基于自注意力的无监督室内定位信号异常检测

doi:10.11959/j.issn.1000-0801.2023258

摘要/Abstract

摘要：

基于射频信号的室内指纹定位技术，以其精度高、部署成本低等优点被广泛运用于室内定位领域。室内信号环境的变化会直接影响定位精度。深度神经网络也用于时序数据异常检测，在此基础上，提出了一种基于自注意力机制的无监督室内定位信号异常检测模型。训练模型的输入是易获得的正常信号环境下无位置标签的指纹数据。该模型的注意力模块关注提取指纹数据中不同信号来源之间的相互关联，结合关联误差和重构误差来放大正常与异常的可区分性，从而提升室内定位信号检测的精度。在实验室采集的蓝牙信号数据集和一个公开的Wi-Fi数据集UJIIndoorLoc中进行性能评估，实验结果表明，与其他算法相比，所提模型具有最好的异常检测性能。

关键词: 异常检测, 指纹定位, 自注意力, 无监督

Abstract:

Indoor fingerprint location technology based on radio signal is widely used in the field of indoor location because of high accuracy and low deployment cost.The change of indoor signal environment will directly affect the positioning accuracy.The deep neural network is also used in anomaly detection of time series data.On this basis, an unsupervised indoor location signal anomaly detection model based on self-attention mechanism was proposed.The input of model is normal fingerprint data that can be easily obtained without location tags.The attention module of the model focuses on extracting the correlation between different signal sources in the fingerprint data.It amplifies the distinguishability between normal and abnormal by combining the association errors and reconstruction errors, thus improving the accuracy of indoor location signal detection.The performance of the proposed model was evaluated in a bluetooth signal dataset collected in the laboratory and a public Wi-Fi dataset named UJIIndoorLoc.The experimental results show that the proposed model has the best anomaly detection performance compared to other algorithms.

Key words: anomaly detection, fingerprint location, self-attention, unsupervised

中图分类号:

TP399

袁江华, 艾浩军. 基于自注意力的无监督室内定位信号异常检测[J]. 电信科学, 2023, 39(12): 42-52.

Jianghua YUAN, Haojun AI. Unsupervised anomaly detection of indoor location signals based on self-attention[J]. Telecommunications Science, 2023, 39(12): 42-52.

图/表 14

图1

图2

图3

图4

表1

表2

表3

表4

表5

表6

表7

图5

图6

表8

参考文献 23

[1]	ZAFARI F , GKELIAS A , LEUNG K K . A survey of indoor localization systems and technologies[J]. IEEE Communications Surveys ＆ Tutorials, 2019,21(3): 2568-2599.
[2]	JI T , LI W H , ZHU X C ,et al. Survey on indoor fingerprint localization for BLE[C]// Proceedings of 2022 IEEE 6th Information Technology and Mechatronics Engineering Conference (ITOEC). Piscataway:IEEE Press, 2022: 129-134.
[3]	SUBHAN F , KHAN A , SALEEM S ,et al. Experimental analysis of received signals strength in bluetooth low energy (BLE) and its effect on distance and position estimation[J]. Transactions on Emerging Telecommunications Technologies, 2022,33(2): e3793.
[4]	YIU S , DASHTI M , CLAUSSEN H ,et al. Wireless RSSI fingerprinting localization[J]. Signal Processing, 2017,131: 235-244.
[5]	PINTO B , BARRETO R , SOUTO E ,et al. Robust RSSI-based indoor positioning system using K-means clustering and Bayesian estimation[J]. IEEE Sensors Journal, 2021,21(21): 24462-24470.
[6]	RUFF L , VANDERMEULEN R , GOERNITZ N ,et al. Deep one-class classification[C]// International conference on machine learn-ing. PMLR, 2018: 4393-4402.
[7]	GUPTA M , GAO J , AGGARWAL C ,et al. Outlier detection for temporal data[J]. Synthesis Lectures on Data Mining and Knowledge Discovery, 2014,5(1): 1-129.
[8]	AMER M , GOLDSTEIN M , ABDENNADHER S . Enhancing one-class support vector machines for unsupervised anomaly detection[C]// Proceedings of the ACM SIGKDD Workshop on Outlier Detection and Description. New York:ACM Press, 2013: 8-15.
[9]	ALGHUSHAIRY O , ALSINI R , SOULE T ,et al. A review of local outlier factor algorithms for outlier detection in big data streams[J]. Big Data and Cognitive Computing, 2020,5(1): 1.
[10]	ZENATI H , FOO C S , LECOUAT B ,et al. Efficient gan-based anomaly detection[J]. arXiv preprint, 2018,arXiv:1802.06222.
[11]	ZONG B , SONG Q , MIN M R ,et al. Deep autoencoding gaussian mixture model for unsupervised anomaly detec-tion[C]// International conference on learning representations. 2018.
[12]	HE S N , JI B , CHAN S H G . Chameleon:survey-free updating of a fingerprint database for indoor localization[J]. IEEE Pervasive Computing, 2016,15(4): 66-75.
[13]	YANG J H , ZHAO X H , LI Z . Crowdsourcing indoor positioning by light-weight automatic fingerprint updating via ensemble learning[J]. IEEE Access, 2019,7: 26255-26267.
[14]	AI H J , HU T , XU T S . RAD-GAN:radio map anomaly detection for fingerprint indoor positioning with GAN[C]// Proceed ings of 2021 International Conference on Indoor Positioning and Indoor Navigation (IPIN). Piscataway:IEEE Press, 2022: 1-8.
[15]	VASWANI A , SHAZEER N , PARMAR N ,et al. Attention is all you need[J]. Advances in neural information processing systems, 2017:30.
[16]	VOITA E , TALBOT D , MOISEEV F ,et al. Analyzing multi-head self-attention:specialized heads do the heavy lifting,the rest can be pruned[J]. arXiv preprint, 2019,arXiv:1905.09418.
[17]	TULI S , CASALE G , JENNINGS N . TranAD:deep transformer networks for anomaly detection in multivariate time series data[J]. Proc VLDB Endow, 2022(15): 1201-1214.
[18]	ZHANG H W , XIA Y Q , YAN T J ,et al. Unsupervised anomaly detection in multivariate time series through transformer-based variational autoencoder[C]// Proceedings of 2021 33rd Chinese Control and Decision Conference (CCDC). Piscataway:IEEE Press, 2021: 281-286.
[19]	CHEN Z K , CHEN D S , ZHANG X ,et al. Learning graph structures with transformer for multivariate time-series anomaly detection in IoT[J]. IEEE Internet of Things Journal, 2022,9(12): 9179-9189.
[20]	XU J H , WU H X , WANG J M ,et al. Anomaly transformer:Time series anomaly detection with association discrepancy[J]. arXiv preprint, 2021,arXiv:2110.02642.
[21]	WANG B , ZHOU S L , LIU W Y ,et al. Indoor localization based on curve fitting and location search using received signal strength[J]. IEEE Transactions on Industrial Electronics, 2015,62(1): 572-582.
[22]	MENDOZA-SILVA G , RICHTER P , TORRES-SOSPEDRA J , ,et al. Long-term WiFi fingerprinting dataset for research on robust indoor positioning[J]. Data, 2018,3(1): 3.
[23]	LIN T Y , WANG Y X , LIU X Y ,et al. A survey of transformers[J]. AI Open, 2022(3): 111-132.

对比项	True	False
Positive	TP	FP
Negative	TN	FN

实验条件	设置
a0	正常条件，所有蓝牙发射功率为0，并安装在固定位置
a1	只将1号蓝牙发射功率设置为-8 dB
a2	只将1号蓝牙发射功率设置为-12 dB
a3	只将1号蓝牙发射功率设置为-16 dB
b1	只将1号蓝牙信标移动到图3所示的B3处
b2	只将1号蓝牙信标移动到图3所示的B2处
b3	只将1号蓝牙信标移动到图3所示的B3处
c	经过检查将蓝牙信标的信号强度和位置重新恢复到正常条件下

模型	a0
GBDT	0.38
LOF	0.23
RAD-GAN	0.20
OCSVM	0.12
LOC-ATTEN	0.12

模型	a1	a2	a3
GBDT	0.44	0.43	0.35
LOF	0.43	0.32	0.25
RAD-GAN	0.41	0.30	0.21
OCSVM	0.45	0.33	0.17
LOC-ATTEN	0.33	0.23	0.15

模型			a1					a2					a3
模型	U ₁	U ₂	U ₃	U ₄	U ₅	U ₁	U ₂	U ₃	U ₄	U ₅	U ₁	U ₂	U ₃	U ₄	U ₅
GBDT	0.16	0.35	0.42	0.56	0.70	0.22	0.24	0.40	0.57	0.71	0.12	0.18	0.34	0.49	0.64
LOF	0.14	0.31	0.38	0.52	0.78	0.10	0.12	0.25	0.48	0.66	0.07	0.10	0.16	0.36	0.56
RAD-GAN	0.16	0.25	0.32	0.57	0.74	0.07	0.07	0.23	0.49	0.64	0.05	0.05	0.09	0.32	0.51
OCSVM	0.10	0.16	0.39	0.75	0.85	0.05	0.06	0.31	0.45	0.78	0.02	0.03	0.07	0.29	0.45
LOC-ATTEN	0.11	0.19	0.31	0.41	0.62	0.06	0.07	0.15	0.33	0.55	0.04	0.05	0.06	0.23	0.37