计算机科学 ›› 2024, Vol. 51 ›› Issue (3): 63-71.doi: 10.11896/jsjkx.221100012
周琳茹, 彭鹏菲
ZHOU Linru, PENG Pengfei
摘要: 合理的任务分析可帮助分析者快速、准确地进行任务规划,目前使用案例推理方法进行任务分析存在分析时间长、分析结果准确性较低等问题。针对该问题,提出了基于注意力-生成对抗网络的任务分析方法。以长短时记忆网络(LSTM)为生成器、循环神经网络(RNN)为判别器,针对离散数据细微梯度的更新无法回传的问题,在生成器中使用rollout policy对生成的不完整序列进行推理补充,在判别器中使用蒙特卡罗(MC)进行数据采样得到完整的数据序列动作价值函数,从而指导生成器的参数更新;针对稀疏数据特征不明显、数据重点不明确等问题,在生成对抗网络训练前加入软注意力机制,为不同特征赋予不同权重从而过滤冗余数据,筛选出重要的特征数据。将该方法与未加入注意力机制的生成对抗网络在同一模拟数据集上进行对比实验,结果表明,加入注意力机制后的方法在精确率(P)、召回率(R)、F1值和准确率(Accuracy)4种评价指标上分别提升了0.088,0.092,0.094和0.068,与其他神经网络推荐算法相比,在P,R,F1值和Accuracy上分别提升了0.1~0.3,0.1~0.2,0.1~0.25和0.07~0.17,证明了该方法的有效性。
中图分类号:
[1]WU T.Research on optimization method of joint operation plan based on cloud simulation[D].China Electronics Technology Group Corporation,2020. [2]SHAO T H,CHEN H H,SHU Z,et al.Mission IntelligentParsing for Unmanned Combat Command and Control[J].Journal of Command and Control,2021,7(2):146-152. [3]QIAN J,LIU Y,LIU C,et al.Study on case analysis and scena-rio deduction based on multi-dimensional scenario space method[J].Systems Engineering-Theory & Practice,2015,35(10):2588-2595. [4]HUANG X P,ZHOU X Z,LU X M,et al.The Comprehension and Cognition to the Evolution Trend of Decision System From the Angle of Human-Oriented—the Construction of Human-Computure Coordination Decision System[J].Chinese Journal of Systems Science,2013,21(1):35-39. [5]PENG Y F,YANG F D.Formal Decription and Decomosition of Space Missions[J].Military Operations Research and Systems Engineering,2021,35(4):62-67. [6]WANG J Y,WANG J T.A Study of Emergency Decision Support System Based on Case-Based Reasoning[J].Journal of Management Science,2003(6):46-51. [7]CHE H Y,GAN R C.Research of CBR and RBR Integrated Method in Architecture Design of Information System[J].Application Research of Computers,2004(3):81-84. [8]DING X C.Research of CBR Software Enterprise KnowledgeManagement and System Development[D].Liaoning Technical University,2004. [9]LAI M.Giraffe:Using Deep Reinforcement Learning to PlayChess[J].arXiv:1509.01549,2015. [10]WU M X.Study of the Method for Solving Complex Problems Based on CBR[D].Shandong University of Science and Techno-logy,2005. [11]CHO K,VAN MERRIËNBOER B,GULCEHRE C,et al.Learning phrase representations using RNN encoder-decoder for statistical machine translation[J].arXiv:1406.1078,2014. [12]SRIVASTAVA N,HINTON G,KRIZHEVSKY A,et al.Dropout:a simple way to prevent neural networks from overfitting[J].Journal of Machine Learning Research,2014,15(1):1929-1958. [13]DIMITRI B.Multiagent Reinforcement Learning:Rollout andPolicy Iteration[J].IEEE/CAA Journal of Automatica Sinica,2021,8(2):249-272. [14]LAI M.Giraffe:Using Deep Reinforcement Learning to Play Chess[J].arXiv:1509.01549,2015. [15]KINGMA D P,BA J.Adam:A method for stochastic optimization[J].arXiv:1412.6980,2014. [16]SRIVASTAVA N,HINTON G,KRIZHEVSKY A,et al.Dropout:a simple way to prevent neural networks from overfitting[J].Journal of Machine Learning Research,2014,15(1):1929-1958. |
|