Research On The Resource Abundance Forecast And Fishing Ground Distribution Of South Pacific Albacore Based On Machine Learning

Albacore tuna(Thunnus alalunga)is widely distributed in the mid-latitude waters of the three oceans.In recent years,China’s fishing vessels in the South Pacific Ocean have increased a lot,and South Pacific albacore tuna has become one of the important fishing objects of China’s pelagic longline fishery.In order to better arrange fishing operations,improve fishing efficiency and reduce the cost of looking for fishing grounds,it is particularly important to improve the accuracy of fishing ground forecast.Since fisheries production data are difficult to obtain and the collection process is susceptible to local climate and policies,resulting in more samples containing missing values in fisheries production data and difficult to effectively extract features from marine environmental data,which adds great difficulty to fishing forecasting.With the development of artificial intelligence,data mining,deep learning,and artificial neural networks based on probability and statistics theories have been applied in many fields,A variety of machine learning methods have also been applied to fishery forecast,which has better effect than the traditional fishery forecast model using statistical principles.The use of machine learning methods for fishery forecasting can better compensate for the problems that traditional forecasting methods are difficult to effectively fit highdimensional marine environmental data and are vulnerable to missing values of fishery production data.The experimental sea area in this study ranges from 110°E-135°W,5°S-40°S.Since the spatial and temporal scales of each environmental factor data are different,the environmental data must be resampled on spatial and temporal scales after replenishing the missing values,and the data are processed by python programming and converted to the same spatial and temporal resolution as the fishery data as the new experimental data.In this study,catch per unit effort(CPUE)was used as the resource abundance index of albacore tuna,and a fishing area of 5°×5° was used.There are two main aspects of this study as follows.(1)To improve the prediction accuracy of fishery resource abundance,CPUE is used as the index of albacore tuna resource abundance,and the marine environmental factors obtained from marine remote sensing and Argo are used to adopt Convolutional Neutral Network(CNN)based on the optimal distributed decision gradient boosting tree(XGBoost)model Based on the XGBoost model,we proposed an improved XGBoost model CNN-SA-XGBoost model by using the Simulate Anneal(SA)algorithm for feature extraction of high-dimensional marine environmental data and hyperparameter optimization of the optimal distributed decision gradient boosting tree(XGBoost)model.XGBoost model to achieve the regression prediction of albacore tuna resource abundance in the South Pacific Ocean.Experiments showed that the CNN-SA-XGBoost model predicted the abundance of albacore tuna in the South Pacific Ocean with a root mean square error of 0.486,which was 12.4% less than that of XGBoost,and 11% less than that of Multiple Linear Regression,Random Forest(RF),BP The prediction error is 11.8%-28.4% lower than that of Multiple Linear Regression,Random Forest(RF),and BP neural network models.Meanwhile,the efficiency of the simulated annealing algorithm optimization algorithm is much higher than that of grid search tuning,and the time cost is reduced by 93.4%.And the improved XGBoost model improves to a certain extent the problem of large prediction error when the traditional resource abundance prediction model faces high-dimensional environmental data and fishery production data with many missing values,and provides a new method for pelagic fishery forecasting.(2)In terms of exploring the spatial and temporal distribution of potential fisheries,the Maximum entropy model(Max Ent)and GIS(geographic information system,GIS)were used in this study to analyze the changes of the potential distribution of albacore tuna in the South Pacific on the spatial and temporal scales.Mass to use Arc GIS secondary development in the first place,target area grid image of each environmental factor by Max Ent models to predict the potential distribution prediction results of albacore tuna,and combined with Arc GIS to draw its potential habitat,distribution according to the potential distribution analysis of the months in the migration direction aim months.Secondly,the importance of each environmental factor in the model was evaluated using the Jackknife method,and the range of environmental factors suitable for albacore tuna was analyzed according to the response curve to provide theoretical guidance for the fishing forecast.This study proposes an improved XGBoost model to predict the abundance of albacore tuna resources in the South Pacific Ocean,which not only improves the accuracy of model prediction,but also reduces the time cost of model prediction,and provides a new idea for the fishing forecast in other seas.Secondly,in order to further analyze the distribution of albacore tuna fisheries in the South Pacific from the perspective of time and space,the Max Ent model is also established to explore the potential distribution of albacore tuna during the peak fishing season and to grasp the spatial and temporal dynamics of the fishery.