«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

期数:

2025年02期

页码:

122-129

栏目:

人工智能

出版日期:

2025-02-10

文章信息/Info

Title:

WSN Data Drift Blind Calibration Algorithm Based on Improved BiLSTM-KF

文章编号:

1673-629X(2025)02-0122-08

作者:

史志强1; 2; 3; 古丽米拉·克孜尔别克1; 2; 3; 韩博1; 2; 3; 张瑛进1; 2; 3

1. 新疆农业大学 计算机与信息工程学院,新疆 乌鲁木齐 830052;

2. 智能农业教育部工程研究中心,新疆 乌鲁木齐 830052;

3. 新疆农业信息化工程技术研究中心,新疆 乌鲁木齐 830052

Author(s):

SHI Zhi-qiang1; 2; 3; GULIMILA Kezierbieke1; 2; 3; HAN Bo1; 2; 3; ZHANG Ying-jin1; 2; 3

1. School of Computer and Information Engineering,Xinjiang Agricultural University,Urumqi 830052,China;

2. Intelligent Agriculture Engineering Research Center of the Ministry of Education,Urumqi 830052, China;

3. Xinjiang Agricultural Informatization Engineering Technology Research Center,Urumqi 830052,China

关键词:

无线传感器网络; 数据漂移; 双向长短期记忆网络; 卡尔曼滤波器; 盲校准

Keywords:

wireless sensor networks; data drift; bidirectional long short-term memory networks; Kalman filters; blind calibration

分类号:

TP274.2

DOI:

10.20165/j.cnki.ISSN1673-629X.2024.0320

摘要:

无线传感器网络数据的准确性对智能系统的决策和环境监测的优化至关重要。 针对传感器设备在工作过程中易受外部环境变化及自身特性的影响产生数据漂移现象,导致数据精确度产生偏差的问题,该文提出一种基于 CNN-BiLSTM-Attention-KF 的无线传感器网络数据漂移校准方法。 首先,利用 CNN 提取数据的局部特征,BiLSTM 捕获时序数据的长期依赖关系,并引入 Attention 增强模型处理数据序列中的关键信息,根据节点数据的时空相关性进行建模,得到待校准节点的预测值。 其次,将其预测值与节点的实际观测值作为卡尔曼滤波器的输入,实现对漂移数据的跟踪和校准。 在公开数据集 IBRL 上进行实验,结果表明该方法在各评价指标有所改善,其中平均绝对误差(MAE)降至 0. 325 5,均方误差(MSE)降至 0. 228 9,相关系数(R2)达到 0. 988 2,均优于其他算法。 CNN-BiLSTM-Attention-KF 具有较好的校准效果,对于传感器在长时间工作中保持数据准确性具有重要意义。

Abstract:

The accuracy of wireless sensor network data is highly important for intelligent system decision-making and environmental mo-nitoring optimization. To counter the problem that sensor equipment is susceptible to the influence of external environmental alterations and its inherent attributes during the working process,subsequently resulting in the data drift phenomenon and the deviation of data accuracy,we proffer a data drift calibration approach based on CNN-BiLSTM-Attention-KF in wireless sensor networks. Initially,CNN was harnessed to extract the local traits of the data,BiLSTM seized the long-term dependency relationship of the time series data,and At-tention was incorporated to augment the model’s capacity to handle the key information in the data series. It was modelled in line with the spatiotemporal correlation of the node data,and the predicted value of the node requiring calibration was acquired. Subsequently,the predicted value and the actual observed value of the node were utilized as the input of the Kalman filter to monitor and calibrate the drift data. Experiments are carried out on the public dataset IBRL evince that the proposed method has manifested improvements in diverse e-valuation indicators. Among them,the mean absolute error (MAE) is lowered to 0. 325 5,the mean square error (MSE) is reduced to 0. 228 9,and the correlation coefficient (R2) ascends to 0. 988 2,all of which outshine other algorithms. The CNN-BiLSTM-Attention-KF methodology exhibits a superb calibration effect and holds great influence for the sensor to uphold data accuracy during prolonged working hours.

常用功能

更新日期/Last Update: 2025-02-10