Hydrological model automatic calibration method and system oriented to user-defined context

Abstract

The invention belongs to a hydrological model parameter automatic calibration technology, and relates to a hydrological model automatic calibration method and system oriented to a user-defined context. According to the method, to-be-optimized parameters of a hydrological model specified by a user and a value range of the to-be-optimized parameters are obtained, and the to-be-optimized parameters are optimized by adopting an automatic hydrological model parameter optimization algorithm combining a meta-knowledge base and a Bayesian optimization algorithm. The system comprises a client, an application service layer and a resource layer. By adopting the method and the device, the optimal model parameter combination can be automatically generated only by inputting related parameters and data by a user, efficient calibration and application of the hydrological model are realized, the workload and the time cost of the user can be remarkably reduced, the user can be supported to customize a space range and parameter setting, and personalized requirements are met. According to the method, the meta-learning algorithm and the Bayesian optimization algorithm are combined, the historical optimization result is fully utilized, and the complex hydrological model parameter optimization process is more efficient and automatic.

Description

Technical Field

[0001] The present invention belongs to the automatic calibration technology of hydrological model parameters, involving automatic preprocessing of model input data for user-defined time and space range and Bayesian optimization algorithm combined with meta-learning. The technology can be widely used in water resources management, flood forecasting, disaster risk assessment and other fields. Background Art

[0002] In recent years, with the impact of global climate change and human activities, accurate prediction of large-scale hydrological processes has become particularly important for national and regional water resources management and flood warning. To meet this challenge, hydrological models such as the VIC (Variable Infiltration Capacity) model have been widely used in research. The input of the VIC model includes datasets such as topography, soil, land use, and climate variables (such as rainfall, temperature, relative humidity, wind speed). Before using these data for simulation, they need to be extensively processed and converted, including mosaicking, clipping, resampling, unit conversion, cleaning, format conversion, etc. These data processing processes require professional knowledge and computer skills, which makes the use of the VIC model mostly limited to research institutions with technical resources. At the same time, with the application of the model in a larger spatial range, the demand for flexibility in input data is increasing.

[0003] Hydrological models usually rely on a combination of physical equations and empirical equations to describe the watershed system. However, since model parameters cannot be obtained through direct measurement, parameter optimization becomes the key to model accuracy. Traditional manual parameter adjustment methods are time-consuming and difficult to reproduce, and automated parameter optimization methods have gradually become a research hotspot. Grid search, random search, and Bayesian optimization are common automatic search methods. However, grid search is prone to the curse of dimensionality. Although random search is simple, it is inefficient. Bayesian optimization builds a proxy model to predict the optimal solution for each step, thereby gradually approaching the global optimal solution and showing higher efficiency.

[0004] However, in complex and changeable hydrological environments, Bayesian optimization also has certain limitations, especially when the initial data is insufficient, the optimization process may become inefficient. At this time, the introduction of meta-learning provides a new perspective for parameter optimization of hydrological models. The core idea of ​​meta-learning is to learn from past optimization tasks and guide the optimization process of new tasks by extracting common knowledge from different tasks. For example, the meta-knowledge base can store optimization parameters and result data from historical tasks, and provide initial conditions in new tasks through knowledge transfer, thereby accelerating the optimization convergence of the model. Meta-learning can help the optimization algorithm quickly adapt to new tasks, reduce the number of initial model sampling, and improve the overall efficiency of the algorithm.

[0005] By combining meta-learning and Bayesian optimization, parameter migration can be achieved between different watersheds, especially in scenarios with similar time and space. This method significantly improves the effect of parameter optimization. Meta-learning not only speeds up the optimization process, but also provides more robust parameter initialization values ​​for the hydrological model, thereby improving the accuracy and adaptability of the model in large-scale watershed simulation. Summary of the invention

[0006] In response to the difficulty in calibrating hydrological model parameters, the present invention provides a method and system for automatic calibration of hydrological models based on user-defined contexts, which can realize automatic optimization of hydrological model parameters, and aims to provide an efficient and flexible method for calibrating hydrological models that meets the personalized needs of users for model calibration.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A method for automatic calibration of a hydrological model for a user-defined context includes the following steps:

[0009] Obtain the parameters to be optimized and their value ranges of the hydrological model specified by the user;

[0010] The hydrological model parameter automatic optimization algorithm combining meta-knowledge base and Bayesian optimization algorithm is used to optimize the parameters to be optimized.

[0011] Furthermore, the hydrological model parameter automatic optimization algorithm combining the meta-knowledge base and the Bayesian optimization algorithm includes the following steps:

[0012] Retrieve the meta-knowledge base based on the task information submitted by the user and measure the similarity between watersheds by constructing spatiotemporal vectors;

[0013] According to the similarity between watersheds, the k parameter sets that are most similar to the target watershed in terms of temporal and spatial characteristics are selected as the initial sampling points for Bayesian optimization;

[0014] Select the objective function for optimizing the hydrological model parameters and calculate the objective function value of the parameter set at the initial sampling point;

[0015] Construct a parameter sample set based on the parameter set of the initial sampling point and the corresponding objective function value;

[0016] Iterative optimization is performed based on a sample set of parameters.

[0017] Furthermore, the meta-knowledge base stores optimization results of previous tasks, including task identifiers, watershed identifiers, watershed boundary vectors, watershed feature vectors, simulation time, optimization parameter sets and their corresponding objective function values.

[0018] Furthermore, the similarity between the watersheds takes into account the characteristics of the watersheds and the differences in different time periods. The measurement formula of the space-time vector is as follows:

[0019] d p (w i ,w j )=||m i -m j || p

[0020] Among them, d p Represents the watershed w i and watershed w j The distance between i and m j The watershed w i and watershed w j The eigenvector of ,||.|| p represents the p-norm.

[0021] Furthermore, the iterative optimization based on the parameter sample set includes:

[0022] Based on the parameter sample set, the objective function is modeled using a Bayesian surrogate model;

[0023] Randomly sample multiple candidate parameter sets in the parameter space;

[0024] Use the Bayesian surrogate model to predict each candidate parameter set and obtain the mean and variance of the corresponding objective function;

[0025] According to the prediction results of the Bayesian surrogate model, the acquisition function value of each candidate parameter set is calculated;

[0026] Select the candidate parameter set with the largest acquisition function value from the candidate parameter set as the candidate point for the next iteration;

[0027] Calculate the objective function value corresponding to the selected candidate point and add it to the parameter sample set;

[0028] Check the convergence conditions and stop the optimization if the preset maximum number of iterations or the threshold of the number of iterations without improvement is reached; otherwise, continue the iteration;

[0029] Output the optimized parameter results and return them to the user.

[0030] Furthermore, the acquisition function is a desired improved acquisition function.

[0031] Furthermore, the hydrological model is a VIC model, and the parameters to be optimized include:

[0032] B: The power of the variable infiltration curve equation, which is the infiltration curve shape parameter that affects rainfall infiltration and surface runoff yield;

[0033] Ds: The ratio of the flow velocity when nonlinear base flow occurs to the maximum base flow velocity Dsmax;

[0034] Dsmax: maximum base flow velocity;

[0035] Ws: the ratio of soil moisture content to saturated moisture content when nonlinear base flow occurs;

[0036] D2: thickness of the second soil layer;

[0037] D3: Thickness of the third soil layer.

[0038] An automatic calibration system of hydrological model for user-defined contexts, comprising a client, an application service layer, and a resource layer;

[0039] The client is responsible for user interaction, including user-defined parameter settings, task submission and result viewing; the user-defined parameter settings are the parameters required for the operation of the hydrological model input and configuration by the user, and submitted to the application service layer for processing;

[0040] The application service layer is responsible for task parameter analysis, data preprocessing, model automatic optimization, task scheduling service and map rendering service; after receiving the task request submitted by the client, the application service layer performs task parameter analysis, and preprocesses the data required by the hydrological model to generate a data set and model configuration file that meets the hydrological model input standard; the model automatic optimization adopts the above method of the present invention to automatically calibrate the hydrological model to obtain parameter optimization results; the task scheduling service is responsible for managing and allocating computing resources; for tasks that require geographic information display, the map rendering service is used to visualize the calculation results on the map;

[0041] The resource layer includes data storage resources and computing resources.

[0042] By integrating the above technologies, the present invention provides a service for automatic optimization of hydrological model parameters. Users only need to input relevant parameters and data, and the system can automatically generate the optimal model parameter combination to achieve efficient calibration and application of the hydrological model. Compared with the prior art, the present invention can automatically generate the input data required for the model, significantly reducing the workload and time cost of users; support users to customize spatial ranges and parameter settings to meet personalized needs; combine meta-learning with Bayesian optimization algorithms, make full use of historical optimization results, and make the complex hydrological model parameter optimization process more efficient and automated. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1It is an architectural diagram of the hydrological model automatic calibration system for user-defined contexts of the present invention.

[0044] Figure 2 The black box function is a process of configuring the hydrological model parameters, running the hydrological model to obtain simulation results, and calculating the objective function value based on these results. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below through specific embodiments and drawings.

[0046] The hydrological model automatic calibration system for user-defined contexts of the present invention adopts a three-layer architecture, namely, a client, an application service layer, and a resource layer. Figure 1 shown.

[0047] The client is mainly responsible for user interaction, including user-defined parameter settings, task submission, and result viewing. In terms of parameter settings, users input and configure the parameters required for model operation through the client interface, including but not limited to the simulated watershed, simulation time period, parameters to be optimized, parameter space, etc., and submit these configuration items to the application service layer for processing. After the task is completed, users can view and analyze the results through the client, including the output data of the calculation model, statistical analysis results, and visualization charts.

[0048] The application service layer is responsible for task parameter parsing, data preprocessing, model automatic optimization, task scheduling service (task distributed scheduling) and map rendering service. First, the application service layer receives the task request submitted by the client, performs task parameter parsing, and extracts user-defined information such as the simulated watershed of the hydrological model, the simulation time period, and the spatial resolution of the results. Data preprocessing refers to the preprocessing of meteorological driving data, vegetation parameter data, soil parameter data, etc. required by the hydrological model, including spatial cropping for the user-selected watershed spatial range, resampling the data according to the resolution, converting the format to the standard format of the hydrological model input data, etc., to generate a data set and model configuration file that meets the hydrological model input standard. The model automatic optimization module mainly obtains the optimization results through multiple iterations of the optimizer. The task scheduling service is responsible for managing and allocating computing resources, including managing the model parameter optimization task queue submitted by the user, judging the computing resource demand according to the simulated time and space range, and optimizing resource allocation. For tasks that require geographic information display, the system provides map rendering services to visualize the calculation results on the map.

[0049] The resource layer includes data storage and computing resources. Data storage includes the storage of vector data (Vector), raster data (Raster), task information (Task), and optimization parameter library. Computing resources include physical servers and virtual computing nodes, which are managed and allocated through task scheduling services.

[0050] The model automatic optimization module in the application service layer of the present invention uses a framework combining meta-knowledge base and Bayesian optimization algorithm to automatically optimize complex hydrological models. Its core process is as follows: Figure 2 As shown in the figure, it mainly includes three parts: meta-knowledge base, meta-guidance and Bayesian optimization.

[0051] The meta-knowledge base stores the optimization results of previous tasks, including task ID (id), watershed ID (wid), watershed boundary vector (geojson), watershed feature vector (feature), simulation time (time), optimization parameter set (x) and its corresponding objective function value (y), etc. The watershed feature vector covers multi-dimensional data such as terrain features, climate index, hydrological characteristics, land cover, soil, and geological characteristics.

[0052] The meta-guidance module provides initial conditions for new optimization tasks based on the meta-knowledge base through knowledge transfer and initial sampling point setting. Specifically, it determines the initial sampling points for the Bayesian optimizer by utilizing the historical optimization results of similar watersheds or time periods.

[0053] After the Bayesian optimizer determines the initial sampling point, it starts the optimization process and obtains the optimization result after multiple iterations. The specific process is as follows:

[0054] First, the actual objective function value of each parameter set in the initial sampling point is calculated. The parameter set and the corresponding objective function value constitute the parameter sample set.

[0055] Secondly, iterative optimization is performed, the steps of which include:

[0056] 1) Train the Bayesian proxy model based on the existing parameter sample set.

[0057] 2) Randomly generate candidate parameter sets;

[0058] 3) Infer based on the Bayesian surrogate model to generate the posterior distribution of the candidate parameter set;

[0059] 4) Calculate the acquisition function of the candidate parameter set;

[0060] 5) Select the candidate parameter set with the largest acquisition function value as the candidate point for the next iteration;

[0061] 6) Calculate the objective function value of the candidate point and update the parameter sample set.

[0062] Finally, after the convergence conditions are met, the optimal solution (optimal combination of hydrological model parameters) and its objective function value are returned.

[0063] The variable infiltration capacity (VIC) land surface model (hereinafter referred to as the VIC model) is a macro-scale model used to simulate and analyze hydrological processes. It can predict floods, droughts and water supply and demand under different meteorological and topographic conditions, and assess the impact of climate change on water resources. It has been applied in hundreds of studies in catchment areas, regions and the world. The following takes the automatic calibration of VIC model parameters as an example to further illustrate the present invention.

[0064] First, the user specifies the parameters to be optimized and their value ranges of the VIC model. Existing research results show that the runoff generation of the VIC model is sensitive to the following six parameters:

[0065] B: The power of the variable infiltration curve equation is the infiltration curve shape parameter that affects rainfall infiltration and surface runoff yield. The higher the value, the less infiltration and the more surface runoff.

[0066] Ds: The ratio of the flow velocity when nonlinear base flow occurs to the maximum base flow velocity Dsmax; when the Ds value increases, the base flow in the lower soil layer will increase at low soil moisture content.

[0067] Dsmax: Maximum base flow velocity, unit: mm / day.

[0068] Ws: The ratio of soil moisture content to saturated moisture content when nonlinear base flow occurs; a high value of Ds will increase the soil moisture content that reflects the nonlinear base flow, which will inevitably delay the occurrence of the flood peak.

[0069] D2: thickness of the second soil layer, in m.

[0070] D3: Thickness of the third soil layer, in meters. The thickness of the soil layers other than the first layer (the second and third layers), if increased, will increase evaporation and delay the peak of base flow.

[0071] Among the above six parameters, the value range of each parameter can be obtained from the literature.

[0072] Then, the user submits the parameter optimization task to the application service layer, and the system uses the hydrological model parameter automatic optimization algorithm combining the meta-knowledge base and the Bayesian optimization algorithm of the present invention to automatically optimize the parameters of the VIC model. The specific steps are as follows:

[0073] Step 1: According to the task information submitted by the user, the meta-knowledge base is retrieved and the similarity between watersheds is measured by constructing a spatiotemporal vector. This similarity not only considers the characteristics of the watershed (such as topography, hydrology, climate, etc.), but also comprehensively considers the differences in different time periods, because the topographic characteristics and environmental factors in different time periods will be different. The measurement formula of the spatiotemporal vector is as follows:

[0074] d p (w i ,w j )=||m i -m j || p (1)

[0075] In formula (1), d p Represents the watershed w i and watershed w j The distance between i and m j The watershed w i and watershed w j The characteristic vector of ,||.|| p represents the p-norm.

[0076] Step 2: Based on the calculated distance d between watersheds p , select k parameter sets X1,X2,...,X that are most similar to the target basin in terms of temporal and spatial characteristics k As the initial sampling point of Bayesian optimization (one parameter set is used as a sampling point, there are k sampling points in total), the sampling point X i It can be expressed as follows:

[0077] X i =(x i1 ,x i2 ,...,x iq )

[0078] Among them, x i1 ,x i2 ,...,x iq Indicates the sampling point X i The sub-parameters of are B, Ds, Dsmax, Ws, D2 and D3 respectively. q represents the parameter dimension, that is, the number of sub-parameters, which is 6 here.

[0079] Step 3: Select the objective function for optimizing the VIC model parameters and calculate the objective function value for each parameter set. Commonly used hydrological model evaluation objective functions include Nash–Sutcliffe efficiency (NSE) sensitive to high flow, flow efficiency coefficient (VE) sensitive to water balance, log Nash efficiency coefficient (logNSE) sensitive to low flow, Kling-Gupta efficiency coefficient (KGB) sensitive to runoff changes, and Box–Cox transformed root mean square error (TRMSE), runoff coefficient error (ROCE) and mean square derivative error (MSDE).

[0080] The calculation formula of flow efficiency coefficient (VE) is as follows:

[0081]

[0082] In formula (2), N is the time step, Q ot is the observed runoff, Q st is the runoff volume simulated by the VIC model.

[0083] The calculation formula of Nash efficiency coefficient (NSE) is as follows:

[0084]

[0085] In formula (3) is the observed mean runoff volume.

[0086] The calculation formula of log Nash efficiency coefficient (logNSE) is as follows:

[0087]

[0088] In formula (4), log is the natural logarithm.

[0089] The Kling-Gupta efficiency coefficient (KGE) is sensitive to runoff changes and is defined as follows:

[0090]

[0091] In formula (5), r is the linear correlation coefficient. σ is the standard deviation, σ s is the standard deviation of the simulation results, σ o is the standard deviation of the observed data, μ is the mean, and μ s is the mean of the simulation results, μ o is the mean of the observed data.

[0092] Step 4: Use the parameter set X = {X1, X2, ..., X k} and the corresponding objective function values ​​y1,y2,...,yk , construct parameter sample set

[0093] Step 5: Set a threshold for the number of iterations without improvement during the optimization process. Stop the iteration when the threshold is reached. The specific steps are as follows:

[0094] 1) Based on the parameter sample set D, the target function is modeled using a Bayesian surrogate model. You can choose a Gaussian process model, a random forest, etc. as the surrogate model. If you choose a Gaussian process model, one of the commonly used kernel functions is the radial basis function (RBF) kernel. The RBF kernel function is defined as follows:

[0095]

[0096] In formula (6), K(x,x′) is the kernel function value between the input vectors x and x′; ‖xx′‖ 2 Represents the square of the Euclidean distance between x and x'; σ is a scale parameter that controls the "width" of the kernel function.

[0097] 2) Candidate parameter set sampling. Randomly sample n candidate parameter sets X in the parameter space ′ ={X1 ′ ,X2 ′ ,...,X ′ n}. These candidate parameter sets are used for prediction and optimization of the Bayesian surrogate model in subsequent steps. Here n is set to 10000.

[0098] 3) Use the Bayesian surrogate model for each candidate parameter set X′ i Make a prediction and get the corresponding mean value μ(X′ i ) and uncertainty (i.e. variance) σ(X′ i ).

[0099] 4) According to the prediction results of the Bayesian surrogate model, the acquisition function value of each candidate parameter set is calculated. Here, the expected improvement (EI) acquisition function is used, which is defined as:

[0100] EI(X)=E[max(0,yy * )] (7)

[0101] y in formula (7) * is the currently known optimal objective function value, and y is the objective function value. The expected improvement combines the mean and uncertainty of the model prediction to select a parameter set that can bring potential improvements.

[0102] 5) Select the candidate parameter set X with the largest EI value from the candidate parameter set next, as the candidate point for the next iteration.

[0103] 6) Calculate the selected candidate point X next The corresponding objective function value y next , and add it to the parameter sample set D.

[0104] 7) Check the convergence conditions. If the preset maximum number of iterations is reached, or the threshold of the number of iterations without improvement is reached (that is, the objective function value does not change significantly in several iterations), stop the optimization; otherwise, return to 1) and continue iterating.

[0105] Finally, the optimized parameter results are output and returned to the user.

[0106] In the above embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are only schematic, and the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the parameters mentioned in the above embodiments can be replaced with other parameters within a reasonable range according to actual applications.

[0107] The integrated modules or units implemented in the form of software functional units in the above embodiments can be stored in a computer-readable storage medium. The above software functional units are stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0108] The specific embodiments of the present invention disclosed above are intended to help understand the content of the present invention and implement it accordingly. It can be understood by those skilled in the art that various replacements, changes and modifications are possible without departing from the spirit and scope of the present invention. The present invention should not be limited to the contents disclosed in the embodiments of this specification, and the scope of protection of the present invention shall be subject to the scope defined in the claims.

Claims

1. A method for automatic calibration of hydrological models for user-defined contexts, characterized in that: The following steps are involved: Obtain the parameters to be optimized and their value ranges of the hydrological model specified by the user; The hydrological model parameter automatic optimization algorithm combining meta-knowledge base and Bayesian optimization algorithm is used to optimize the parameters to be optimized.

2. The method according to claim 1, characterized in that The hydrological model parameter automatic optimization algorithm combining the meta-knowledge base and the Bayesian optimization algorithm comprises the following steps: Retrieve the meta-knowledge base based on the task information submitted by the user and measure the similarity between watersheds by constructing spatiotemporal vectors; According to the similarity between watersheds, the k parameter sets that are most similar to the target watershed in terms of temporal and spatial characteristics are selected as the initial sampling points for Bayesian optimization; Select the objective function for optimizing the hydrological model parameters and calculate the objective function value of the parameter set at the initial sampling point; Construct a parameter sample set based on the parameter set of the initial sampling point and the corresponding objective function value; Iterative optimization is performed based on a sample set of parameters.

3. The method according to claim 2, characterized in that The meta-knowledge base stores the optimization results of previous tasks, including task identification, watershed identification, watershed boundary vector, watershed feature vector, simulation time, optimization parameter set and its corresponding objective function value.

4. The method according to claim 2, characterized in that: The similarity between the watersheds takes into account the characteristics of the watersheds and the differences in different time periods. The measurement formula of the space-time vector is as follows: d p (w i ,w j )=||m i -m j || p Among them, d p Represents the watershed w i and watershed w j The distance between i and m j The watershed w i and watershed w j The eigenvector of ,||.|| p represents the p-norm.

5. The method according to claim 2, characterized in that: The iterative optimization based on the parameter sample set includes: Based on the parameter sample set, the objective function is modeled using a Bayesian surrogate model; Randomly sample multiple candidate parameter sets in the parameter space; Use the Bayesian surrogate model to predict each candidate parameter set and obtain the mean and variance of the corresponding objective function; According to the prediction results of the Bayesian surrogate model, the acquisition function value of each candidate parameter set is calculated; Select the candidate parameter set with the largest acquisition function value from the candidate parameter set as the candidate point for the next iteration; Calculate the objective function value corresponding to the selected candidate point and add it to the parameter sample set; Check the convergence conditions and stop the optimization if the preset maximum number of iterations or the threshold of the number of iterations without improvement is reached; otherwise, continue the iteration; Output the optimized parameter results and return them to the user.

6. The method according to claim 5, characterized in that The acquisition function is a desired improved acquisition function.

7. The method according to claim 1, characterized in that The hydrological model is a VIC model, and its parameters to be optimized include: B: the power of the variable infiltration curve equation, which is the infiltration curve shape parameter that affects rainfall infiltration and surface runoff production; Ds: the ratio of the flow velocity when nonlinear base flow occurs to the maximum base flow velocity Dsmax; Dsmax: maximum base flow velocity; Ws: the ratio of soil moisture content to saturated moisture content when nonlinear base flow occurs; D2: thickness of the second soil layer; D3: Thickness of the third soil layer.

8. A hydrological model automatic calibration system for user-defined contexts, characterized in that: Includes client, application service layer, and resource layer; The client is responsible for user interaction, including user-defined parameter settings, task submission and result viewing; the user-defined parameter settings are the parameters required for the operation of the hydrological model input and configuration by the user, and submitted to the application service layer for processing; The application service layer is responsible for task parameter analysis, data preprocessing, model automatic optimization, task scheduling services and map rendering services; After receiving the task request submitted by the client, the application service layer performs task parameter analysis and preprocesses the data required by the hydrological model to generate a data set and a model configuration file that meets the hydrological model input standard; the model automatic optimization adopts the method described in any one of claims 1 to 7 to automatically calibrate the hydrological model to obtain parameter optimization results; The task scheduling service is responsible for managing and allocating computing resources; for tasks that require geographic information display, the computing results are visualized on the map through the map rendering service; The resource layer includes data storage resources and computing resources.

9. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, the computer program is configured to be executed by the processor, and the computer program comprises instructions for executing the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a computer, the method according to any one of claims 1 to 7 is implemented.

Images