Automatic model parameter repairing method and system based on Spec standard
By combining ridge regression and dichotomous search methods, the problem of difficult to smooth the optimized model parameter trend in the prior art is solved, and the dual optimization of parameter stability and standard compliance is achieved.
Patent Information
- Application Number
- CN202411951131.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-04-15
AI Technical Summary
The existing technology is difficult to smoothly optimized model parameter trends under the premise of meeting the specification Spec requirements, resulting in unstable parameters and affecting system reliability.
The automatic repair method of model parameters based on Spec standard is adopted, combined with ridge regression and dichotomy search, and post-processing is performed to smooth the parameter trend. Ridge regression eliminates local fluctuations, and dichotomous search adjusts parameters to meet the specification Spec.
Effectively eliminate the violent fluctuations and unstable trends that may occur during the optimization process of parameters, ensure that the parameters are both smooth and meet engineering specifications, and improve the effect of parameter repair and the stability of the system.
Smart Images
Figure CN119831072A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information technology, and in particular to a method and system for automatically repairing model parameters based on Spec standards. Background Art
[0002] In many machine learning and control system modeling and parameter extraction, model parameters (such as a0, a1, vth0, etc.) and their trends are one of the key indicators for evaluating system performance and delivery. The changing trend of these parameters usually needs to be kept smooth to meet the physical meaning and stability requirements of the system. However, in practical applications, the optimization process often faces two major challenges: one is how to accurately control the target accuracy, and the other is how to optimize the smoothed parameter trend to avoid instability or excessive fluctuations.
[0003] Although optimization algorithms usually find the optimal solution on the objective function and generate a set of optimized parameters. However, these optimized parameters often have certain trend instability problems, such as Figure 1 As shown, it manifests as rapid fluctuations or mutations in local values. This instability may affect the reliability of the system and even lead to wrong decisions in practical applications, such as in industrial control, autonomous driving or intelligent recognition systems.
[0004] In order to ensure the stability of the system, it is usually necessary to reduce parameter fluctuations through smoothing. Existing technologies mainly focus on adjusting the objective function during the optimization process, or smoothing parameters through post-processing techniques such as filtering and regularization. However, these methods have the following defects: ① Although the gradient descent method is often used for optimization, the gradient descent method usually focuses on the local optimal solution and cannot effectively smooth the global parameter trend, which easily leads to parameter fluctuations; ② Regularization method: Regularization (such as L1 / L2) reduces fluctuations by adding penalty terms, but it is not specifically optimized for the smoothness of parameter trends and cannot fully meet physical and application requirements. Therefore, existing technologies usually find it difficult to smooth the optimized parameter trends while meeting the requirements of the specification. Summary of the invention
[0005] In view of the above problems, the present invention proposes a method and system for automatically repairing model parameters based on the Spec standard, which performs post-processing after model parameter optimization, combines ridge regression with binary search, effectively smooths parameter trends and ensures that they meet the requirements of the engineering specification Spec. The above invention object of the present invention is achieved through the following technical solutions: The present invention provides a method for automatically repairing model parameters based on the Spec standard, comprising the following steps: Step S100: receiving model parameters obtained by optimization algorithm ; Step S200: Based on model parameters , design feature construction to capture the changing trend of model parameters, and fit the features through the ridge regression model to calculate the smoothed parameter values ; Step S300: Set the initial interval of binary search to , and set the preset search precision As a termination condition; Step S400: Execute the binary search to iteratively search and adjust the upper and lower limits of the initial interval until the termination condition is met. Or when the iteration number threshold is reached, the smoothing parameter is returned. As the final result.
[0006] Further, step S200 includes: Step S210: Based on model parameters , construct the feature vector for the input of the ridge regression model , used to capture the changing trend of model parameters, feature construction includes polynomial features, nonlinear transformation features or differential features; Step S220: Transform the feature vector Input into the ridge regression model and solve the smoothed model parameters by minimizing the ridge regression objective function formula .
[0007] Further, in step S210, Polynomial features include quadratic or higher-degree polynomials; Nonlinear transformation features include and logarithmic transformation; Cross-features involve generating new features by combining cross-terms between different parameters.
[0008] Further, in step S220, the ridge regression objective function formula is, ; in, is the target value, is the input feature vector, are model parameters, is the sample size, is the L2 norm, which is the sum of squares of the model parameters and is used to control the complexity of the model. is a regularization hyperparameter that controls the strength of regularization.
[0009] Further, in step S220, the smoothed model parameters for, ; Among them, X is the N*M input matrix, N is the number of samples, and M is the feature dimension; y is an N*1 column vector, which is the target value of each sample. is the regularization term, is the identity matrix, is the regularization hyperparameter.
[0010] Further, step S400 includes: Step S410: Initial interval [ , The upper and lower limits of ] are set as the parameter values in step S200 and the model parameters in step S100 ; Step S420: Execute the binary search to iterate the interval to adjust the upper and lower limits of the initial interval, wherein in each iteration, the middle value is calculated. ; and check whether the intermediate value meets the engineering specification Spec. If the engineering specification Spec is met, the upper limit is updated for , otherwise update the lower limit for ; until Or the number of iterations reaches the threshold, and the search stops; Step S430: When the binary search is completed, the current middle value is returned As the smoothed parameter , and meet the specification Spec.
[0011] Based on the same inventive concept, the present invention provides a model parameter automatic repair system based on the Spec standard, which adopts the model parameter automatic repair method as described above, including: Response module, used to receive the model parameters obtained by the optimization algorithm ; Smoothing module, based on model parameters , design feature construction to capture the changing trend of model parameters, and fit the features through the ridge regression model to calculate the smoothed parameter values ; Optimize the processing module and set the initial interval of binary search to , and set the preset search precision As the termination condition; perform the binary search in the interval to adjust the upper and lower limits of the initial interval until the termination condition is met Or when the iteration number threshold is reached, the smoothing parameter is returned. As the final result.
[0012] Furthermore, the smoothing processing module includes: Feature construction unit for model parameters , construct the feature vector for the input of the ridge regression model , used to capture the changing trend of model parameters, feature construction includes polynomial features, nonlinear transformation features or differential features; Regression fitting unit, used to transform the feature vector Input into the ridge regression model and solve the smoothed model parameters by minimizing the ridge regression objective function formula .
[0013] Furthermore, in the regression fitting unit, the smoothed model parameters for, ; Among them, X is the N*M input matrix, N is the number of samples, and M is the feature dimension; y is an N*1 column vector, which is the target value of each sample. is the regularization term, is the identity matrix, is the regularization hyperparameter.
[0014] Furthermore, the optimization processing module includes: The interval initialization unit is used to initialize the initial interval [ , ] is set to the parameter value and model parameters ; The binary search execution unit is used to perform binary search in the interval to adjust the upper and lower limits of the initial interval, wherein in each iteration, the middle value is calculated ; and check whether the intermediate value meets the engineering specification Spec. If the engineering specification Spec is met, the upper limit is updated for , otherwise update the lower limit for ; until Or the number of iterations reaches the threshold, and the search stops; The result output unit is used to return the current intermediate value after the binary search is completed. As the smoothed parameter , and meet the specification Spec.
[0015] Compared with the prior art, the present invention has at least one of the following beneficial effects: The present invention performs post-processing after model parameter optimization, combines ridge regression with binary search, and effectively smoothes parameter trends and ensures that they meet the requirements of engineering specifications Spec. Specifically, the optimized parameters are smoothed by ridge regression to effectively eliminate the violent fluctuations and unstable trends that may occur in the parameters during the optimization process. At the same time, without manual intervention, the optimized parameters are automatically adjusted to achieve smoothness while meeting the predetermined specifications (spec). Finally, ridge regression is combined with binary search to make full use of the smoothing effect of ridge regression and the precise adjustment ability of binary search, thereby achieving dual optimization of parameter smoothing and specification compliance. This combined method not only improves the effect of parameter repair, but also ensures that the final parameters achieve the best balance between smoothness and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 It is a diagram of rapid fluctuation or change of local values in this embodiment; Figure 2 The steps of the automatic repair method of model parameters based on the Spec standard of the present invention are as follows Figure 1 ; Figure 3 The steps of the automatic repair method of model parameters based on the Spec standard of the present invention are as follows Figure 2 ; Figure 4 is the initial kop and the parameter changes after optimization in this embodiment Figure 1 ; Figure 5 This is the smoothed kop and parameter changes after adjusting the spec value in this embodiment Figure 1 ; Figure 6 This is the smoothed kop and parameter changes after adjusting the spec value in this embodiment Figure 2 . DETAILED DESCRIPTION
[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0018] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.
[0019] First embodiment In the process of model building and parameter optimization, the trend change of parameters is one of the key performance indicators in system delivery. For example, parameters such as a0, a1, vth0, etc. usually need to have a smooth trend, which not only ensures the physical rationality of the model, but also ensures the stability of the system. However, in practical applications, these optimized parameters often show unstable trends, such as local fluctuations or mutations, such as Figure 1 Especially in complex systems or engineering applications, drastic fluctuations in parameters may have an adverse effect on subsequent processes. Therefore, how to smooth these optimized parameters to ensure that they meet the engineering specifications without losing optimization accuracy becomes a major challenge.
[0020] At present, most commonly used solutions focus on post-processing during or after the optimization process to improve the smoothness of parameters. Traditional technical methods include but are not limited to gradient descent, regularization-based optimization methods (such as L1 regularization, L2 regularization), and smoothing techniques such as low-pass filtering and moving average. Although these methods can optimize parameters or reduce fluctuations to a certain extent, they usually have the following defects: (1) Gradient descent is a classic optimization algorithm that is widely used in objective function optimization. It gradually adjusts parameters to minimize the loss function by calculating the gradient of the objective function. However, since our parameter settings are independent of each other in the objective function, the optimized parameters often only focus on the local optimal point, so it is impossible to control the smoothness of the parameter trend.
[0021] (2) Regularization methods (such as L1 and L2 regularization) introduce additional penalty terms in the optimization objective function to limit the fluctuation range or sparsity of the parameters, thereby suppressing overfitting. These methods have a certain effect in reducing the amplitude of parameter fluctuations, but they still have shortcomings in controlling the smoothness of parameter trends. For example, they are highly dependent on the initial values of the parameters before optimization, and their effects are greatly affected by the regularization coefficient. Improper selection may lead to poor optimization results.
[0022] (3) Post-processing smoothing methods: There are also some methods that perform post-processing to smooth parameters after the optimization is completed, such as low-pass filtering, moving average and other techniques. However, since these methods cannot take into account the specification requirements at the same time, they may cause the optimization target to deviate and the optimization effect to be poor.
[0023] Based on the existing technical problems, the inventor innovatively proposed a parameter automatic repair method based on spec constraints, which realizes the smoothing of optimized parameters by combining Ridge Regression and Binary Search. This method overcomes the shortcomings of the existing technology through the following two-step optimization strategy: (1) First, the ridge regression model is used to smooth the optimized parameters to eliminate local fluctuations and ensure a smooth transition of parameter trends. Ridge regression introduces regularization terms to make parameter changes more stable, thereby reducing the instability problem caused by the gradient descent method.
[0024] (2) Based on ridge regression smoothing, the binary search method is used to further adjust the optimized parameters to ensure that they meet the engineering specifications. By iteratively searching within the initial interval and gradually approaching the smoothing parameters that meet the specifications, the binary search method can achieve the optimal balance between smoothness and optimization accuracy.
[0025] The present invention provides a method for automatically repairing model parameters based on Spec standards, such as Figure 2 As shown, the following steps are included: Step S100: receiving model parameters obtained by optimization algorithm ; Step S200: Based on model parameters , design feature construction to capture the changing trend of model parameters, and fit the features through the ridge regression model to calculate the smoothed parameter values ; It should be specifically noted that in the smoothing implementation of the W / L (width / length) trend parameter, the Ridge Regression method is used because the W / L trend parameter usually has a certain regularity, but is also easily disturbed by noise. Through ridge regression, local fluctuations can be effectively removed, making the optimized parameter trend smoother, thus meeting physical meaning and engineering requirements. It is worth noting that the implementation method of the smoothing part is not limited to ridge regression. In other application scenarios, we can replace it with other smoothing algorithms as needed, as long as it can output smoothing parameters that meet the requirements. ; Step S300: Set the initial interval of binary search to , and set the preset search precision As a termination condition; Step S400: Execute the binary search to iteratively search and adjust the upper and lower limits of the initial interval until the termination condition is met. Or when the iteration number threshold is reached, the smoothing parameter is returned. As the final result.
[0026] Further, step S200 includes: Step S210: Based on model parameters , construct the feature vector for the input of the ridge regression model , used to capture the changing trend of model parameters, feature construction includes polynomial features, nonlinear transformation features or differential features; Step S220: Transform the feature vector Input into the ridge regression model and solve the smoothed model parameters by minimizing the ridge regression objective function formula .
[0027] Further, in step S210, Polynomial features include quadratic or higher-degree polynomials; Nonlinear transformation features include and logarithmic transformation; Cross-features involve generating new features by combining cross-terms between different parameters.
[0028] Further, in step S220, the ridge regression objective function formula is, ; in, is the target value, is the input feature vector, are model parameters, is the sample size, is the L2 norm, which is the sum of squares of the model parameters and is used to control the complexity of the model. is a regularization hyperparameter that controls the strength of regularization.
[0029] Further, in step S220, the smoothed model parameters for, ; Among them, X is the N*M input matrix, N is the number of samples, and M is the feature dimension; y is an N*1 column vector, which is the target value of each sample. is the regularization term, is the identity matrix, is the regularization hyperparameter.
[0030] Further, step S400 includes: Step S410: Initial interval [ , ] is set to the parameter value and model parameters ; Step S420: Execute the binary search to iterate the interval to adjust the upper and lower limits of the initial interval, wherein in each iteration, the middle value is calculated. ; and check whether the intermediate value meets the engineering specification Spec. If the engineering specification Spec is met, the upper limit is updated for , otherwise update the lower limit for ; until Or the number of iterations reaches the threshold, and the search stops; Step S430: When the binary search is completed, the current middle value is returned As the smoothed parameter , and meet the specification Spec.
[0031] The specific code is: def binary_search(w_opt, w_smooth, epsilon): # Initialization w_low = w_smooth w_high = w_opt # Binary search while norm(w_high - w_low) > epsilon: w_mid = 0.5 * (w_high + w_low) if check_rule(w_mid): # Check if the current value meets the specification # If the specification is met, update w_high w_high = w_mid else: # If the specification is not met, update w_low w_low = w_mid # Search ends, return the final parameter value w_search = w_high return w_search It should be noted that in the rule checking phase, in order to ensure that the search results are compatible with the optimized parameters For situations where the specified spec may not be fully met, two criteria are designed: ①check_rule(w_{search}) = 0, indicating the current search result Fully comply with regulatory requirements; ②check_rule(w_{search}) \leq check_rule(w_{opt}): means even if Not fully compliant as long as it is no worse than existing parameters Through this flexible rule judgment, the algorithm can tolerate some results that do not fully meet the specifications, while ensuring that the search process can optimize parameters and improve accuracy and stability.
[0032] Finally, when the maximum number of objective function calls and iteration thresholds are not set, the time complexity of the algorithm is O(log2(1 / ϵ))×obj_func_time, where ϵ is the search accuracy and obj_func_time is the time required to calculate the objective function; if the maximum number of objective function calls max_func_calls is set, the time complexity of the algorithm is: min(O(log2(1 / ϵ))×obj_func_time, max_func_calls × obj_func_time), where the time complexity is determined by the search accuracy requirement and the objective function calculation time. By setting the maximum number of objective function calls, the running time of the algorithm can be limited when a quick response is required to avoid excessive iterations.
[0033] Combination Figure 3 From the perspective of the invention, the steps of the present invention include the following: Step 1: Receive optimized model parameters As parameter input, it provides basic data for subsequent data smoothing and searching; Step 2: Design feature structures based on experience, such as quadratic polynomials or 1 / x, and use ridge regression to fit the trend of the optimization parameters, and calculate the parameter values after regression smoothing through parameter smoothing algorithms : ; get The analytical solution is: ; Among them, X is an N*M matrix, N is the number of samples, M is the feature dimension, I is the M*M identity matrix, and y is an N-dimensional vector.
[0034] Step 3: Initialization = , = , calculate the objective function through simulation, and determine whether the termination condition is currently met. The termination condition of binary search requires that , where ε is the search accuracy of the preset value.
[0035] Step 4: Set the search range to [ , ] = [ , ], and set the search precision to ε, iteratively narrow the search interval through bisection bidirectional search, and finally find the smoothing parameter that meets the specification spec; calculate the intermediate value in each iteration , and check whether it meets the specification. If it meets the specification, update the upper limit; otherwise, update the lower limit and continue iterating.
[0036] Step 5: For better smoothing effect, you can set the maximum number of iterations. If the current number of iterations does not reach the maximum value or , then return to Step 2 to continue smoothing calculation; if the maximum number of iterations is reached or , then the final smoothing parameter is output.
[0037] The technical effect of the present invention is mainly reflected in the problem of unstable parameter trend after smooth optimization. In actual application scenarios, the present invention can support multi-parameter, multi-objective function, multi-rule, and two-dimensional WL value optimization. The initial kop and optimized parameters are as follows Figure 4 As shown in the figure, set spec = 0.01. In this case, since spec is large, the parameters after smoothing also meet the rule, so the search ends quickly. The smoothed kop and parameters are as follows Figure 5 As shown in the figure, after regression smoothing, the obtained parameters have met the rule, that is, the smoothed parameters have met the set accuracy requirements. Since the specification requirements are loose, the smoothed parameters have met the standards, so no further binary search is required and the search ends quickly. Set spec=0.005. In this case, since the spec is small, the smoothed parameters do not meet the rule and a binary search is required. The smoothed kop and parameters are as follows Figure 6 As shown, Second embodiment Based on the same inventive concept, the present invention provides a model parameter automatic repair system based on the Spec standard, which adopts the model parameter automatic repair method as described above, including: Response module, used to receive the model parameters obtained by the optimization algorithm ; Smoothing module, based on model parameters , design feature construction to capture the changing trend of model parameters, and fit the features through the ridge regression model to calculate the smoothed parameter values ; Optimize the processing module and set the initial interval of binary search to , and set the preset search precision As the termination condition; perform the binary search in the interval to adjust the upper and lower limits of the initial interval until the termination condition is met Or when the iteration number threshold is reached, the smoothing parameter is returned. As the final result.
[0038] Furthermore, the smoothing processing module includes: Feature construction unit for model parameters , construct the feature vector for the input of the ridge regression model , used to capture the changing trend of model parameters, feature construction includes polynomial features, nonlinear transformation features or differential features; Regression fitting unit, used to transform the feature vector Input into the ridge regression model and solve the smoothed model parameters by minimizing the ridge regression objective function formula .
[0039] Furthermore, in the regression fitting unit, the smoothed model parameters for, ; Among them, X is the N*M input matrix, N is the number of samples, and M is the feature dimension; y is an N*1 column vector, which is the target value of each sample. is the regularization term, is the identity matrix, is the regularization hyperparameter.
[0040] Furthermore, the optimization processing module includes: The interval initialization unit is used to initialize the initial interval [ , ] is set to the parameter value Model parameters ; The binary search execution unit is used to perform binary search in the interval to adjust the upper and lower limits of the initial interval, wherein in each iteration, the middle value is calculated ; and check whether the intermediate value meets the engineering specification Spec. If the engineering specification Spec is met, the upper limit is updated for , otherwise update the lower limit for ; until Or the number of iterations reaches the threshold, and the search stops; The result output unit is used to return the current intermediate value after the binary search is completed. As the smoothed parameter , and meet the specification Spec.
[0041] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
[0042] It should be noted that the above embodiments can be freely combined as needed. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention.
Claims
1. A method for automatically repairing model parameters based on Spec standard, characterized in that: The following steps are involved: Step S100: receiving model parameters obtained by optimization algorithm ; Step S200: Based on the model parameters , design feature construction to capture the changing trend of the model parameters, fit the features through the ridge regression model, and calculate the smoothed parameter values ; Step S300: Set the initial interval of binary search to , and set the preset search precision As a termination condition; Step S400: Execute the binary search to iteratively search and adjust the upper and lower limits of the initial interval until the termination condition is met. Or when the number of iterations reaches the threshold, the smoothing parameter is returned. As the final result.
2. The automatic repair method for model parameters according to claim 1, characterized in that: The step S200 includes: Step S210: Based on the model parameters , construct the feature vector used as the input feature vector of the ridge regression model , used to capture the changing trend of the model parameters, the feature construction includes polynomial features, nonlinear transformation features or differential features; Step S220: The feature vector Input into the ridge regression model, and solve the smoothed model parameters by minimizing the ridge regression objective function formula .
3. The automatic repair method for model parameters according to claim 1, characterized in that: In step S210, The polynomial features include quadratic or higher-order polynomials; The nonlinear transformation features include and logarithmic transformation; The cross-features include generating new features by combining cross-terms between different parameters.
4. The automatic repair method for model parameters according to claim 2, characterized in that: In step S220, the ridge regression objective function formula is: ； in, is the target value, is the input feature vector, are the model parameters, is the sample size, is the L2 norm, i.e., the sum of squares of the model parameters, used to control the complexity of the model, is a regularization hyperparameter that controls the strength of regularization.
5. The automatic repair method for model parameters according to claim 3, characterized in that: In step S220, the smoothed model parameters for, ； Where X is an N*M input matrix, N is the number of samples, and M is the feature dimension; y is an N*1 column vector, which is the target value for each sample. is the regularization term, is the identity matrix, is the regularization hyperparameter.
6. The automatic repair method for model parameters according to claim 1, characterized in that: The step S400 includes: Step S410: The initial interval [ , The upper and lower limits of ] are set to the parameter values in step S200 and the model parameters in step S100 ; Step S420: Execute the binary search to iteratively search and adjust the upper and lower limits of the initial interval in the interval, wherein, in each iteration, the middle value is calculated. ; and check whether the intermediate value meets the engineering specification Spec, when the If the engineering specification Spec is met, the upper limit is updated for , otherwise update the lower limit for ; until the Or the number of iterations reaches a threshold, and the search is stopped; Step S430: When the binary search is completed, the current middle value is returned As the smoothed parameter , and meets the specification Spec.
7. A model parameter automatic repair system based on Spec standard, using any one of the model parameter automatic repair methods according to claims 1 to 6, characterized in that: include: Response module, used to receive the model parameters obtained by the optimization algorithm ; Smoothing module, based on the model parameters , design feature construction to capture the changing trend of the model parameters, fit the features through the ridge regression model, and calculate the smoothed parameter values ; Optimize the processing module and set the initial interval of binary search to , and set the preset search precision As the termination condition; perform the binary search in the interval to iteratively adjust the upper and lower limits of the initial interval until the termination condition is met Or when the number of iterations reaches the threshold, the smoothing parameter is returned. As the final result.
8. The model parameter automatic repair system according to claim 7, characterized in that: The smoothing processing module comprises: A feature construction unit for constructing a , construct the feature vector used as the input feature vector of the ridge regression model , used to capture the changing trend of the model parameters, the feature construction includes polynomial features, nonlinear transformation features or differential features; Regression fitting unit, used to transform the feature vector Input into the ridge regression model, and solve the smoothed model parameters by minimizing the ridge regression objective function formula .
9. The model parameter automatic repair system according to claim 8, characterized in that: In the regression fitting unit, the smoothed model parameters for, ； Where X is an N*M input matrix, N is the number of samples, and M is the feature dimension; y is an N*1 column vector, which is the target value for each sample. is the regularization term, is the identity matrix, is the regularization hyperparameter.
10. The model parameter automatic repair system according to claim 7, characterized in that: The optimization processing module includes: An interval initialization unit is used to initialize the initial interval [ , ] is set to the parameter value and the model parameters ; The binary search execution unit is used to perform binary search in the interval to iteratively search and adjust the upper and lower limits of the initial interval, wherein in each iteration, the intermediate value is calculated ; and check whether the intermediate value meets the engineering specification Spec, when the If the engineering specification Spec is met, the upper limit is updated for , otherwise update the lower limit for ; until the Or the number of iterations reaches a threshold, and the search is stopped; The result output unit is used to return the current intermediate value after the binary search is completed. As the smoothed parameter , and meets the specification Spec.