基于北斗船位数据的拖网渔船捕捞努力量算法研究

Research on fishing effort algorithm of trawler based on Beidou Satellite navigation data

渔船捕捞作业行为识别算法准确率直接影响捕捞努力量的估算精度,现有的捕捞作业行为识别算法存在特征局限及算力要求高等问题。基于2019—2021年辽宁省拖网渔船的北斗船位数据计算捕捞努力量,构造含空间信息的特征向量和基于极限梯度提升算法的拖网渔船捕捞作业行为识别模型。结果表明,模型的准确率、敏感性、特异性和马修斯相关系数分别为96.47%、99.58%、95.02%和0.9234。捕捞努力量平均绝对误差为0.1009 kW·h,均方根误差为0.9851 kW·h。研究证实,提出的捕捞努力量算法精度较高,可为渔业资源评估和限额捕捞管理提供基础数据支撑。

Fishing effort refers to the amount of work invested in a fishing ground using a particular fishing method over a defined period of time.It is a pivotal reference for managing fishery production,evaluating fishery resources,and assessing ecological vulnerability.Accurately identifying and evaluating the temporal and spatial characteristics of fishing effort is essential for developing effective fishing quota strategies and assessing fishery resources.Vessel monitoring system(VMS)offers a fishing vessel monitoring mechanism capable of obtaining dynamic information as fishing vessel position,speed,and transmission time.Methods based on VMS data enlarge the scope of data sources available for fishery scientific research and introduce novel approaches to estimate fishing effort.This article proposed a method for estimating fishing effort based on Beidou Satellite navigation data to distinguish trawling fishing behavior.The experimental data was Beidou Satellite navigation data for fishing vessels in Liaoning Province,with a total of 853 fishing vessels and 48256060 satellite navigation data records,collected from September 2019 to January 2022.The investigation area was the sea area between 37°-40°N and 119°-124°E,including parts of the Bohai Sea and the Yellow Sea.The satellite navigation data of each vessel included latitude,longitude,speed,transmission time,and fishing area information.During the data preprocessing process,data with missing,incomplete,or out-of-range longitude,latitude,speed,and time information were removed based on the overall data range.The data segments were divided based on the entry and exit time of the fishing port.As the satellite reception process might be affected by signal fluctuations,and there might be human factors such as obstruction of the transmission source,if the time interval between two reporting positions exceeded 3 h,the trajectory segment would be split,and trajectory segments with less than 5 data points would be deleted.The vessel position data for 12 trawling fishing vessels in Liaoning Province from September 2019 to January 2022 were calibrated.The important parameters of fishing vessel operating characteristics were extracted from the calibrated data,including time interval,distance,the shortest distance to the coastline,theoretical speed,current speed,time,and month,to construct the feature vector.The XGBoost algorithm was used to train the trawling fishing behavior recognition model,and 60362 randomly selected vessel position data were used as the experimental dataset.After selecting the optimal hyperparameters with five-fold cross-validation and completing the model training,external validation was performed using 114734 vessel position records.Extreme learning machine and random forest were selected for comparison with XGBoost to test the generalization ability of the algorithms.The experimental results showed that the accuracy,sensitivity,specificity,and Matthews correlation coefficient of the model were 96.47%,99.58%,95.02%,and 0.9234,respectively.The mean absolute error and root mean square error of the fishing effort were 0.1009 kW·h and 0.9851 kW·h,respectively.In order to further improve the accuracy of the fishing behavior recognition model and mitigate the error associated with fishing effort,three aspects can be considered:feature optimization,hyperparameter setting,and expansion of the data scale.Regarding feature optimization,while this article has introduced the shore distance as spatial information,fishing boat operations possess greater spatial complexity.Therefore,effective spatial information between ships can be extracted to serve as feature parameters and improve the accuracy of the model.Concerning the optimization of hyperparameter settings,this article has employed cross-validation and grid search methods to determine the hyperparameters.However,relevant literatures suggest that heuristic algorithms such as gray wolf optimization,egret swarm optimization algorithm,and fruit fly optimization algorithm can also be utilized to optimize the selection of hyperparameters.Moving forward,appropriate heuristic algorithms can be selected to optimize the hyperparameter selection of XGBoost and bolster the accuracy of the model.Additionally,as the annotation of ship data requires a considerable amount of effort,this article has selected a limited amount of trawler data.It is recommended that the data volume should be expanded.The research methods and results can provide important data support for fishery resources assessment.