Intelligent interpretable forecasting method for northwest pacific ocean squid center fishery

Abstract

The invention discloses an interpretable intelligent forecasting method for a northwest pacific ocean squid center fishing ground, and belongs to the technical field of ocean fishing condition analysis and processing. Comprising the steps that historical fishery data and corresponding environment data are acquired, and the fishing catch per unit fishing effort per month is obtained; fitting the fishing catch per unit fishing effort per month and the environment data, and screening out an environment variable with a dome-shaped residual error as an interpretable environment variable; constructing a fishery forecasting model based on an extreme gradient lifting tree; according to the month to which the target date belongs, a central fishing ground for squid inhabiting in the northwest Pacific Ocean is obtained through spatial intersection and spatial superposition qualitative analysis; and quantitatively predicting a central fishery in which the squid inhabits in the northwest Pacific Ocean based on the fishery forecasting model. According to the method, the credibility of fishery managers and fishermen on forecast products is enhanced through the interpretable environment variables screened by the generalized additive model, and the extreme gradient boosting tree fishery forecast model constructed on the basis of the interpretable environment variables can effectively capture the change of the squid resource quantity.

Description

Technical Field

[0001] The invention relates to the technical field of marine fishery situation analysis and processing, and in particular to an interpretable intelligent forecasting method for the northwest Pacific squid central fishing ground. Background Art

[0002] The Northwest Pacific squid (Ommastrephes bartramii) is an important economic cephalopod widely distributed in the North Pacific. The confluence of the Kuroshio warm current and the Oyashio cold current in the Northwest Pacific forms a complex marine environment. Driven by factors such as temperature and nutrients, squid migrate between spawning grounds in subtropical waters and feeding grounds in the subarctic waters. my country has been commercially fishing squid in the Northwest Pacific since 1994. According to the distribution area and peak spawning period, it is divided into winter-spring groups in the east and autumn groups in the west, with 170°E as the boundary. Currently, the winter-spring group is the main fishing group. Due to the short capture period, there is a high demand for the prediction of the abundance of squid resources in the Northwest Pacific.

[0003] Scholars at home and abroad have explored the impact of marine environmental factors such as sea surface temperature (SST) and seawater chlorophyll concentration (Chlorophyll-a, Chl-a) on the abundance of squid resources. These studies have revealed the mechanism of dynamic changes in squid resources to a certain extent. However, marine habitat data are of various types and large scale. Although AI algorithms have the characteristics of high precision, their interpretability is poor. Therefore, when predicting the abundance of squid resources in the northwest Pacific, identifying central fishing grounds by screening environmental factors with strong interpretability can not only enhance the trust of fishery managers and fishermen in forecast products and improve the public's acceptance of blue aquatic products, but also effectively distinguish low-yield and high-yield areas of fishing grounds, thereby improving prediction accuracy. In addition, from the perspective of squid fishery resource management and protection in the northwest Pacific, high-yield fishing areas are key areas for resource protection. Therefore, accurately predicting and determining the location of these areas is of great significance for the management and protection of squid resources in the northwest Pacific. Summary of the invention

[0004] In view of the defects of the prior art, the present invention provides an interpretable intelligent forecasting method for the northwest Pacific squid center fishing ground, which takes into account the marine environmental factors experienced by squid during its life history, and reconstructs a fishing ground forecasting model based on the extreme gradient boosting tree (eXtreme Gradient Boosting, XGBoost) on the basis of screening interpretable environmental variables by a generalized additive model (Generalized Additive Model, GAM), and takes the catch per unit effort (CPUE) that reflects the fishing situation as a benchmark indicator to improve the prediction accuracy of squid resource abundance and narrow the uncertainty range.

[0005] In order to achieve the above object, the present invention provides an interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground, comprising the following steps:

[0006] (1) Obtaining fishery data and corresponding environmental data for N historical months; the fishery data includes operation location, operation date, catch, and number of fishing vessels; the environmental factors include water temperature, temperature difference, salinity, dissolved oxygen, primary productivity, phytoplankton concentration, chlorophyll concentration, sea surface height, mixed layer depth, and ocean front;

[0007] (2) calculating the monthly catch per unit of fishing effort based on the historical fishery data according to the operation location and operation date;

[0008] (3) matching the environmental data with the fishery data, and fitting the monthly catch per unit fishing effort and the environmental data based on a generalized additive model, and selecting environmental variables with dome-shaped fitting curves as interpretable environmental variables;

[0009] (4) constructing a fishery forecast model based on an extreme gradient boosting tree, and using the monthly unit fishing effort catch and interpretable environmental variables to train the fishery forecast model for each month;

[0010] (5) Based on the month to which the target date belongs, the high-yield area, medium-yield area, and low-yield area of ​​the Northwest Pacific squid fishery are obtained through spatial intersection and spatial superposition;

[0011] (6) extracting the same environmental variables in the target date based on the interpretable environmental variables selected in the month to which the target date belongs, and inputting them into the fishing ground forecast model to quantitatively predict the central fishing ground suitable for the habitat of the Northwest Pacific squid.

[0012] Furthermore, N is 6; the obtaining of the fishery data and the corresponding environmental data for N historical months is to obtain the fishery data and the corresponding environmental data of squid from June to November 1995 to 2022.

[0013] Furthermore, the monthly catch per unit fishing effort is specifically:

[0014]

[0015] Among them: CPUE ymij 、Catch ymij and Effort ymij They represent the catch per unit fishing effort, catch volume and number of fishing vessels at longitude i, latitude j, month m and year y respectively.

[0016] Furthermore, the step (3) is specifically as follows:

[0017] (3.1) Convert all environmental data into a grid that matches the historical fishery data to accommodate the spatial resolution of the fishery data for each month and match them based on the operation location and operation date;

[0018] (3.2) fitting the monthly catch per unit fishing effort and the corresponding environmental data based on the generalized additive model;

[0019] (3.3) A minimum threshold is set. In the fitting curve of the monthly unit fishing effort catch and the corresponding environmental data, when the monthly unit fishing effort catch of each month exceeds the set minimum threshold, the fitting curve takes a dome shape. This environmental variable is the screened explainable environmental variable of that month.

[0020] Furthermore, the minimum threshold is represented by a percentile, that is, according to the data distribution, a specified percentage of data items are lower than this threshold; the dome shape shows a trend of first increasing and then decreasing.

[0021] Furthermore, in step (4), the parameters of the fishery forecast model are optimized by an automatic tuning method to minimize the mean square error (MSE) to evaluate the performance.

[0022] Furthermore, the step (5) is specifically as follows:

[0023] (5.1) Set three percentile thresholds corresponding to high-yield areas, medium-yield areas, and low-yield areas;

[0024] (5.2) sorting the monthly catch per unit fishing effort corresponding to each interpretable environmental variable selected according to the month to which the target date belongs, extracting the monthly catch per unit fishing effort exceeding the set percentile threshold, and the concentration of the interpretable environmental variable corresponding to the monthly catch per unit fishing effort is the concentration range below the threshold;

[0025] (5.3) extracting the same interpretable environmental variables in the target date according to the interpretable environmental variables selected from the month to which the target date belongs;

[0026] (5.4) Under the same threshold, the spatial locations where each explanatory environmental variable is within its corresponding concentration range on the target date are screened out, and then the spatial intersection analysis is performed on the spatial locations of all the screened explanatory environmental variables, that is, the same spatial locations are retained, and the different spatial locations are discarded;

[0027] (5.5) For the three percentile thresholds, the spatial locations of the high-yield area, medium-yield area, and low-yield area of ​​the Northwest Pacific squid fishery are obtained according to steps (5.2) to (5.4).

[0028] Furthermore, the percentile thresholds corresponding to the high-yield area, medium-yield area and low-yield area are 80%, 60% and 40% respectively.

[0029] Beneficial effects of the present invention:

[0030] The present invention enhances the trust of fishery managers and fishermen in forecast products through interpretable environmental variables screened by generalized additive models, and improves the public's acceptance of blue aquatic products. The extreme gradient boosting tree fishery forecast model constructed based on interpretable environmental variables can effectively capture the changes in squid resources, and uses the catch per unit effort (CPUE) as a benchmark indicator to reflect the fishing situation. A more accurate central fishery forecast model is established, which improves the prediction accuracy of squid resource abundance and reduces the uncertainty range. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The present invention is a flowchart of an intelligent forecasting method for the Northwest Pacific squid center fishing ground that can be explained by an embodiment of the present invention.

[0032] Figure 2 This is an example diagram of the screening of key environmental factors in an embodiment of the present invention.

[0033] Figure 3 This is a schematic diagram of the Northwest Pacific Squid Central Fishery according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. The directions or positional relationships indicated by the terms "upper", "lower", "front", "back", "left", "right", "bottom", "inside", "outside", etc. are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention.

[0035] like Figure 1 As shown, an embodiment of the present invention provides an interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground, comprising the following steps:

[0036] S101. Obtain fishery data and corresponding environmental data for N historical months; the fishery data include operation location (longitude and latitude), operation date (year and month), catch (tons), number of fishing vessels (number of vessels); the environmental factors include water temperature, temperature difference, salinity, dissolved oxygen, primary productivity, phytoplankton concentration, chlorophyll concentration, sea surface height, mixed layer depth and ocean front.

[0037] In the embodiment of the present invention, the fishery data comes from the China Offshore Fisheries Data Center of Shanghai Ocean University, with a "monthly" time resolution and a spatial resolution of 0.25° latitude × 0.25° longitude. The data range is from June to November 1995 to 2022, and the spatial range covers the main fishing grounds of 145°-170°W and 35°-50°S. The data content includes the operation location (latitude and longitude), operation date (year, month), catch (Catch, unit: tons), and number of fishing vessels (Vessel, unit: only). Environmental variables include water temperature, salinity and temperature difference at 5 depth layers (surface, 47 meters, 92 meters, 156 meters, 222 meters), sea surface height, mixed layer depth and ocean front with a time resolution of months and a spatial resolution of 0.25°×0.25°, as well as primary productivity, phytoplankton concentration, dissolved oxygen concentration and chlorophyll concentration at 5 depth layers with a time resolution of months and a spatial resolution of 0.083°×0.083°.

[0038] S102. Calculate the monthly unit fishing effort catch according to the historical fishery data according to the operation location and operation date.

[0039]

[0040] Among them: CPUEy mij 、Catch ymij and Effort ymijThey represent the catch per unit fishing effort, catch volume and number of fishing vessels at longitude i, latitude j, month m and year y respectively.

[0041] S103, matching the environmental data with the fishery data, and fitting the monthly catch per unit fishing effort and the environmental data based on a generalized additive model, and selecting environmental variables with dome-shaped fitting curves as interpretable environmental variables.

[0042] First, the fishery data and environmental data need to be grid-matched. The grid cells of primary productivity, phytoplankton concentration, dissolved oxygen concentration, and chlorophyll concentration in the environmental data selected in the embodiment of the present invention are relatively small, so it is necessary to apply the spatial averaging method to resample the data from a higher resolution of 0.083° to a coarser resolution of 0.25°. The specific method is: calculate the arithmetic mean of all values ​​in the smaller grid cells (0.083° resolution) of each larger grid cell (0.25° resolution) as the value of the larger grid cell.

[0043] Then the catch per unit fishing effort and environmental data were fitted based on the generalized additive model.

[0044] Finally, a minimum threshold is set. In each unit fishing effort catch and environmental data fitting curve, when the unit fishing effort catch exceeds the set minimum threshold, the fitting curve takes on a dome shape, that is, it shows a trend of first increasing and then decreasing (e.g. Figure 2 ), this environmental variable is the filtered interpretable environmental variable. Here, the lowest threshold is expressed as percentile, which is a method to determine the threshold based on data distribution. Percentile is a statistic that describes the location of data distribution. It indicates that in a set of data, a specified percentage of data items are lower than this value. For example, the 40th percentile means that 40% of the data items are lower than this value.

[0045] S104, constructing a fishing ground forecast model based on the extreme gradient boosting tree (eXtreme Gradient Boosting, XGBoost), and using the monthly unit fishing effort catch and interpretable environmental variables to train the fishing ground forecast model for each month.

[0046] The XGB model is rebuilt based on the selected interpretable environmental variables. The parameters are optimized by automatic tuning to minimize the mean square error (MSE) to evaluate the performance, and finally the optimal parameters are used to train the final forecast model.

[0047] S105. According to the month mentioned in the target date, the high-yield, medium-yield and low-yield areas of the Northwest Pacific squid fishery are obtained through spatial intersection and spatial superposition.

[0048] (1) Three percentile thresholds are set to correspond to high-yield area, medium-yield area and low-yield area, respectively. In the embodiment of the present invention, the three percentile thresholds are 40%, 60% and 80%, respectively, corresponding to the low-yield area, medium-yield area and high-yield area of ​​the Northwest Pacific squid fishing ground.

[0049] (2) The monthly catch per unit fishing effort corresponding to each interpretable environmental variable selected according to the month to which the target date (a certain year, a certain month, a certain day) belongs is sorted, and the monthly catch per unit fishing effort that exceeds the set percentile threshold is extracted. The concentration of the corresponding interpretable environmental variable is the concentration range below the threshold.

[0050] like Figure 2 As shown, the response value represents the catch per unit fishing effort (CPUE), the habitat factor represents the interpretable environmental variables, the solid line represents the curve of CPUE fluctuating with the concentration range of interpretable environmental data, and the upper and lower dotted lines represent the confidence intervals.

[0051] (3) Extract the same interpretable environmental variables in the target date (year, month, and day) based on the interpretable environmental variables selected based on the month to which the target date belongs.

[0052] (4) Under the same threshold, the spatial locations of each explanatory environmental variable within its corresponding concentration range on the target date are screened out, and then the spatial locations of all screened out environmental variables are subjected to spatial intersection analysis, that is, the same spatial locations corresponding to each explanatory environmental variable are retained, and the different spatial locations are discarded.

[0053] (5) For the three percentile thresholds, the spatial locations of the high-yield area, medium-yield area, and low-yield area of ​​the Northwest Pacific squid fishery are obtained according to steps (2) to (4).

[0054] like Figure 3 The upper middle figure shows the overlay of all explainable environmental variables at different thresholds (40%, 60% and 80%), which correspond to the low-yield, medium-yield and high-yield areas of the Northwest Pacific squid fishery. The darkest color corresponds to the low-yield area, and the darkest color corresponds to the medium-yield area.

[0055] S106, extracting the same environmental variables in the target date based on the interpretable environmental variables screened out in the month to which the target date belongs, and inputting them into the fishing ground forecast model to quantitatively predict the central fishing ground suitable for the habitat of Northwest Pacific squid.

[0056] like Figure 3 The lower middle figure shows a fishing ground forecast map given by the fishing ground forecast model based on the extreme gradient boosting tree. The relative abundance is the predicted catch per unit fishing effort.

[0057] By combining the results of step S104 and step S106, the intelligent forecast of the Northwest Pacific squid center fishing ground can be obtained.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the principles and spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. An interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground, characterized in that: The steps include: (1) Obtaining fishery data and corresponding environmental data for N historical months; the fishery data includes operation location, operation date, catch, and number of fishing vessels; the environmental factors include water temperature, temperature difference, salinity, dissolved oxygen, primary productivity, phytoplankton concentration, chlorophyll concentration, sea surface height, mixed layer depth, and ocean front; (2) calculating the monthly catch per unit of fishing effort based on the historical fishery data according to the operation location and operation date; (3) matching the environmental data with the fishery data, and fitting the monthly catch per unit fishing effort and the environmental data based on a generalized additive model, and selecting environmental variables with dome-shaped fitting curves as interpretable environmental variables; (4) constructing a fishery forecast model based on an extreme gradient boosting tree, and using the monthly unit fishing effort catch and interpretable environmental variables to train the fishery forecast model for each month; (5) Based on the month to which the target date belongs, the high-yield area, medium-yield area, and low-yield area of ​​the Northwest Pacific squid fishery are obtained through spatial intersection and spatial superposition; (6) extracting the same environmental variables in the target date based on the interpretable environmental variables selected in the month to which the target date belongs, and inputting them into the fishing ground forecast model to quantitatively predict the central fishing ground suitable for the habitat of the Northwest Pacific squid.

2. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 1 is characterized by: The N is 6; the obtaining of the fishery data and the corresponding environmental data for N historical months is to obtain the fishery data and the corresponding environmental data of squid from June to November 1995 to 2022.

3. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 1 is characterized in that: The monthly catch per unit of fishing effort is as follows: Among them: CPUE ymij 、Catch ymij and EFFort ymij They represent the catch per unit fishing effort, catch volume and number of fishing vessels at longitude i, latitude j, month m and year y respectively.

4. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 1 is characterized in that: The step (3) is specifically: (3.1) Convert all environmental data into a grid that matches the historical fishery data to accommodate the spatial resolution of the fishery data for each month and match them based on the operation location and operation date; (3.2) fitting the monthly catch per unit fishing effort and the corresponding environmental data based on the generalized additive model; (3.3) A minimum threshold is set. In the fitting curve of the monthly unit fishing effort catch and the corresponding environmental data, when the monthly unit fishing effort catch of each month exceeds the set minimum threshold, the fitting curve takes a dome shape. This environmental variable is the screened explainable environmental variable of that month.

5. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 4 is characterized in that: The lowest threshold is represented by a percentile, that is, according to the data distribution, a specified percentage of data items is lower than this threshold; the dome shape shows a trend of first increasing and then decreasing.

6. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 1 is characterized by: In step (4), the parameters of the fishing ground forecast model are optimized by automatic tuning to minimize the mean square error (MSE) to evaluate the performance.

7. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 1 is characterized in that: The step (5) is specifically as follows: (5.1) Set three percentile thresholds corresponding to high-yield areas, medium-yield areas, and low-yield areas; (5.2) sorting the monthly catch per unit fishing effort corresponding to each interpretable environmental variable selected according to the month to which the target date belongs, extracting the monthly catch per unit fishing effort exceeding the set percentile threshold, and the concentration of the interpretable environmental variable corresponding to the monthly catch per unit fishing effort is the concentration range below the threshold; (5.3) extracting the same interpretable environmental variables in the target date according to the interpretable environmental variables selected from the month to which the target date belongs; (5.4) Under the same threshold, the spatial locations where each explanatory environmental variable is within its corresponding concentration range on the target date are screened out, and then the spatial intersection analysis is performed on the spatial locations of all the screened explanatory environmental variables, that is, the same spatial locations are retained, and the different spatial locations are discarded; (5.5) For the three percentile thresholds, the spatial locations of the high-yield area, medium-yield area, and low-yield area of ​​the Northwest Pacific squid fishery are obtained according to steps (5.2) to (5.4).

8. The interpretable intelligent forecasting method for the Northwest Pacific squid center fishing ground according to claim 7 is characterized by: The percentile thresholds corresponding to the high-yield area, the medium-yield area and the low-yield area are 80%, 60% and 40% respectively.

Images