layer of the multilayer perceptron, and replacing the first three layers of the hidden layer of the multilayer perceptron with the attention mechanism module to obtain the improved multilayer perceptron.
[0068] This embodiment introduces the attention mechanism module into the hidden layer of the multilayer perceptron. The attention mechanism is a neural network structure that simulates the human visual attention mechanism, which can automatically focus on important parts and ignore irrelevant information when processing information. By embedding the attention mechanism module into the hidden layer of the multilayer perceptron, this embodiment can make the model more efficient and accurate when processing input data.
[0069] In terms of specific implementation, this embodiment replaces the first three layers of the hidden layer of the multilayer perceptron with the attention mechanism module. These three hidden layers were originally responsible for extracting the deep features of the input data, but after the introduction of the attention mechanism module, they can better capture the key information in the input data and pass it to the subsequent network layers. In this way, when the model is performing feature extraction and classification, it can pay more attention to the parts that have an important impact on the task objectives, thereby improving the performance of the model.
[0070] Further optimizing the scheme, the process of obtaining the ablation rate prediction model includes:
[0071] Pre-training the improved multi-layer perceptron through a BERT pre-training model to obtain a pre-training model;
[0072] The improved multi-layer perceptron is iteratively trained by using the preprocessed data set as a sample set and the energy balance data as input features until the pre-trained model reaches the point where the relationship between the input parameters and the snow accumulation rate is automatically learned, and the iteration is terminated to obtain the ablation rate prediction model.
[0073] The mean square error loss function is selected to measure the difference between the predicted value and the actual snow melting rate. Through repeated iterative training, the MLP model automatically adjusts the weights and biases to maximize the learning of the complex relationship between the input features (meteorological, surface and energy balance data) and the target output (snow melting rate).
[0074] Mean Squared Error (MSE) is a loss function commonly used in regression tasks, which is used to measure the average value of the squared difference between the model's predicted value and the true value. Its expression is as follows:
[0075] Assuming there are (n) samples, the predicted value is (\hat{y}_i), and the true value is (y_i), then the expression of the mean square error loss function is:
[0076] [\text{MSE}=\frac{1}{n}\sum_{i=1}^{n}(\hat{y}_i-y_i)^2]
[0077] in:
[0078] (\hat{y}_i) is the model's prediction for the (i)th sample,
[0079] (y_i) is the true value of the (i)th sample,
[0080] (n) is the sample size.
[0081] The smaller the MSE, the smaller the difference between the model's predicted value and the true value, that is, the better the model's fitting effect.
[0082] A further optimization scheme is to calculate the snow ablation rate based on the ablation rate prediction model, and the process of obtaining the rate calculation result includes: obtaining meteorological data, surface data and snow data in the real-time area, inputting the meteorological data, surface data and snow data in the real-time area into the ablation rate prediction model for calculation, and obtaining the snow ablation rate prediction curve.
[0083] In order to further measure the prediction effect of the model, the model prediction rate curve is drawn for each sub-region to analyze the quality of the model prediction results in a more intuitive way. The specific method is to use the established regional spatial panel data model to predict the data, with the error between the actual value and the predicted value (sorted from small to large) as the horizontal axis and the proportion of the sample number to the total as the vertical axis, and draw the model prediction rate curve for each region.
[0084] The specific implementation method of the model prediction rate curve is as follows: define the sample with prediction error lower than the set value as a good prediction sample point, and evaluate the model by the ratio of good prediction samples to prediction error. In this embodiment, the prediction error is controlled within 15%-25%. The specific method is to calculate the error between the fitting value and the actual value of each sample point and arrange them in order from small to large, and draw a cumulative curve of n sample points (1 / n, 2 / n, ... (n-1) / n, n), where the horizontal axis is the prediction error and the vertical axis represents the proportion of good prediction samples.
[0085] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A method for rapidly determining snow melting rate, characterized in that: The following steps are involved: Dividing the target snow accumulation area into grids, obtaining meteorological data, surface data and snow accumulation data in each grid area, and preprocessing the meteorological data, surface data and snow accumulation data to obtain a preprocessed data set; The energy budget balance data of the snow surface in each grid area is obtained through the distributed energy balance model; Constructing an improved multilayer perceptron, and training the improved multilayer perceptron using the preprocessed data set and the energy balance data to obtain an ablation rate prediction model; The snow ablation rate is calculated based on the ablation rate prediction model to obtain a rate calculation result.
2. The method for rapidly determining snow melting rate according to claim 1, characterized in that: The process of meshing the snow target area includes: Acquire digital elevation model data of a target snow accumulation area, perform smoothing processing and missing value filling on the digital elevation model data, and obtain processed model data; The processing model data is gridded based on an adaptive grid method, wherein the grid is an orthogonal grid.
3. The method for rapidly measuring snow melting rate according to claim 1, characterized in that: The process of obtaining meteorological data, surface data, and snow cover data in each grid area includes: Obtain meteorological parameter data for each grid area based on real-time data from weather stations; Obtain terrain information within each grid area, and simultaneously obtain soil type and vegetation coverage data within each grid area; Snow cover data is obtained, and a time series of the snow cover data is obtained.
4. The method for rapidly determining snow melting rate according to claim 1, characterized in that: The meteorological data, surface data and snow data are preprocessed to obtain a preprocessed data set, including: Checking the meteorological data, surface data and snow data, cleaning damaged data, and deleting data associated with the damaged data; The cleaned data is normalized by z-score to obtain the preprocessed data set.
5. The method for rapidly measuring snow melting rate according to claim 1, characterized in that: The process of obtaining the energy balance data of the snow surface in each grid area through the distributed energy balance model includes: Construct a distributed energy balance model based on energy balance theory; The energy balance data of the snow surface in each grid area is obtained through the meteorological data, surface data and snow data in each grid area and the distributed energy balance model.
6. The method for rapidly measuring snow melting rate according to claim 1, characterized in that: The process of building an improved multilayer perceptron includes: Constructing a Transformer encoder, training the Transformer encoder, and obtaining a parameter output vector of the Transformer encoder; A multilayer perceptron is constructed, the parameter output vector is used as the initial input of the multilayer perceptron, and an attention mechanism module is introduced into the hidden layer of the multilayer perceptron to obtain the improved multilayer perceptron.
7. The method for rapidly determining snow melting rate according to claim 6, characterized in that: The attention mechanism module is introduced into the hidden layer of the multilayer perceptron to obtain the improved multilayer perceptron, comprising: constructing a multilayer perceptron based on a feedforward neural network, using the parameter output vector as the starting input parameter of the input layer of the multilayer perceptron, and replacing the first three layers of the hidden layer of the multilayer perceptron with the attention mechanism module to obtain the improved multilayer perceptron.
8. The method for rapidly measuring snow melting rate according to claim 1, characterized in that: The process of obtaining the ablation rate prediction model includes: Pre-training the improved multi-layer perceptron through a BERT pre-training model to obtain a pre-training model; The improved multi-layer perceptron is iteratively trained by using the preprocessed data set as a sample set and the energy balance data as input features until the pre-trained model reaches the point where the relationship between the input parameters and the snow accumulation rate is automatically learned, and the iteration is terminated to obtain the ablation rate prediction model.
9. The method for rapidly measuring snow melting rate according to claim 1, characterized in that: The snow ablation rate is calculated based on the ablation rate prediction model, and the process of obtaining the rate calculation result includes: obtaining meteorological data, surface data and snow data of the real-time area, inputting the meteorological data, surface data and snow data of the real-time area into the ablation rate prediction model for calculation, and obtaining the snow ablation rate prediction curve.
