摘要
[目的]为了解决现有意图预判方法存在的准确性较低、计算时间较长等问题,尤其在处理“新意图”时,其在原有任务上的性能通常会显著下降。因此,需要研究面向水面目标意图预判的可增智能方法。[方法]采用长短期记忆网络(LSTM)对同一水面目标的状态信息数据集进行训练;通过提取水面目标轨迹数据的特征,分析水面目标运动信息与意图的关联策略;提出基于支持向量机(SVM)的水面目标意图动态预判算法。当目标出现新意图时,该算法依据预设的触发条件,对SVM分类器进行实时优化,进而实现对水面目标意图的增量式预判。[结果]试验结果表明,动态预判模型的预测误差降低了15.52%,意图预判的准确率提高了22.55%。[结论]基于LSTM-SVM的可增智能学习技术不仅能够提高水面目标意图预判的准确率,而且能不断“扩增”新的水面目标意图类型。
[Purpose]To address the limitations of existing intention prediction methods,such as reduced accuracy and prolonged computation times,especially when confronted with"new intentions,"where performance on original tasks typically degrades significantly,it is necessary to develop an incremental learning method for surface target intention prediction.[Method]The long short-term memory(LSTM)networks is used to train on the state information datasets from identical water surface targets.By extracting features from the trajectory data of water surface targets,this study analyzes the correlation strategy between water surface targets movement information and intentions.A dynamic intention prediction algorithm for water surface targets based on support vector machine(SVM)is proposed.When a target exhibits a new intention,the algorithm performs real-time optimization on the SVM classifier according to preset trigger conditions,thereby realizing the incremental prediction of the water surface target's intention.[Result]Experimental results demonstrate a 15.52%reduction in prediction error for the dynamic intention prediction model and a 22.55%increase in the accuracy of intention prediction.[Conclusion]The LSTM-SVM-based incremental intelligent learning technique not only enhances the precision of intention prediction for water surface targets,but also continuously"expands"the types of new water surface target intentions.
作者
史岳橙
李军
高睿
李梅
SHI Yuecheng;LI Jun;GAO Rui;LI Mei(Key Laboratory of Marine Intelligent Equipment and System of Ministry of Education,Shanghai Jiao Tong University,Shanghai 200240,China;Marine Design&Research Institute of China,Shanghai 200011,China;Shanghai Marine Electronic Equipment Research Institute,Shanghai 201108,China)
出处
《船舶工程》
北大核心
2025年第11期126-136,152,共12页
Ship Engineering
基金
上海市浦江人才项目(22PJ1405400)
国家自然科学基金青年科学基金项目(C类)(52301402)。
关键词
水面目标
运动状态预测
意图预判
可增智能
water surface target
movement state prediction
intention prediction
incremental intelligent
