基于随机权神经网络集成模型的实时遥测数据处理  

An Ensemble Based on Neural Networks with Random Weights for Real-timeTelemetry Data Processing

作　　者：贾海艳[1] Jia Haiyan(Unit 92941 Element of the PLA,Huludao 125000,China)

机构地区：[1]中国人民解放军92941部队,辽宁葫芦岛125000

出　　处：《计算机测量与控制》2020年第6期52-55,共4页Computer Measurement &Control

摘　　要：飞行任务中的遥测数据是快速产生的时间序列数据流,其受测量设备和空间环境等因素影响易产生数据的漂移,由于过程进化属性,其数据分布属性也会发生变化,传统单一数据预测模型无法反应数据自身特征属性的这一变化;因此,提出一种联合具有随机权重的神经网络和装袋算法的集成方法实现对遥测数据的在线回归预测,设计的算法能根据数据特征属性变化而进行自主更新;利用基模型的多样性和低训练复杂度,同时满足数据处理的精度和实时性要求;通过实验仿真,结果表明该方法能明显抑制遥测数据的漂移现象,数据的预测精度提高近10m。Telemetry data of flight task are time-series data streams sequentially and rapidly.Data drift Influenced by measurement devices and spatial environment is produced,the evolving nature of processes may often cause changes of data distribution,which is difficult to detect and causes loss of accuracy in data prediction accuracy.a single forecast models can t adapt the change of data feature attribute.As a consequence,an ensemble model was put off that combine neural networks with random weights algorithms and bagging algorithm,it is able to update actively according to possible changes in the data distribution.By use of the diversity of base model,low training complexity and dynamic updating mechanisms,it is a accurate algorithm that can operate in a computational time.Through experimental simulation,the results show that the method can obviously restrain the drift of telemetry data,and the prediction accuracy of the data is improved by nearly 10 m.

关 键 词：遥测数据 集成 神经网络 随机权值 优化 

分 类 号：TP2[自动化与计算机技术—检测技术与自动化装置]