River ecological health assessment method based on comprehensive indexes

Abstract

The invention discloses a river ecological health assessment method based on omnibearing indexes, and the key points of the technical scheme are as follows: through constructing a multi-dimensional assessment matrix containing a plurality of key indexes, the scheme can comprehensively cover a plurality of aspects of a river ecological system, such as water quality conditions, biocenosis structures, species diversity and the like, so that the comprehensiveness of assessment is ensured; the weight of each dimension is dynamically calculated and adjusted by using the grid search optimization and gradient descent algorithm, the subjectivity and one-sidedness of a traditional method are avoided, the scientificity of evaluation is improved, the accuracy and reliability of an evaluation result are ensured through verification of an independent data set, a powerful scientific support is provided for river ecological protection and sustainable development, and the method has a wide application prospect. According to the method, various data of the river ecosystem can be comprehensively and systematically collected and analyzed by constructing the matrix containing a plurality of key indexes, a solid data basis is provided for subsequent evaluation, and the accuracy of data used in the evaluation process is ensured through the data preprocessing step.

Description

Technical Field

[0001] The present invention relates to the technical field of river health assessment, and in particular to a river ecological health assessment method based on all-round indicators. Background Art

[0002] With the rapid development of industrialization and urbanization, rivers, as an important natural resource, not only support the survival and reproduction of aquatic organisms, but also provide humans with a variety of service functions such as drinking water, irrigation, and shipping. Therefore, it is particularly important to conduct scientific assessments of the health of river ecosystems.

[0003] Most of the current mainstream evaluation schemes are based on single-dimensional analysis models, such as the biological integrity index method, the comprehensive water quality identification index method, and the species diversity index method. Although these methods can reveal the ecological health of rivers and basins from specific dimensions such as water quality, community structure, and species diversity, they have the problems of low data application rate and one-sided evaluation dimensions. In order to overcome the limitations of single-dimensional evaluation, some pioneers in the industry have tried to combine the above methods to obtain a more comprehensive evaluation through expert experience or calculation of the average value of various index results. However, this method relies more on expert experience and is more subjective. It fails to make full use of indicators of each dimension objectively and scientifically. The existing mainstream evaluation schemes have integrated multiple evaluation perspectives to a certain extent, but the scientificity and accuracy of their results are still questionable. Most of these methods have the problem of "one size fits all", that is, they fail to fully consider the uniqueness and complexity of different river ecosystems, resulting in the evaluation results that may not be accurate and comprehensive enough. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a river ecological health assessment method based on comprehensive indicators, which solves the need for a method that can comprehensively consider multiple dimensions and scientifically calculate the weights of each dimension to achieve a comprehensive and accurate assessment of the river ecological health status. This method should be able to dynamically adjust the weight system to better adapt to the characteristics and changes of different river ecosystems.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] A river ecological health assessment method based on a full range of indicators, including:

[0007] (1) Construct a multidimensional evaluation index matrix: Based on the data accumulated from the field survey, a multidimensional evaluation index matrix is ​​constructed that includes multiple key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity, and river geological quality;

[0008] (2) Data preprocessing: The collected river ecosystem data are cleaned, outliers and missing values ​​are removed, and standardized to eliminate the dimensional differences between different indicators;

[0009] (3) Define the objective function: The sum of squared errors between the predicted result and the objective result is defined as the objective function. The mathematical expression of the objective function is:

[0010] (4) Initialization parameter set: Randomly initialize a coefficient parameter set containing weights and bias terms as the initial parameters of the model;

[0011] (5) Model training: Using the gradient descent algorithm, the objective function is minimized by iteratively updating the parameter set until the sum of squared errors reaches a preset threshold or the number of iterations reaches a maximum value;

[0012] (6) Solving the optimal coefficient parameter set: Through repeated iterations and calculations, an optimal coefficient parameter set is solved, which can best fit the inherent laws in the data set and reveal the complex relationship between various indicators of the river ecosystem;

[0013] (7) Model validation: Select an independent data set as the validation data set, apply the optimal coefficient parameter set to the matrix operation of the validation data set, and evaluate the prediction accuracy and generalization ability of the model;

[0014] (8) Output results: Use the optimal coefficient parameter set obtained to comprehensively evaluate the health status of the river ecosystem and output the evaluation results.

[0015] Preferably, the method for constructing the multidimensional evaluation index matrix includes but is not limited to quantitative analysis of key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, and constructing a matrix based on these quantitative analysis results. After constructing the multidimensional evaluation index matrix, a scoring mode based on weight factors is adopted. This mode takes into account multiple key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, and performs weighted processing according to the importance of each indicator. First, each indicator is quantitatively analyzed, and a multidimensional evaluation index matrix is ​​constructed based on these quantitative analysis results. The weight factor of each indicator is determined according to a grid search optimization algorithm that dynamically adjusts the weights of evaluation indicators in each dimension. After constructing the multidimensional evaluation index matrix, we use these weight factors to weight the score of each module to ensure that in subsequent model training, each indicator can affect the final result according to its importance. At the same time, the model fusion method adopts the multiplication fusion formula Here factor is a hyperparameter, which indicates the combination weight; β is a hyperparameter used to improve the ratio and nonlinear processing; αi It is also a hyperparameter, indicating sensitivity.

[0016]

[0017] Preferably, the data preprocessing step also includes normalizing the data to ensure that data of different dimensions are comparable in subsequent calculations.

[0018] Preferably, when the gradient descent algorithm iteratively updates the parameters, the weight and bias terms are dynamically adjusted according to the error gradient between the predicted result and the objective result. The parameter update formula of the gradient descent algorithm is: To minimize the objective function.

[0019] Preferably, the optimal coefficient parameter set includes weights and bias items corresponding to each evaluation indicator, and these parameters can fit the inherent laws in the historical data to the greatest extent and accurately reflect the complex relationship between various indicators of the river ecosystem.

[0020] Preferably, the validation dataset is two years of data independent of the training dataset, and is used to evaluate the prediction accuracy and generalization ability of the model on unknown data.

[0021] A river ecological health assessment system based on a full range of indicators, including:

[0022] (a) Data collection module, used to collect relevant data on river ecosystems;

[0023] (b) a data processing module, which is used to clean the collected data, remove outliers and missing values, and perform standardization;

[0024] (c) Model building module, used to construct a multi-dimensional evaluation index matrix and set the objective function;

[0025] (d) a parameter solving module, which is used to initialize parameters and iteratively update parameters through a gradient descent algorithm to solve for the optimal coefficient parameter set;

[0026] (e) The validation evaluation module is used to select an independent validation data set, apply the solved optimal coefficient parameter set to the matrix operation of the validation data set, and evaluate the prediction accuracy and generalization ability of the model.

[0027] Preferably, the model building module also includes a grid search optimization algorithm for dynamically adjusting the weights of evaluation indicators of each dimension. The mathematical representation of the grid search optimization algorithm can be simplified into the following steps:

[0028] S1: Define the parameter space, and set the hyperparameter set to be P = p1, p2, ..., p n , where p irepresents the i-th hyperparameter, which can be in the range of V i =v i1 ,v i2 ,…,v im .

[0029] S2: Exhaustive search: For each hyperparameter p i , traverse all possible values ​​v ij , generate all possible parameter combinations C = c1, c2, ..., c k , where c j Represents the jth parameter combination.

[0030] S3: Model training and evaluation: For each parameter combination C j , use the training data to train the model, and use the validation set to evaluate the performance of the model, and get the performance index S(c j ).

[0031] S4: Select the optimal parameter combination: Select the parameter combination with the best performance index Optimal parameter combination.

[0032] Preferably, in the defined parameter space, the hyperparameter space is set to Ω, which is a multidimensional space, and each dimension corresponds to a hyperparameter. For example, if we want to optimize two hyperparameters α and β, then Ω is the Cartesian product of the value ranges of α and β, that is, Ω = α × β.

[0033] Preferably, the parameter space is defined in the hyperparameter space Ω, and we define a parameter grid G, which is a subset of Ω and contains all hyperparameter combinations to be searched. For example, if the candidate values ​​of α are α1, α2, and the candidate values ​​of β are β1, β2, then the parameter grid G ​​= (α1, β1), (α1, β2), (α2, β1), (α2, β2).

[0034] Preferably, in the model training and evaluation, for each set of hyperparameter combinations (αi, βj) in the parameter grid G, we train a model Mi,j and evaluate its performance on the validation set. The evaluation indicator is usually a scalar value, such as accuracy, cross entropy loss, etc., denoted as f(Mi,j).

[0035] Preferably, in the selection of the optimal parameter combination, we select the hyperparameter combination that makes the evaluation index f optimal as the optimal parameter combination. That is, we look for (α, β)∈G such that f(M*,*)=min(f(Mi,j))|(αi,βj)∈G.

[0036] In summary, the present invention mainly has the following beneficial effects:

[0037] By constructing a multidimensional evaluation matrix containing multiple key indicators, the program can comprehensively cover multiple aspects of river ecosystems, such as water quality, biological community structure, species diversity, etc., to ensure the comprehensiveness of the evaluation. It uses grid search optimization and gradient descent algorithms to dynamically calculate and adjust the weights of each dimension, avoiding the subjectivity and one-sidedness of traditional methods, improving the scientific nature of the evaluation, and ensuring the accuracy and reliability of the evaluation results through verification by independent data sets, providing strong scientific support for river ecological protection and sustainable development.

[0038] By constructing a matrix containing multiple key indicators, various data of river ecosystems can be collected and analyzed comprehensively and systematically, providing a solid data foundation for subsequent evaluation. The multiplication model fusion method is used to effectively avoid the weakening of the importance of each sub-goal after the addition of new goals. Through the data preprocessing step, the accuracy and consistency of the data used in the evaluation process are ensured, thereby improving the reliability of the evaluation results. Through optimization methods such as the gradient descent algorithm, iteratively updating the parameter set to minimize the objective function is a key step in solving the optimal coefficient parameter set, so that the evaluation method can adaptively adjust the weights to better reflect the actual situation of the river ecosystem. By introducing a verification unit and using an independent data set to verify the evaluation results, it can ensure that the evaluation process is not affected by data bias or model overfitting. This verification mechanism not only helps to discover possible problems in the evaluation process, but also can further adjust and optimize the evaluation model by comparing the verification results with the expected results, thereby improving the scientificity and accuracy of the evaluation results. By encoding the evaluation method as a computer program and storing it on a readable storage medium, users can easily obtain and use this tool without having to deeply understand the internal details of the evaluation method, which not only improves the accessibility and ease of use of the evaluation method, but also promotes the widespread dissemination and application of the evaluation method. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a data flow diagram of the present invention;

[0040] Figure 2 is a system block diagram of the present invention;

[0041] Figure 3 It is a flowchart of the algorithm of the present invention;

[0042] Figure 4 It is a flowchart of the verification experiment of the present invention. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] refer to Figure 1-Figure 4 , a river ecological health assessment method based on a full range of indicators, including:

[0045] (1) Construct a multidimensional evaluation index matrix: Based on the data accumulated from the field survey, a multidimensional evaluation index matrix is ​​constructed that includes multiple key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity, and river geological quality;

[0046] (2) Data preprocessing: The collected river ecosystem data are cleaned, outliers and missing values ​​are removed, and standardized to eliminate the dimensional differences between different indicators;

[0047] (3) Define the objective function: The sum of squared errors between the predicted result and the objective result is defined as the objective function. The mathematical expression of the objective function is:

[0048] (4) Initialization parameter set: Randomly initialize a coefficient parameter set containing weights and bias terms as the initial parameters of the model;

[0049] (5) Model training: Using the gradient descent algorithm, the objective function is minimized by iteratively updating the parameter set until the sum of squared errors reaches a preset threshold or the number of iterations reaches a maximum value;

[0050] (6) Solving the optimal coefficient parameter set: Through repeated iterations and calculations, an optimal coefficient parameter set is solved, which can best fit the inherent laws in the data set and reveal the complex relationship between various indicators of the river ecosystem;

[0051] (7) Model validation: Select an independent data set as the validation data set, apply the optimal coefficient parameter set to the matrix operation of the validation data set, and evaluate the prediction accuracy and generalization ability of the model;

[0052] (8) Output results: Use the optimal coefficient parameter set obtained to comprehensively evaluate the health status of the river ecosystem and output the evaluation results.

[0053] The method for constructing the multidimensional evaluation index matrix includes but is not limited to quantitative analysis of key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, and constructing a matrix based on these quantitative analysis results. After constructing the multidimensional evaluation index matrix, a scoring mode based on weight factors is adopted. This mode takes into account multiple key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, and performs weighted processing according to the importance of each indicator. First, each indicator is quantitatively analyzed, and a multidimensional evaluation index matrix is ​​constructed based on these quantitative analysis results. The weight factor of each indicator is determined according to the grid search optimization algorithm that dynamically adjusts the weights of the evaluation indicators of each dimension. After constructing the multidimensional evaluation index matrix, we use these weight factors to weight the score of each module to ensure that in the subsequent model training, each indicator can affect the final result according to its importance. At the same time, the model fusion method adopts the multiplication fusion formula Here factor is a hyperparameter, which indicates the combination weight; β is a hyperparameter used to improve the ratio and nonlinear processing; α i It is also a hyperparameter, indicating sensitivity. The data preprocessing step also includes normalization of the data to ensure that data of different dimensions are comparable in subsequent calculations. When the gradient descent algorithm iteratively updates the parameters, it dynamically adjusts the weights and bias terms according to the error gradient between the predicted results and the objective results. The parameter update formula of the gradient descent algorithm is To minimize the objective function, the optimal coefficient parameter set includes the weights and bias items corresponding to each evaluation indicator. These parameters can fit the inherent laws in the historical data to the greatest extent and accurately reflect the complex relationship between various indicators of the river ecosystem. The validation data set is two years of data independent of the training data set, which is used to evaluate the prediction accuracy and generalization ability of the model on unknown data.

[0054] refer to Figure 1-Figure 4 , a river ecological health assessment system based on a full range of indicators, including:

[0055] (a) Data collection module, used to collect relevant data on river ecosystems;

[0056] (b) a data processing module, which is used to clean the collected data, remove outliers and missing values, and perform standardization;

[0057] (c) Model building module, used to construct a multi-dimensional evaluation index matrix and set the objective function;

[0058] (d) a parameter solving module, which is used to initialize parameters and iteratively update parameters through a gradient descent algorithm to solve for the optimal coefficient parameter set;

[0059] (e) The validation evaluation module is used to select an independent validation data set, apply the solved optimal coefficient parameter set to the matrix operation of the validation data set, and evaluate the prediction accuracy and generalization ability of the model.

[0060] The model building module also includes a grid search optimization algorithm for dynamically adjusting the weights of evaluation indicators of each dimension. The mathematical representation of the grid search optimization algorithm can be simplified into the following steps:

[0061] S1: Define the parameter space, and set the hyperparameter set to be P = p1, p2, ..., p n , where p i represents the i-th hyperparameter, which can be in the range of V i =v i1 ,v i2 ,…,v im .

[0062] S2: Exhaustive search: For each hyperparameter p i , traverse all possible values ​​v ij , generate all possible parameter combinations C = c1, c2, ..., c k , where c j Represents the jth parameter combination.

[0063] S3: Model training and evaluation: For each parameter combination C j , use the training data to train the model, and use the validation set to evaluate the performance of the model, and get the performance index S(c j ).

[0064] S4: Select the optimal parameter combination: Select the parameter combination with the best performance index Optimal parameter combination.

[0065] The parameter space is defined in the hyperparameter space Ω. We define a parameter grid G, which is a subset of Ω and contains all the hyperparameter combinations to be searched. For example, if the candidate values ​​of α are α1, α2, and the candidate values ​​of β are β1, β2, then the parameter grid G ​​= (α1, β1), (α1, β2), (α2, β1), (α2, β2). In the model training and evaluation, for each set of hyperparameter combinations (αi, βj) in the parameter grid G, we train a model Mi,j and evaluate its performance on the validation set. The evaluation index is usually a scalar value, such as accuracy, cross entropy loss, etc., denoted as f(Mi,j). In the selection of the optimal parameter combination, we select the hyperparameter combination that makes the evaluation index f optimal as the optimal parameter combination. That is, find (α, β)∈G so that f(M*,*)=min(f(Mi,j))||(αi, βj)∈G.

[0066] Working principle: Please refer to Figure 1-Figure 4 As shown, when using it, first, collect relevant data of river ecosystems, including indicators of water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, clean the collected data, remove outliers and missing values, and perform standardization to eliminate the dimensional differences between different indicators and improve the accuracy of subsequent analysis. Based on the preprocessed data, construct a multi-dimensional evaluation index matrix. Each row of the matrix represents the observation data at a time point, and each column represents an evaluation index, so as to comprehensively reflect the health status of the river ecosystem. The sum of squares of the error between the predicted results and the objective results is defined as the objective function. As the optimization goal, the gradient descent algorithm is selected as the optimization algorithm, and the parameters are iteratively updated to minimize The objective function is optimized, a coefficient parameter set including weights and bias terms is initialized as the initial parameters, the evaluation index matrix is ​​linearly transformed using the current parameter set to obtain a prediction result vector, the error between the prediction result and the objective result is calculated, and the parameter set is updated according to the gradient of the error. The iterative process is repeated until the sum of squared errors reaches a preset threshold or the number of iterations reaches a maximum value, thereby solving an optimal coefficient parameter set. An independent validation data set is selected, and the optimal coefficient parameter set obtained is applied to the matrix operation of the validation data set to obtain a prediction vector. The consistency between the prediction vector and the objective result is compared, the prediction accuracy and generalization ability of the model are evaluated, and the verified model is applied to the actual river ecological health assessment, and corresponding ecological protection and management decisions are made according to the assessment results.

[0067] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A river ecological health assessment method based on comprehensive indicators, characterized in that: include: (1) Construct a multidimensional evaluation index matrix: Based on the data accumulated from the field survey, a multidimensional evaluation index matrix is ​​constructed that includes multiple key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity, and river geological quality; (2) Data preprocessing: The collected river ecosystem data were cleaned through online search, outliers and missing values ​​were removed, and standardized to eliminate the dimensional differences between different indicators; (3) Define the objective function: The sum of squared errors between the predicted result and the objective result is defined as the objective function. The mathematical expression of the objective function is: (4) Initialization parameter set: Randomly initialize a coefficient parameter set containing weights and bias terms as the initial parameters of the model; (5) Model training: Using the gradient descent algorithm, the objective function is minimized by iteratively updating the parameter set until the sum of squared errors reaches a preset threshold or the number of iterations reaches a maximum value; (6) Solving the optimal coefficient parameter set: Through repeated iterations and calculations, an optimal coefficient parameter set is solved, which can best fit the inherent laws in the data set and reveal the complex relationship between various indicators of the river ecosystem; (7) Model validation: Select an independent data set as the validation data set, apply the optimal coefficient parameter set to the matrix operation of the validation data set, and evaluate the prediction accuracy and generalization ability of the model; (8) Output results: Use the optimal coefficient parameter set obtained to comprehensively evaluate the health status of the river ecosystem and output the evaluation results.

2. A river ecological health assessment method based on all-round indicators according to claim 1, characterized in that: The method for constructing the multidimensional evaluation index matrix includes but is not limited to quantitative analysis of key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, and constructing a matrix based on these quantitative analysis results. After constructing the multidimensional evaluation index matrix, a scoring mode based on weight factors is adopted. This mode takes into account multiple key indicators such as water quality, biological community structure, species diversity, wetland integrity, water salinity and river geological quality, and performs weighted processing according to the importance of each indicator. First, each indicator is quantitatively analyzed, and a multidimensional evaluation index matrix is ​​constructed based on these quantitative analysis results. The weight factor of each indicator is determined according to the grid search optimization algorithm that dynamically adjusts the weights of the evaluation indicators of each dimension. After constructing the multidimensional evaluation index matrix, we use these weight factors to weight the score of each module to ensure that in the subsequent model training, each indicator can affect the final result according to its importance. At the same time, the model fusion method adopts the multiplication fusion formula Here factor is a hyperparameter, indicating the combination weight; β is a hyperparameter used to improve the ratio and nonlinear processing; α i It is also a hyperparameter, indicating sensitivity.

3. A river ecological health assessment method based on all-round indicators according to claim 1, characterized in that: The data preprocessing step also includes normalizing the data to ensure that data of different dimensions are comparable in subsequent calculations.

4. A river ecological health assessment method based on all-round indicators according to claim 1, characterized in that: The gradient descent algorithm dynamically adjusts the weights and bias terms according to the error gradient between the predicted result and the objective result when iteratively updating the parameters. The parameter update formula of the gradient descent algorithm is: To minimize the objective function.

5. A river ecological health assessment method based on all-round indicators according to claim 1, characterized in that: The optimal coefficient parameter set includes weights and bias items corresponding to each evaluation indicator. These parameters can fit the inherent laws in historical data to the greatest extent and accurately reflect the complex relationship between various indicators of the river ecosystem.

6. A river ecological health assessment method based on all-round indicators according to claim 1, characterized in that: The validation dataset is two years of data independent of the training dataset and is used to evaluate the prediction accuracy and generalization ability of the model on unknown data.

7. A river ecological health assessment system based on all-round indicators, characterized in that: include: (a) Data collection module, used to collect relevant data on river ecosystems; (b) a data processing module, which is used to clean the collected data, remove outliers and missing values, and perform standardization; (c) Model building module, used to construct a multi-dimensional evaluation index matrix and set the objective function; (d) a parameter solving module, which is used to initialize parameters and iteratively update parameters through a gradient descent algorithm to solve for the optimal coefficient parameter set; (e) The validation evaluation module is used to select an independent validation data set, apply the solved optimal coefficient parameter set to the matrix operation of the validation data set, and evaluate the prediction accuracy and generalization ability of the model.

8. A river ecological health assessment system based on all-round indicators according to claim 7, characterized in that: The model building module also includes a grid search optimization algorithm for dynamically adjusting the weights of evaluation indicators of each dimension. The mathematical representation of the grid search optimization algorithm can be simplified into the following steps: S1: Define the parameter space, and set the hyperparameter set to be P = p1, p2, ..., p n , where p i represents the i-th hyperparameter, which can be in the range of V i =v i1 ,v i2 ,…,v im . S2: Exhaustive search: For each hyperparameter p i , traverse all possible values ​​v ij , generate all possible parameter combinations C = c1, c2, ..., c k , where c j Represents the jth parameter combination. S3: Model training and evaluation: For each parameter combination C j , use the training data to train the model, and use the validation set to evaluate the performance of the model, and get the performance index S(c j ). S4: Select the optimal parameter combination: Select the parameter combination with the best performance index Optimal parameter combination.

9. The grid search optimization algorithm according to claim 8, characterized in that: In the parameter space definition, the hyperparameter space is Ω, which is a multidimensional space, and each dimension corresponds to a hyperparameter. For example, if we want to optimize two hyperparameters α and β, then Ω is the Cartesian product of the value ranges of α and β, that is, Ω = α × β.

10. The grid search optimization algorithm according to claim 7, characterized in that: Defining parameter space In the hyperparameter space Ω, we define a parameter grid G, which is a subset of Ω and contains all the hyperparameter combinations to be searched. For example, if the candidate values ​​of α are α1, α2, and the candidate values ​​of β are β1, β2, then the parameter grid G ​​= (α1, β1), (α1, β2), (α2, β1), (α2, β2).

11. The grid search optimization algorithm according to claim 7, characterized in that: In the model training and evaluation, for each set of hyperparameter combinations (αi, βj) in the parameter grid G, we train a model Mi,j and evaluate its performance on the validation set. The evaluation indicator is usually a scalar value, such as accuracy, cross entropy loss, etc., denoted as f(Mi,j).

12. The grid search optimization algorithm according to claim 7, characterized in that: In the above selection of the optimal parameter combination, we select the hyperparameter combination that makes the evaluation index f optimal as the optimal parameter combination. That is, we look for (α, β)∈G such that f(M*,*)=min(f(Mi,j))||(αi,βj)∈G.

Images