Fishing gear selectivity parameter calibration method based on double booting

Abstract

The present application relates to the field of fishing gear parameter calibration, and discloses a fishing gear selective parameter calibration method based on double booting method, comprising the following steps: carrying out fishing gear selectivity test, collecting and pre-processing fishing data to construct a standard database; constructing a logistic regression selection model and determining initial parameters to obtain a target fitting model; introducing double booting method, carrying out sample level and residual level booting simulation, and quantifying double uncertainty; and outputting optimal fishing gear parameters through parameter fusion calibration optimization. The present application creatively adopts double booting method, solves the problems of low calibration precision and insufficient robustness of traditional methods, and improves parameter reliability.

Description

Technical Field

[0001] This invention relates to the field of fishing gear parameter calibration, and in particular to a method for calibrating selective parameters of fishing gear based on the double bootstrap method. Background Technology

[0002] Selectivity parameters of fishing gear, such as mesh selectivity, body length selection threshold, and escape probability, are core indicators for evaluating the rationality of fishing gear and achieving sustainable utilization of fishery resources. Their calibration accuracy directly determines the effectiveness of gear structure optimization and the efficiency of fishery resource protection. Currently, the calibration of fishing gear selectivity parameters mainly relies on traditional statistical modeling methods. This involves collecting catch data through field fishing experiments and combining this data with algorithms such as maximum likelihood estimation to solve for model parameters, thereby determining the selectivity characteristics of the fishing gear and providing a basis for gear design and operational optimization.

[0003] Current technical solutions have significant shortcomings and are unable to meet the requirements for high-precision and high-robustness calibration. On the one hand, traditional calibration methods estimate parameters based on only a single or small batch of original test samples, which is highly dependent on the randomness of sample sampling. When the original sample size is insufficient or the sample distribution is uneven, it is easy to lead to large deviations in parameter estimation, making it impossible to accurately represent parameter fluctuations in actual fishing scenarios. On the other hand, existing methods often ignore the errors generated during model fitting, only considering the uncertainty of sample sampling, without quantifying and correcting the model's own fitting deviation. This results in insufficient robustness of the calibrated parameters, poor adaptability to different environmental conditions, and difficulty in accurately guiding the optimization of fishing gear parameters. In addition, some calibration methods do not take into account the influence of covariates in the field environment, further reducing the practicality and specificity of parameter calibration, and failing to effectively solve the problem of accurately matching fishing gear selectivity with the protection of target fish species resources.

[0004] The fishing gear selectivity parameter calibration method based on the double boot method proposed in this patent can effectively solve the defects of the existing technology. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method for calibrating fishing gear selectivity parameters based on the double boot method.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The first aspect of this invention provides a method for calibrating fishing gear selectivity parameters based on the double-boot method, comprising the following steps: Selective tests were conducted on the fishing gear, and raw data related to the catch were collected and preprocessed after the tests. By combining a standard database of catches, a logistic regression selection model adapted to the length selection rules of catches is constructed, and the initial parameters of the model are calculated and determined to obtain the target fitting model. Based on the target fitting model, a double bootstrap method is introduced to perform the first-level sample-level bootstrap simulation and the second-level residual-level bootstrap simulation respectively. The target fitting model after uncertainty simulation is subjected to dual bootstrap parameter fusion and calibration optimization, and the parameters of the fishing gear to be calibrated are optimized based on the calibration results.

[0007] Furthermore, in a preferred embodiment of the present invention, the selective testing of fishing gear, and the subsequent collection of raw data related to the catch and data preprocessing, specifically include: Select the fishing gear that needs parameter calibration, mark it as the fishing gear to be calibrated, and build the main and auxiliary net fishing structure for the statement to be calibrated; Among them, if the catch escapes from the main net while in the fishing gear to be calibrated, it will be intercepted by the secondary net; Based on the fishing gear to be calibrated, multiple batches of parallel fishing and harvesting were carried out under conditions of fixed operating water depth, navigation speed, and release time throughout the entire process, and the catch was sorted into two types of samples after the fishing was completed. Among them, the first type of sample is the catch left in the main net, which is defined as the retained sample; the second type of sample is the catch intercepted in the secondary net, which is labeled as the escape sample. The length of the catch was measured for both types of samples, and a basic dataset of the catch was constructed based on the total number of tails and biological weight of different types of catches. During the construction of the basic dataset of catches, on-site environmental parameters are recorded synchronously, and the on-site environmental parameters are time-aligned with the data in the basic dataset of catches to output the target dataset of catches. Within the target dataset of the catch, raw data cleaning is performed, including removing abnormal and invalid data. After the raw data cleaning, the catch retention status is set with the length of the catch as the independent variable. The catches caught in the main net are marked as valid and the catches caught in the secondary net are marked as escaped. The data is then transformed into a unified structure to construct a standard database of the catch.

[0008] Furthermore, in a preferred embodiment of the present invention, the step of constructing a logistic regression selection model adapted to the length selection rules of catches by combining a standard database of catches, and calculating and determining the initial parameters of the model to obtain a target fitting model, specifically involves: Referring to the historical data network, a logistic regression selection model used to adapt to the long-term screening pattern of fish catch is retrieved from the historical data network and calibrated as the target benchmark model; Based on the standard database of fish catches, we extract fish catch length data, fish catch retention status and on-site environmental parameters. Among them, fish catch length data is used as the core independent variable of the target benchmark model, fish catch retention status is used as the classification dependent variable of the target benchmark model, and finally on-site environmental parameters are used as backup covariates for model correction. By combining the target benchmark model, a correlation mapping relationship is constructed between the length data of the catch and the survival probability of the catch of the fishing gear to be calibrated. Two parameters to be solved in the target benchmark model are determined, including the overall position of the selection curve in the target benchmark model, which is used to screen the standard body length data of the catch, and the slope of the selection curve, which is used to determine the discrimination sensitivity of the length data of the catch. The correlation mapping relationship between the classification dependent variable of the target benchmark model and the length data of the catch and the catch retention probability of the fishing gear to be calibrated is matched to construct the likelihood estimation system of the target benchmark model. The maximum likelihood estimation method is used to iteratively optimize and solve the two parameters to be solved in the target benchmark model. When the number of iterations for optimization is equal to the preset value, the optimal initial parameters of the target benchmark model are output. The optimal initial parameters of the target benchmark model are then substituted into the target benchmark model to obtain the initial fitting model of the target benchmark model, which is then calibrated as the target fitting model.

[0009] Furthermore, in a preferred embodiment of the present invention, the step of introducing a double bootstrap method based on the target fitting model to perform a first-level sample-level bootstrap simulation and a second-level residual-level bootstrap simulation specifically involves: Based on the standard database of catches, all data in the standard database of catches are designated as the sample dataset; The double bootstrap method is introduced, which involves random sampling with replacement of the sample dataset and a preset standard number of sample datasets to generate a first-level bootstrap sample dataset with a number equal to the standard number of sample datasets. The first-layer bootstrap sample dataset is imported into the target fitting model to solve the sample-level parameters using the maximum likelihood estimation method. Data outlier preprocessing is performed on all the solved sample-level parameters to obtain an effective sample-level parameter cluster. Among them, the data outlier preprocessing is to remove the sample set parameters that do not conform to the biological laws of fish from all the sample-level parameters obtained by the solution; Characterization experiments were conducted on the effective sample-level parameter cluster to obtain the parameter fluctuation range of the effective sample-level parameter cluster. Based on the parameter fluctuation range of the effective sample-level parameter cluster, a second-level residual-level bootstrap simulation was performed.

[0010] Furthermore, in a preferred embodiment of the present invention, the step of combining the parameter fluctuation range of the effective sample-level parameter cluster to perform the second-level residual-level bootstrap simulation processing specifically includes: Based on the fish survival status in the standard fish database, combined with the fish survival probability output when the target fitting model is running, the standardized fitting residual of each sample in the effective sample-level parameter cluster is calculated, and the standardized fitting residuals of all samples are integrated to obtain the global residual dataset. Based on the double bootstrap method, random sampling with replacement is performed on the global residual dataset, and the standard number of residual sampling is preset to be equal to the standard number of sample dataset. Based on the standard number of residual sampling, a second layer of residual bootstrap dataset is generated. The second residual bootstrap dataset of each group is associated and reconstructed with the catch length data and catch retention status in the catch standard database to generate a fitting error correction simulation sample set corresponding to the number of groups of the second residual bootstrap dataset. All the fitting error correction simulation sample sets are imported into the target fitting model, and the corresponding residual level parameters are solved based on the maximum likelihood estimation method. The residual level parameter clusters are then obtained. Data outlier preprocessing is performed on the residual parameter cluster. Specifically, outlier preprocessing involves removing residual parameters that do not conform to the biological laws of fish to obtain an effective residual parameter cluster. Characterization experiments were conducted on the effective residual level parameter cluster to obtain the parameter fluctuation range of the effective residual level parameter cluster. Combined with the parameter fluctuation range of the effective sample level parameter cluster, a two-level uncertainty simulation of the target fitting model based on the double bootstrap method was completed.

[0011] Furthermore, in a preferred embodiment of the present invention, the step of performing dual bootstrap parameter fusion and calibration optimization on the target fitting model after uncertainty simulation, and optimizing the parameters of the fishing gear to be calibrated based on the calibration results, specifically involves: The effective residual level parameter cluster and the effective sample level parameter cluster are merged across the entire domain to construct a total parameter sample cluster, wherein the total parameter sample cluster covers the parameter fluctuation range corresponding to the two parameter clusters. The mean-based algorithm is used to perform benchmark calibration on the total parameter sample cluster. After benchmark calibration, the parameter confidence interval within the total parameter sample cluster is calculated to determine the upper and lower limits of the parameter confidence interval and to determine the parameter fluctuation boundary. The field environmental parameters from the standard database of fish catches are imported into the target fitting model after uncertainty simulation to construct a target fitting model with environmental covariates, which is then calibrated as an environmental covariance fitting model. In the environmental covariate fitting model, the maximum likelihood estimation method is used to solve the corresponding coefficient parameters of the environmental covariates, and the parameter confidence intervals within the total parameter sample cluster are combined to output the calibrated optimal parameter set of the fishing gear. The calibrated optimal parameter set of fishing gear is substituted into the environmental covariance fitting model to obtain the final calibration model; Based on historical data networks, retrieve the range of variable structural parameters of the fishing gear to be calibrated, and determine whether the optimal parameter set of the calibrated fishing gear is within the corresponding range of variable structural parameters. If so, then the fishing gear parameters to be calibrated will be optimized and adjusted according to the optimal parameter set of the calibrated fishing gear. If not, the variable structural parameter range of the fishing gear to be calibrated will be substituted into the final calibration model for parameter simulation, matching the qualified calibration parameters of the fishing gear to be calibrated, and used to optimize and adjust the fishing gear parameters.

[0012] This invention addresses the technical deficiencies in the prior art and offers the following advantages: It conducts selective fishing gear experiments, collects and preprocesses catch data to construct a standard database; it constructs a logistic regression selection model and determines initial parameters to obtain a target fitting model; it introduces a double bootstrap method to perform sample-level and residual-level bootstrap simulations, quantifying dual uncertainties; and it outputs optimal fishing gear parameters through parameter fusion calibration optimization. This invention creatively employs the double bootstrap method to solve the problems of low calibration accuracy and insufficient robustness in traditional methods, thereby improving parameter reliability. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0014] Figure 1 A flowchart of a fishing gear selectivity parameter calibration method based on the double bootstrap method is shown; Figure 2 A flowchart of a method for performing two-level uncertainty simulation on a target fitted model is shown. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 A flowchart illustrating a fishing gear selectivity parameter calibration method based on the double boot method is shown, including the following steps: S102: Conduct selective tests on fishing gear, and collect raw data related to the catch after the tests and perform data preprocessing; S104: Combining the standard database of catches, construct a logistic regression selection model that adapts to the length selection rules of catches, and calculate and determine the initial parameters of the model to obtain the target fitting model; S106: Based on the target fitting model, a double bootstrap method is introduced to perform the first-level sample-level bootstrap simulation and the second-level residual-level bootstrap simulation respectively. S108: Perform dual bootstrap parameter fusion and calibration optimization on the target fitting model after uncertainty simulation, and optimize the parameters of the fishing gear to be calibrated based on the calibration results.

[0018] Furthermore, in a preferred embodiment of the present invention, the selective testing of fishing gear, and the subsequent collection of raw data related to the catch and data preprocessing, specifically include: Select the fishing gear that needs parameter calibration, mark it as the fishing gear to be calibrated, and build the main and auxiliary net fishing structure for the statement to be calibrated; Among them, if the catch escapes from the main net while in the fishing gear to be calibrated, it will be intercepted by the secondary net; Based on the fishing gear to be calibrated, multiple batches of parallel fishing and harvesting were carried out under conditions of fixed operating water depth, navigation speed, and release time throughout the entire process, and the catch was sorted into two types of samples after the fishing was completed. Among them, the first type of sample is the catch left in the main net, which is defined as the retained sample; the second type of sample is the catch intercepted in the secondary net, which is labeled as the escape sample. The length of the catch was measured for both types of samples, and a basic dataset of the catch was constructed based on the total number of tails and biological weight of different types of catches. During the construction of the basic dataset of catches, on-site environmental parameters are recorded synchronously, and the on-site environmental parameters are time-aligned with the data in the basic dataset of catches to output the target dataset of catches. Within the target dataset of the catch, raw data cleaning is performed, including removing abnormal and invalid data. After the raw data cleaning, the catch retention status is set with the length of the catch as the independent variable. The catches caught in the main net are marked as valid and the catches caught in the secondary net are marked as escaped. The data is then transformed into a unified structure to construct a standard database of the catch.

[0019] It should be noted that the fishing gear needs to be set up in a main net operation and a secondary net interception system. The main net is responsible for routine catching and screening, while all escaped fish are intercepted and captured by the meticulous secondary net. The purpose is to achieve full traceability of escaped samples, eliminate the error of missing escape data from a physical mechanism, and significantly improve the accuracy of subsequent selection parameter calculations. Maintaining fixed operating water depth, navigation speed, and deployment time throughout the process can reduce errors from random operations and eliminate interference from differences in human factors and the environment. After the fishing is completed, the catch from the main net is defined and retained, while the catch from the secondary net is defined as escaped fish. Standardized classification rules are established, and the catch size is statistically analyzed using body length as the core indicator, along with quantity and weight, to build a basic raw data table. Body length data is the core independent variable for selective modeling of fishing gear, supporting subsequent parameter fitting. It is also synchronously bound to on-site environmental parameters, retaining environmental covariates, and reserving data interfaces for later model expansion, parameter environment correction, and differential calibration.

[0020] Furthermore, in a preferred embodiment of the present invention, the step of constructing a logistic regression selection model adapted to the length selection rules of catches by combining a standard database of catches, and calculating and determining the initial parameters of the model to obtain a target fitting model, specifically involves: Referring to the historical data network, a logistic regression selection model used to adapt to the long-term screening pattern of fish catch is retrieved from the historical data network and calibrated as the target benchmark model; Based on the standard database of fish catches, we extract fish catch length data, fish catch retention status and on-site environmental parameters. Among them, fish catch length data is used as the core independent variable of the target benchmark model, fish catch retention status is used as the classification dependent variable of the target benchmark model, and finally on-site environmental parameters are used as backup covariates for model correction. By combining the target benchmark model, a correlation mapping relationship is constructed between the length data of the catch and the survival probability of the catch of the fishing gear to be calibrated. Two parameters to be solved in the target benchmark model are determined, including the overall position of the selection curve in the target benchmark model, which is used to screen the standard body length data of the catch, and the slope of the selection curve, which is used to determine the discrimination sensitivity of the length data of the catch. The correlation mapping relationship between the classification dependent variable of the target benchmark model and the length data of the catch and the catch retention probability of the fishing gear to be calibrated is matched to construct the likelihood estimation system of the target benchmark model. The maximum likelihood estimation method is used to iteratively optimize and solve the two parameters to be solved in the target benchmark model. When the number of iterations for optimization is equal to the preset value, the optimal initial parameters of the target benchmark model are output. The optimal initial parameters of the target benchmark model are then substituted into the target benchmark model to obtain the initial fitting model of the target benchmark model, which is then calibrated as the target fitting model.

[0021] It should be noted that the logistic regression selection model chosen to fit the fish length screening pattern is a universally available model. Subsequently, body length is used as the core influencing factor, survival / escape labels as classification results, and environmental parameters are reserved for later correction, strictly defining the model input hierarchy. Next, it is necessary to establish a mapping relationship between body length and survival probability to determine the core ball-carrying parameters. That is, to use the model to establish the correspondence between "how big the fish is → how likely it is to be caught in the net"; defining two physically meaningful parameters: one controls the standard screening body length (the left and right positions of the curve), and the other controls the sensitivity of distinguishing between large and small fish (the steepness of the curve). These two core parameters correspond to the two major performance characteristics of fishing gear screening benchmark size and the ability to distinguish between large and small fish.

[0022] The algorithm matches the actual survival / escape sample labels with the model's predicted survival probabilities to construct a global sample inference logic. Through a maximum likelihood iterative algorithm, it continuously fine-tunes two core parameters to reduce the deviation between the model's predictions and actual fishing results. The iteration stops after a preset number of iterations, outputting the target fitted model. This model can transform mathematical parameters into practical performance indicators that fishing gear can optimize.

[0023] Furthermore, in a preferred embodiment of the present invention, the step of performing dual bootstrap parameter fusion and calibration optimization on the target fitting model after uncertainty simulation, and optimizing the parameters of the fishing gear to be calibrated based on the calibration results, specifically involves: The effective residual level parameter cluster and the effective sample level parameter cluster are merged across the entire domain to construct a total parameter sample cluster, wherein the total parameter sample cluster covers the parameter fluctuation range corresponding to the two parameter clusters. The mean-based algorithm is used to perform benchmark calibration on the total parameter sample cluster. After benchmark calibration, the parameter confidence interval within the total parameter sample cluster is calculated to determine the upper and lower limits of the parameter confidence interval and to determine the parameter fluctuation boundary. The field environmental parameters from the standard database of fish catches are imported into the target fitting model after uncertainty simulation to construct a target fitting model with environmental covariates, which is then calibrated as an environmental covariance fitting model. In the environmental covariate fitting model, the maximum likelihood estimation method is used to solve the corresponding coefficient parameters of the environmental covariates, and the parameter confidence intervals within the total parameter sample cluster are combined to output the calibrated optimal parameter set of the fishing gear. The calibrated optimal parameter set of fishing gear is substituted into the environmental covariance fitting model to obtain the final calibration model; Based on historical data networks, retrieve the range of variable structural parameters of the fishing gear to be calibrated, and determine whether the optimal parameter set of the calibrated fishing gear is within the corresponding range of variable structural parameters. If so, then the fishing gear parameters to be calibrated will be optimized and adjusted according to the optimal parameter set of the calibrated fishing gear. If not, the variable structural parameter range of the fishing gear to be calibrated will be substituted into the final calibration model for parameter simulation, matching the qualified calibration parameters of the fishing gear to be calibrated, and used to optimize and adjust the fishing gear parameters.

[0024] It should be noted that merging the two layers to form an integrated total parameter sample cluster fully incorporates all parameter fluctuation ranges corresponding to sampling errors and model fitting errors. The aim is to form a global parameter distribution base, avoiding calibration bias caused by single-dimensional parameters. Subsequently, a comprehensive benchmark calibration calculation is performed on the total parameter sample cluster. Then, the confidence limits of core parameters are calculated using interval statistics, clarifying the maximum reasonable fluctuation range of each key parameter and quantifying the parameter reliability boundary. Field environmental parameters such as water temperature, flow velocity, and water depth, synchronously retained and time-aligned during the previous data acquisition phase, are retrieved and embedded into the target fitting model that has undergone uncertainty simulation. An upgraded covariate fitting model with environmental correction factors is constructed to compensate for the shortcomings of traditional models in handling differences in the aquatic environment. Finally, the maximum likelihood estimation logic is reused to solve for the correction coefficients corresponding to the environmental covariates. Then, the global parameter confidence intervals are integrated to comprehensively output the final optimal parameter set that takes into account both errors and environmental impacts, used to generate the final calibration model. Based on the final calibration model, the theoretically optimal parameters are compared to see if they are within the allowable range of actual production, and adjustments are then made.

[0025] Figure 2 The flowchart illustrates a method for performing two-level uncertainty simulation on a target fitted model, including the following steps: S202: Based on the target fitting model, a double bootstrap method is introduced to perform the first-level sample-level bootstrap simulation and the second-level residual-level bootstrap simulation respectively. S204: Combine the parameter fluctuation range of the effective sample-level parameter cluster to perform a second-level residual-level shoegear simulation.

[0026] Furthermore, in a preferred embodiment of the present invention, the step of introducing a double bootstrap method based on the target fitting model to perform a first-level sample-level bootstrap simulation and a second-level residual-level bootstrap simulation specifically involves: Based on the standard database of catches, all data in the standard database of catches are designated as the sample dataset; The double bootstrap method is introduced, which involves random sampling with replacement of the sample dataset and a preset standard number of sample datasets to generate a first-level bootstrap sample dataset with a number equal to the standard number of sample datasets. The first-layer bootstrap sample dataset is imported into the target fitting model to solve the sample-level parameters using the maximum likelihood estimation method. Data outlier preprocessing is performed on all the solved sample-level parameters to obtain an effective sample-level parameter cluster. Among them, the data outlier preprocessing is to remove the sample set parameters that do not conform to the biological laws of fish from all the sample-level parameters obtained by the solution; Characterization experiments were conducted on the effective sample-level parameter cluster to obtain the parameter fluctuation range of the effective sample-level parameter cluster. Based on the parameter fluctuation range of the effective sample-level parameter cluster, a second-level residual-level bootstrap simulation was performed.

[0027] It should be noted that designating all data within the standard fish catch database as the sample dataset ensures a unified, clean, and traceable sampling population, preventing dirty or non-standard data from interfering with the bootstrap simulation results. The double bootstrap method prioritizes the original data and uses repeated sampling with replacement to create multiple sets of simulated data for calculating error fluctuations. For example, in this application, bootstrap can be understood as releasing 100 fish back into the mix, recording one, releasing it again, and repeating this process until 100 fish are collected, creating a new set of simulated data. Repeating this process multiple times yields multiple sets of experimental data.

[0028] Each set of simulated boot-growing samples was substituted into the target fitting model defined earlier, and the corresponding model parameters were solved using a unified maximum likelihood estimation method. Then, a unified screening and filtering process was implemented, retaining compliant parameters to form a parameter cluster. The aim was to generate unique parameters for each set of experimental data, forming multiple parameter samples that reflected the range of parameter fluctuations. Targeted screening thresholds were also set to eliminate calculated parameters that violated fish growth patterns.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of combining the parameter fluctuation range of the effective sample-level parameter cluster to perform the second-level residual-level bootstrap simulation processing specifically includes: Based on the fish survival status in the standard fish database, combined with the fish survival probability output when the target fitting model is running, the standardized fitting residual of each sample in the effective sample-level parameter cluster is calculated, and the standardized fitting residuals of all samples are integrated to obtain the global residual dataset. Based on the double bootstrap method, random sampling with replacement is performed on the global residual dataset, and the standard number of residual sampling is preset to be equal to the standard number of sample dataset. Based on the standard number of residual sampling, a second layer of residual bootstrap dataset is generated. The second residual bootstrap dataset of each group is associated and reconstructed with the catch length data and catch retention status in the catch standard database to generate a fitting error correction simulation sample set corresponding to the number of groups of the second residual bootstrap dataset. All the fitting error correction simulation sample sets are imported into the target fitting model, and the corresponding residual level parameters are solved based on the maximum likelihood estimation method. The residual level parameter clusters are then obtained. Data outlier preprocessing is performed on the residual parameter cluster. Specifically, outlier preprocessing involves removing residual parameters that do not conform to the biological laws of fish to obtain an effective residual parameter cluster. Characterization experiments were conducted on the effective residual level parameter cluster to obtain the parameter fluctuation range of the effective residual level parameter cluster. Combined with the parameter fluctuation range of the effective sample level parameter cluster, a two-level uncertainty simulation of the target fitting model based on the double bootstrap method was completed.

[0030] It's important to note that the first step is to compare the actual survival / escape results of each fish with the theoretical survival probability calculated by the target fitted model. The difference between the model prediction and the actual data is calculated row by row, resulting in the standardized residual. These residuals are then aggregated to build a global residual library. Subsequently, random sampling with replacement is performed to construct multiple sets of residual simulation datasets, ensuring that the two-layer bootstrap simulations are of equal size and dimensionality, allowing for mathematical comparability during parameter merging and calibration. The innovation lies in using the residual bootstrap mechanism without altering the original fish catch data, maintaining a unified sampling standard, and ensuring a symmetrical and logically rigorous two-layer bootstrap structure. Later, the generated error-corrected simulation sample set retains fishery characteristics and undergoes a new maximum likelihood estimation, allowing for the calculation of multiple sets of model parameters affected by the fitting error, which are then aggregated into a large cluster of residual parameters. Finally, unqualified parameters were removed, and a characterization experiment was conducted on the effective residual parameter cluster to obtain the parameter fluctuation range of the effective residual parameter cluster. This achieved full coverage of both random errors in fishing at sea and fitting errors in modeling calculations, thus optimizing the model.

[0031] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for calibrating fishing gear selectivity parameters based on the double-bootlet method, characterized in that, Includes the following steps: Selective tests were conducted on the fishing gear, and raw data related to the catch were collected and preprocessed after the tests. By combining a standard database of catches, a logistic regression selection model adapted to the length selection rules of catches is constructed, and the initial parameters of the model are calculated and determined to obtain the target fitting model. Based on the target fitting model, a double bootstrap method is introduced to perform the first-level sample-level bootstrap simulation and the second-level residual-level bootstrap simulation respectively. The target fitting model after uncertainty simulation is subjected to dual bootstrap parameter fusion and calibration optimization, and the parameters of the fishing gear to be calibrated are optimized based on the calibration results.

2. The method for calibrating fishing gear selectivity parameters based on the double-bootlet method according to claim 1, characterized in that, The selective testing of fishing gear, followed by the collection of raw data related to the catch and data preprocessing, specifically involves: Select the fishing gear that needs parameter calibration, mark it as the fishing gear to be calibrated, and build the main and auxiliary net fishing structure for the statement to be calibrated; Among them, if the catch escapes from the main net while in the fishing gear to be calibrated, it will be intercepted by the secondary net; Based on the fishing gear to be calibrated, multiple batches of parallel fishing and harvesting were carried out under conditions of fixed operating water depth, navigation speed, and release time throughout the entire process, and the catch was sorted into two types of samples after the fishing was completed. Among them, the first type of sample is the catch left in the main net, which is defined as the retained sample; the second type of sample is the catch intercepted in the secondary net, which is labeled as the escape sample. The length of the catch was measured for both types of samples, and a basic dataset of the catch was constructed based on the total number of tails and biological weight of different types of catches. During the construction of the basic dataset of catches, on-site environmental parameters are recorded synchronously, and the on-site environmental parameters are time-aligned with the data in the basic dataset of catches to output the target dataset of catches. Within the target dataset of the catch, raw data cleaning is performed, including removing abnormal and invalid data. After the raw data cleaning, the catch retention status is set with the length of the catch as the independent variable. The catches caught in the main net are marked as valid and the catches caught in the secondary net are marked as escaped. The data is then transformed into a unified structure to construct a standard database of the catch.

3. The method for calibrating fishing gear selectivity parameters based on the double-bootlet method according to claim 1, characterized in that, The method involves combining a standard database of catches to construct a logistic regression selection model that adapts to the length selection rules of catches, and calculating and determining the initial parameters of the model to obtain the target fitting model, specifically: Referring to the historical data network, a logistic regression selection model used to adapt to the long-term screening pattern of fish catch is retrieved from the historical data network and calibrated as the target benchmark model; Based on the standard database of fish catches, we extract fish catch length data, fish catch retention status and on-site environmental parameters. Among them, fish catch length data is used as the core independent variable of the target benchmark model, fish catch retention status is used as the classification dependent variable of the target benchmark model, and finally on-site environmental parameters are used as backup covariates for model correction. By combining the target benchmark model, a correlation mapping relationship is constructed between the length data of the catch and the survival probability of the catch of the fishing gear to be calibrated. Two parameters to be solved in the target benchmark model are determined, including the overall position of the selection curve in the target benchmark model, which is used to screen the standard body length data of the catch, and the slope of the selection curve, which is used to determine the discrimination sensitivity of the length data of the catch. The correlation mapping relationship between the classification dependent variable of the target benchmark model and the length data of the catch and the catch retention probability of the fishing gear to be calibrated is matched to construct the likelihood estimation system of the target benchmark model. The maximum likelihood estimation method is used to iteratively optimize and solve the two parameters to be solved in the target benchmark model. When the number of iterations for optimization is equal to the preset value, the optimal initial parameters of the target benchmark model are output. The optimal initial parameters of the target benchmark model are then substituted into the target benchmark model to obtain the initial fitting model of the target benchmark model, which is then calibrated as the target fitting model.

4. The method for calibrating fishing gear selectivity parameters based on the double-bootlet method according to claim 1, characterized in that, The target fitting model introduces a double bootstrap method, performing a first-level sample-level bootstrap simulation and a second-level residual-level bootstrap simulation, specifically as follows: Based on the standard database of catches, all data in the standard database of catches are designated as the sample dataset; The double bootstrap method is introduced, which involves random sampling with replacement of the sample dataset and a preset standard number of sample datasets to generate a first-level bootstrap sample dataset with a number equal to the standard number of sample datasets. The first-layer bootstrap sample dataset is imported into the target fitting model to solve the sample-level parameters using the maximum likelihood estimation method. Data outlier preprocessing is performed on all the solved sample-level parameters to obtain an effective sample-level parameter cluster. Among them, the data outlier preprocessing is to remove the sample set parameters that do not conform to the biological laws of fish from all the sample-level parameters obtained by the solution; Characterization experiments were conducted on the effective sample-level parameter cluster to obtain the parameter fluctuation range of the effective sample-level parameter cluster. Based on the parameter fluctuation range of the effective sample-level parameter cluster, a second-level residual-level bootstrap simulation was performed.

5. The method for calibrating fishing gear selectivity parameters based on the double-bootlet method according to claim 4, characterized in that, The second-level residual-level bootstrap simulation processing, which combines the parameter fluctuation range of the effective sample-level parameter cluster, is specifically as follows: Based on the fish survival status in the standard fish database, combined with the fish survival probability output when the target fitting model is running, the standardized fitting residual of each sample in the effective sample-level parameter cluster is calculated, and the standardized fitting residuals of all samples are integrated to obtain the global residual dataset. Based on the double bootstrap method, random sampling with replacement is performed on the global residual dataset, and the standard number of residual sampling is preset to be equal to the standard number of sample dataset. Based on the standard number of residual sampling, a second layer of residual bootstrap dataset is generated. The second residual bootstrap dataset of each group is associated and reconstructed with the catch length data and catch retention status in the catch standard database to generate a fitting error correction simulation sample set corresponding to the number of groups of the second residual bootstrap dataset. All the fitting error correction simulation sample sets are imported into the target fitting model, and the corresponding residual level parameters are solved based on the maximum likelihood estimation method. The residual level parameter clusters are then obtained. Data outlier preprocessing is performed on the residual parameter cluster. Specifically, outlier preprocessing involves removing residual parameters that do not conform to the biological laws of fish to obtain an effective residual parameter cluster. Characterization experiments were conducted on the effective residual level parameter cluster to obtain the parameter fluctuation range of the effective residual level parameter cluster. Combined with the parameter fluctuation range of the effective sample level parameter cluster, a two-level uncertainty simulation of the target fitting model based on the double bootstrap method was completed.

6. The method for calibrating fishing gear selectivity parameters based on the double-bootlet method according to claim 1, characterized in that, The process involves dual bootstrap parameter fusion and calibration optimization of the target fitting model after uncertainty simulation, followed by optimization of the fishing gear parameters to be calibrated based on the calibration results. The effective residual level parameter cluster and the effective sample level parameter cluster are merged across the entire domain to construct a total parameter sample cluster, wherein the total parameter sample cluster covers the parameter fluctuation range corresponding to the two parameter clusters. The mean-based algorithm is used to perform benchmark calibration on the total parameter sample cluster. After benchmark calibration, the parameter confidence interval within the total parameter sample cluster is calculated to determine the upper and lower limits of the parameter confidence interval and to determine the parameter fluctuation boundary. The field environmental parameters from the standard database of fish catches are imported into the target fitting model after uncertainty simulation to construct a target fitting model with environmental covariates, which is then calibrated as an environmental covariance fitting model. In the environmental covariate fitting model, the maximum likelihood estimation method is used to solve the corresponding coefficient parameters of the environmental covariates, and the parameter confidence intervals within the total parameter sample cluster are combined to output the calibrated optimal parameter set of the fishing gear. The calibrated optimal parameter set of fishing gear is substituted into the environmental covariance fitting model to obtain the final calibration model; Based on historical data networks, retrieve the range of variable structural parameters of the fishing gear to be calibrated, and determine whether the optimal parameter set of the calibrated fishing gear is within the corresponding range of variable structural parameters. If so, then the fishing gear parameters to be calibrated will be optimized and adjusted according to the optimal parameter set of the calibrated fishing gear. If not, the variable structural parameter range of the fishing gear to be calibrated will be substituted into the final calibration model for parameter simulation, matching the qualified calibration parameters of the fishing gear to be calibrated, and used to optimize and adjust the fishing gear parameters.

Images