上海交通大学学报

上海交通大学学报（自然版）

• 生物医学工程 • 上一篇    下一篇

随机协同分解PSO优化的Takagi-Sugeno模糊神经网络临床路径变异处理

杜刚1，江志斌1，刁晓娣2,姚阳3
  

  1. （1.上海交通大学 工业工程与物流工程系， 上海 200240； 2.上海市普陀区中心医院， 上海 200062；3.上海市第六人民医院， 上海 200233）
  • 收稿日期:2009-10-15 修回日期:1900-01-01 出版日期:2010-08-31 发布日期:2010-08-31

Variances Handling for Clinical Pathway Based on Takagi-Sugeno FNNs with Random Cooperative Decomposing PSO Optimization

DU Gang1,JIANG Zhibin1,DIAO Xiaodi2,YAO Yang3
  

  1. (1.Department of Industrial Engineering & Logistics Management, Shanghai Jiaotong University, Shanghai 200240, China; 2.Shanghai Putuo District Central Hospital, Shanghai 200062, China; 3.Shanghai No.6 People’s Hospital, Shanghai 200233, China)
  • Received:2009-10-15 Revised:1900-01-01 Online:2010-08-31 Published:2010-08-31

摘要： 提出了一种随机协同分解粒子群（RCDPSO）优化的TakagiSugeno（TS）模糊神经网络临床路径变异处理方法.在子种群的协同进化过程中，执行顺序随机确定后，选取表现最优的粒子进行分解，对表现最差的粒子进行交叉和变异，并保持子种群的总数不变，既保证了收敛速率，又增加了种群的全局搜索能力.在此基础上，加入了变异扰动机制，增加了种群的多样性，防止种群陷入局部最优.最后以骨肉瘤术前化疗临床路径变异（肝中毒）为例，进行实例验证.结果表明，在处理临床路径变异方面，RCDPSO优化的TS模糊神经网络与其他算法优化的TS模糊神经网络相比，预测能力较强、鲁棒性更佳，大幅度提高了临床路径变异处理的精度和效率.

关键词: 临床路径, TS模糊神经网络, 随机协同分解粒子群优化, 卡尔曼滤波算法, 变异, 骨肉瘤

Abstract: A variances handling method for clinical pathway was proposed, which is based on TakagiSugeno (TS) fuzzy neural networks (FNNs) with random cooperative decomposing particle swarm optimization (RCDPSO). During the process of cooperative coevolution with random execution sequence, a decomposing algorithm was adopted for the particles with the highest performance, and crossover and mutations were adopted for the particles with the worst performance. Therefore, it not only ensures the convergence rate, but also improves performance of the algorithm in global search. Moreover, the variation disturbing mechanism was introduced to strengthen the diversity of population and avoid plunging into local optimum. Finally, a case study on liver poisoning of osteosarcoma preoperative chemotherapy was used to validate the proposed method. The result demonstrates that TS FNNs based on the RCDPSO achieves superior performance in prediction and robustness to TS FNNs based on other algorithms, which makes variances handling of clinical pathway more effective.

中图分类号: