大数据中面向乱序数据的改进型BP算法

针对大数据中的乱序数据缺少关联规则的问题，提出了一种动态调整的改进型BP 算法，运用了动态自适应结构调整机制，根据环境要求自适应调整网络训练结构，自动删除无效训练神经元，优化迭代训练过程；并在网络学习过程中动态调整网络参数中的三因子，即动量因子、权学习指数、阈学习指数，来达到加快学习响应速度、增强网络稳定性的目的. 仿真结果表明，通过动态自适应调整结构、动态调整三因子的改进型算法，能够获得更多的收敛次数，并能有效地提高收敛率，进而提高整体网络性能.

An out-of-order data streams oriented BP algorithm for association rules of big data

Because of the difficulty of getting the association rules over out-of-order streams for big data, this paper presents an improved BP algorithm based on dynamical adjustment. In this algorithm, a dynamic adaptive structural is settled according to environmental requirements, which can automatically remove invalid training neurons, and optimize the iterative training process. Furthermore, the algorithm dynamically adjusts three factors (i.e. momentum factor, right learning index and threshold learning factor) during the learning process to speed up the response time and enhance the stability of the network. In the simulation, three factors are changed by a self-adapting adjusting mechanism. By comparing with the traditional BP algorithm, it gets more convergence times and convergence rate. Results also indicate that the proposed algorithm can obviously improve efficiency and finally obtain the association rules over out-of-order data streams.