CN116757098B - An automated verification method based on SWAT model multi-objective optimization - Google Patents

Abstract

The invention discloses an automatic verification method based on SWAT model multi-objective optimization, which comprises the following steps: s1, initializing a verification program of a SWAT model; s2, setting verification parameters, wherein the specific setting process is as follows: s21, setting an optimization algorithm, the initial population number, the iteration number, the sub population number and the cross variation value of the optimization algorithm, and setting the process number; s22, setting a parameter set range for verification and a parameter adjustment mode of each parameter in the model verification process, wherein the parameter adjustment mode comprises a relative change rate method, an absolute change rate method and a replacement rate method; s3, iteratively operating the SWAT model; s4, evaluating the model by using an optimization algorithm; s5, running the SWAT model for checking completion; the invention can perform multi-objective optimization and multi-site verification, improves the verification efficiency, and solves the problems of identical parameter and different parameter of the model.

Description

An automated verification method based on SWAT model multi-objective optimization

Technical field

The invention relates to the field of computer processing technology, and specifically relates to an automated verification method based on SWAT model multi-objective optimization.

Background technique

The SWAT (Soil and Water Assessment Tool) model is currently one of the most widely used hydrological and water quality models. This model plays an obvious role in hydrology and water quality simulation and watershed management scenario analysis. It is of great help to watershed managers in understanding the hydrology and water quality cycle process in the watershed and formulating watershed management plans. However, depending on the settings of the SWAT model, it may have thousands or more parameters, which need to be verified before they can be used in actual research. The calibration methods of the SWAT model are generally divided into two methods: manual calibration and automatic calibration. Manual calibration requires users to have high professional knowledge and be very familiar with the model, and since the SWAT model has many parameters, using manual calibration to verify the model will consume a lot of time and energy. The essence of automatic calibration is a trial and error method. The calibration process requires continuous iteration by computers to optimize the fit between simulated data and monitoring data, and finally find a set of parameter combinations that achieve a better fit. With the development of computer technology and artificial intelligence, the current optimization methods used in hydrology and water quality model verification mainly include SUFI-2, POS, NSGA3 and SA, etc. These methods have now been integrated into software such as SWAT-CUP and R-SWAT. Although these algorithms and software speed up the verification process of the SWAT model, they still have three shortcomings: (1) Using these algorithms cannot solve the problem of the same parameters and different effects of the model and the same effect of different parameters; (2) Integration of the two software The algorithms are all single-objective or dual-objective optimization algorithms. These algorithms may have problems such as slow convergence when dealing with multi-site model verification tasks; (3) Software generally uses multi-threading to speed up model verification, but SWAT The model is a computationally intensive model, and the multi-process model iteration method can more effectively speed up model verification. Therefore, there is still a large room for improvement in the verification process of the SWAT model, and further optimization is still needed to meet people's actual use needs. Based on this, we invented an automated verification method for the SWAT model, which integrates SWAT model parameter modification, multi-objective optimization algorithm and multi-process operation mode. This method can provide users with multiple sets of model parameter solutions, improve model verification efficiency, and solve the problem of models with the same parameters having different effects and different parameters having the same effect.

Contents of the invention

The purpose of the present invention is to provide an automated verification method based on SWAT model multi-objective optimization. The automated verification method based on SWAT model multi-objective optimization integrates SWAT model parameter modification, multi-objective optimization algorithm and multi-process operation mode. , providing users with multiple sets of model parameter solutions, which can carry out multi-objective optimization and multi-site verification, improve verification efficiency, and solve the problem of models with the same parameters having different effects and different parameters having the same effect.

In order to achieve the above objects, the present invention adopts the following technical solutions:

An automated verification method based on SWAT model multi-objective optimization, including the following steps:

S1. Initialize the verification program of the SWAT model, and construct a framework structure that couples the SWAT model with the multi-objective optimization method. Under this framework, the multi-objective optimization method is applied to the SWAT model verification. The specific initialization process is:

S11. Copy the input and output folders of the constructed SWAT model from the ArcSWAT model;

S12. Determine the hydrology and water quality indicators for verification, and create an EXCEL file of monitoring data for the corresponding indicators;

S13. Export the location table in the SWAT model to the location_table folder;

S2. Set the verification parameters. The specific setting process is:

S21. Set the optimization algorithm used and the initial population number, iteration number, sub-population number and cross-mutation value of the optimization algorithm, and set the number of processes used;

S22. Set the parameter set range for verification and the parameter adjustment method of each parameter during the model verification process. The parameter adjustment methods include the relative change rate determination method, the absolute change rate determination method and the replacement rate determination method;

The calculation formula of the relative change rate determination method is: ,

The calculation formula of the absolute change rate determination method is: ,

The calculation formula of the substitution rate determination method is: ,

in, is the modified parameter value,/> is the original value of the parameter in the SWAT model, and x is the parameter change factor;

S3. Iteratively run the SWAT model;

S4. Use optimization algorithms to evaluate the model;

S5. Run the verified SWAT model.

Preferably, the location table in step S13 includes an hru table, a rch table and a sub table.

Preferably, step S3 iteratively runs the SWAT model according to the initial population and the number of sub-populations of each generation set in step S21. The specific process is:

S31. Use the sampling method to collect parameter samples from the parameter set range in step S22 according to the initial population number or sub-population number, and generate a parameter table set;

S32. Run the SWAT model iteratively according to the number of processes set in step S21.

Preferably, the sampling method in step S31 is random sampling, binomial random sampling or hyper-Latin cube sampling; the specific process of step S32 is:

S321. Correlate the parameter table in step S31 with the position table in step S13, and generate a position parameter table that needs to be modified;

S322. Filter out the corresponding files from the input and output folders according to the position parameter table, and store them in the memory;

S323. For the files that need to be modified in the position parameter table, use different functions to modify them according to the attributes of each file, and import the modified and current files into the running folder for model simulation;

S324. After the SWAT model in the running folder has been run, perform the following operations. First, restore the model files in the running folder to the original state of the model based on the files filtered in step S322, and then use the read file module to read the running files. The model results in the folder are read out. The read files include hydrological response unit data, sub-basin data and river data, and the read files, parameter tables and location parameter tables are saved together in the output folder for use with monitoring data. Carry out the fitting degree of the simulation values and monitoring values of the calculation model.

Preferably, the specific process of step S4 is:

S41. Calculate the fitting degree index based on the EXCEL file of the monitoring data created in step S12 and the simulation value read in step S324. The calculation formula of the fitting degree index is: Among them,/> Indicates the goodness of fit,/> represents the i-th data, represents the Nash coefficient,/> Represents the i-th monitoring value,/> Represents the average of all monitoring values,/> Represents the i-th analog value,/> Represents the average value of simulated values;

S42. All simulation results in the initial population and sub-population are calculated according to the fitting index in step S41 above, and all fitness indicators are stored in the memory;

S43. Repeat the calculation process of step S41 for all SWAT simulation results of the initial population and sub-populations;

S44. According to the optimization algorithm and the number of iterations set in step S21, repeat steps S31 to S42, and screen out non-dominated solutions in the initial population and subpopulations;

S45. When all iteration calculations are completed, calculate all non-dominated solutions and obtain the Parteo optimal solution.

Preferably, the specific process of step S5 is:

S51. Compare the parameter tables of the SWAT model corresponding to all Parteo optimal solutions, and select solutions that meet the watershed simulation conditions as the final parameter set of the model;

S52. Substitute the final parameter set into the SWAT model to complete automatic model verification.

After adopting the above technical solution, the present invention has the following beneficial effects:

1. The present invention can perform multi-objective optimization and multi-site verification. A variety of optimization algorithms are embedded in the SWAT model verification process (as shown in Table 1). Compared with the previous single-objective optimization algorithm, it can target multiple The calibration results are analyzed using several hydrology and water quality monitoring stations and multiple fitness evaluation indicators. These optimization algorithms effectively optimize the calibration process.

Table 1 Optimization algorithm used

2. The present invention can effectively improve the verification efficiency. It adds a multi-process framework to the SWAT model and can use the computer CPU to simulate multiple SWAT models at the same time. This makes the verification efficiency of the present invention higher than that of SWAT-CUP and R. -SWAT.

3. The present invention can effectively solve the problem of models with the same parameters having different effects and different parameters having the same effect. It uses an optimization algorithm to verify the SWAT model. It will return multiple parameter sets. The user can compare the simulation results of several parameter sets. , and combined with professional knowledge to determine a parameter set that is more consistent with the hydrology and water quality simulation conditions of the basin, which can solve the problem of the same parameters and different effects of the model and the same effect of different parameters.

Description of the drawings

Figure 1 is a flow chart of the present invention;

Figure 2 is a flow chart of a single SWAT model verification of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with examples. It should be understood that the specific embodiments described here are only used to explain the present invention and are not intended to limit the present invention.

As shown in Figures 1 to 2, an automated verification method based on SWAT model multi-objective optimization includes the following steps:

S1. Initialize the verification program of the SWAT model, and construct a framework structure that couples the SWAT model with the multi-objective optimization method. Under this framework, the multi-objective optimization method is applied to the SWAT model verification. The specific initialization process is:

S11. Copy the input and output folders of the constructed SWAT model from the ArcSWAT model;

S12. Determine the hydrology and water quality indicators for verification, and create an EXCEL file of monitoring data for the corresponding indicators;

S13. Export the location table in the SWAT model to the location_table folder;

The location table described in step S13 includes hru table, rch table and sub table;

S2. Set the verification parameters. The specific setting process is:

S21. Set the optimization algorithm used and the initial population number, iteration number, sub-population number and cross-mutation value of the optimization algorithm, and set the number of processes used;

S22. Set the parameter set range for verification and the parameter adjustment method of each parameter during the model verification process. The parameter adjustment methods include the relative change rate determination method, the absolute change rate determination method and the replacement rate determination method;

The calculation formula of the relative change rate determination method is: ,

The calculation formula of the absolute change rate determination method is: ,

The calculation formula of the substitution rate determination method is: ,

in, is the modified parameter value,/> is the original value of the parameter in the SWAT model, and x is the parameter change factor;

S3. Iteratively run the SWAT model;

Step S3 iteratively runs the SWAT model according to the initial population and the number of sub-populations of each generation set in step S21. The specific process is:

S31. Use the sampling method to collect parameter samples from the parameter set range in step S22 according to the initial population number or sub-population number, and generate a parameter table set;

S32. Iteratively run the SWAT model according to the number of processes set in step S21;

The sampling method described in step S31 is random sampling, binomial random sampling or hyper-Latin cube sampling; the specific process of step S32 is:

S321. Correlate the parameter table in step S31 with the position table in step S13, and generate a position parameter table that needs to be modified;

S322. Filter out the corresponding files from the input and output folders according to the position parameter table, and store them in the memory;

S323. For the files that need to be modified in the position parameter table, use different functions to modify them according to the attributes of each file, and import the modified and current files into the running folder for model simulation;

S324. After the SWAT model in the running folder has been run, perform the following operations. First, restore the model files in the running folder to the original state of the model based on the files filtered in step S322, and then use the read file module to read the running files. The model results in the folder are read out. The read files include hydrological response unit data, sub-basin data and river data, and the read files, parameter tables and location parameter tables are saved together in the output folder for use with monitoring data. Perform the fitting of the calculation model simulation values and monitoring values;

S4. Use optimization algorithms to evaluate the model;

The specific process of step S4 is:

S41. Calculate the fitting degree index based on the EXCEL file of the monitoring data created in step S12 and the simulation value read in step S324. The calculation formula of the fitting degree index is: Among them,/> Indicates the goodness of fit,/> represents the i-th data, represents the Nash coefficient,/> Represents the i-th monitoring value,/> Represents the average of all monitoring values,/> Represents the i-th analog value,/> Represents the average value of simulated values;

S42. All simulation results in the initial population and sub-population are calculated according to the fitting index in step S41 above, and all fitness indicators are stored in the memory;

S43. Repeat the calculation process of step S41 for all SWAT simulation results of the initial population and sub-populations;

S44. According to the optimization algorithm and the number of iterations set in step S21, repeat steps S31 to S42, and screen out non-dominated solutions in the initial population and subpopulations;

S45. When all iteration calculations are completed, calculate all non-dominated solutions and obtain the Parteo optimal solution;

S5. Run the verified SWAT model;

The specific process of step S5 is:

S51. Compare the parameter tables of the SWAT model corresponding to all Parteo optimal solutions, and select solutions that meet the watershed simulation conditions as the final parameter set of the model;

S52. Substitute the final parameter set into the SWAT model to complete automatic model verification.

The above are only preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of changes or modifications within the technical scope disclosed in the present invention. All substitutions are within the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (2)

1. An automatic verification method based on SWAT model multi-objective optimization is characterized by comprising the following steps:

s1, initializing a verification program of a SWAT model, constructing a framework structure for coupling the SWAT model with a multi-objective optimization method, and under the framework, applying the multi-objective optimization method to the verification of the SWAT model, wherein the specific initialization process is as follows:

s11, copying an input/output folder of the constructed SWAT model from the ArcSWAT model;

s12, determining hydrologic water quality indexes for verification, and creating an EXCEL file of monitoring data of the corresponding indexes;

s13, exporting a position table in the SWAT model to a location_table folder;

the position table in the step S13 comprises a hru table, an rch table and a sub table;

s2, setting verification parameters, wherein the specific setting process is as follows:

s21, setting an optimization algorithm, the initial population number, the iteration number, the sub population number and the cross variation value of the optimization algorithm, and setting the process number;

s22, setting a parameter set range for verification and a parameter adjustment mode of each parameter in the model verification process, wherein the parameter adjustment mode comprises a relative change rate method, an absolute change rate method and a replacement rate method;

the calculation formula of the relative change calibration method is as follows:，

the calculation formula of the absolute change calibration method is as follows:，

the calculation formula of the replacement rating method is as follows:，

wherein,is a modified parameter value,/->Is the original value of the parameter in the SWAT model, x is the parameter variation factor;

s3, iteratively operating the SWAT model; step S3, performing iterative operation on the SWAT model according to the initial population quantity, the iterative times and the sub population quantity set in the step S21, wherein the specific process is as follows:

s31, acquiring parameter samples from the parameter set range in the step S22 by using a sampling method according to the initial population quantity or the sub population quantity, and generating a parameter table set;

s32, performing iterative operation on the SWAT model according to the number of processes set in the step S21;

the sampling method in the step S31 is random sampling, binomial random sampling or superpulling Ding Lifang body sampling; the specific process of step S32 is:

s321, corresponding the parameter table in the step S31 to the position table in the step S13, and generating a position parameter table to be modified;

s322, selecting corresponding files from the input/output folders according to the position parameter table, and storing the files in the memory;

s323, modifying files to be modified in the position parameter table by using different functions according to the attribute of each file, and importing the files which are modified and kept in the present state into an operation folder to perform model simulation;

s324, after the SWAT model in the operation folder is operated, firstly recovering the model file in the operation folder to the original state of the model according to the files screened in the step S322, then reading out the model result in the operation folder by using a reading file module, wherein the reading file comprises hydrological response unit data, sub-river basin data and river data, and storing the reading file, a parameter table and a position parameter table in an output folder together for carrying out fitting degree on calculation model simulation values and monitoring values with the monitoring data;

s4, evaluating the model by using an optimization algorithm;

the specific process of step S4 is:

s41, calculating a fitness index according to the EXCEL file of the monitoring data created in the step S12 and the simulation value read in the step S324, wherein the calculation formula of the fitness index is as follows: wherein (1)>Represents goodness of fit, ->The data of the i-th data is represented,representing Nash coefficient,/->Represents the ith monitored value,/-)>Mean value of all monitoring values, +.>Represents the i-th analog value,/, for>Representing an average value of the analog values;

s42, calculating all simulation results in the initial population and the sub population according to the fitting indexes of the step S41, and storing all fitting degree indexes in a memory;

s43, repeating the calculation process of the step S41 on all SWAT simulation results of the initial population and the sub population;

s44, repeating the steps S31-S42 according to the optimization algorithm and the iteration times set in the step S21, and screening non-dominant solutions from the initial population and the sub population;

s45, when all iteration times operation is finished, calculating all non-dominant solutions and obtaining a Parteo optimal solution;

s5, running the SWAT model for checking completion.

2. The automatic verification method based on the multi-objective optimization of the SWAT model as claimed in claim 1, wherein the specific process of the step S5 is as follows:

s51, comparing parameter tables of SWAT models corresponding to all Parteo optimal solutions, and screening solutions meeting the watershed simulation conditions as final parameter sets of the models;

s52, substituting the final parameter set into the SWAT model to complete automatic verification of the model.