Detección no supervisada de anomalías en series temporales de presión: un benchmark sintético de prueba de concepto en el contexto GasTrack
DOI:
https://doi.org/10.66104/j44hww62Palabras clave:
Detección de anomalías, Series temporales, Internet de las Cosas, Sensores de presión, Aprendizaje automáticoResumen
Este trabajo presenta un benchmark sintético de prueba de concepto para la detección no supervisada de anomalías en series temporales de presión en el contexto del sistema GasTrack. El estudio no pretende diagnosticar fugas: una anomalía de presión es un indicio de comportamiento inusual, compatible con fugas, variaciones del consumo, fallos de válvulas o defectos del propio sensor. Se compararon una regla estadística robusta, Isolation Forest, Local Outlier Factor (LOF) y One-Class SVM en ventanas de 24 lecturas separadas por 5 min (extensión temporal de 115 min, aproximadamente dos horas). El protocolo separa 5.000 ventanas normales de entrenamiento, 2.000 ventanas independientes de calibración del umbral (percentil 99 de las puntuaciones) y una prueba con 2.500 ventanas normales y 2.500 anómalas. El vector contiene once atributos de tendencia, dispersión, brusquedad y planicidad; no se utiliza ninguna variable duplicada. En 20 semillas, el LOF obtuvo un F1 de 0,9660 ± 0,0042, próximo a la regla robusta (0,9434 ± 0,0095). La diferencia absoluta de F1 fue de 0,0226 (IC del 95%: 0,0189–0,0263). La ventaja del LOF en las ventanas correspondientes a sensores bloqueados depende de la representación, especialmente del atributo de planicidad. Con una prevalencia del 1%, el valor predictivo positivo esperado del LOF cae a 0,485, por lo que el F1 de la prueba balanceada no debe interpretarse como desempeño en producción. Se utilizaron Python 3.12.10 y scikit-learn 1.7.2; el código del generador y de la evaluación se proporciona como material suplementario.Descargas
Referencias
BREUNIG, Markus M.; KRIEGEL, Hans-Peter; NG, Raymond T.; SANDER, Jörg. LOF: identifying density-based local outliers. In: ACM SIGMOD INTERNATIONAL CONFERENCE ON MANAGEMENT OF DATA, 2000, Dallas. Proceedings [...]. New York: ACM, 2000. p. 93-104. DOI: https://doi.org/10.1145/335191.335388. DOI: https://doi.org/10.1145/342009.335388
EL BARKANI, Majda; BENAMAR, Nabil; TALEI, Hanae; BAGAA, Miloud. Gas Leakage Detection Using Tiny Machine Learning. Electronics, [S. l.], v. 13, n. 23, art. 4768, 2024. DOI: https://doi.org/10.3390/electronics13234768. DOI: https://doi.org/10.3390/electronics13234768
LIU, Fei Tony; TING, Kai Ming; ZHOU, Zhi-Hua. Isolation Forest. In: IEEE INTERNATIONAL CONFERENCE ON DATA MINING, 8., 2008, Pisa. Proceedings [...]. [S. l.]: IEEE, 2008. p. 413-422. DOI: https://doi.org/10.1109/ICDM.2008.17. DOI: https://doi.org/10.1109/ICDM.2008.17
SCHÖLKOPF, Bernhard et al. Estimating the Support of a High-Dimensional Distribution. Neural Computation, [S. l.], v. 13, n. 7, p. 1443-1471, 2001. DOI: https://doi.org/10.1162/089976601750264965. DOI: https://doi.org/10.1162/089976601750264965
SOUZA, Paulo Silas Severo de et al. Detecting abnormal sensors via machine learning: An IoT farming WSN-based architecture case study. Measurement, [S. l.], v. 164, art. 108042, 2020. DOI: https://doi.org/10.1016/j.measurement.2020.108042. DOI: https://doi.org/10.1016/j.measurement.2020.108042
TAN, Qiong et al. A new sensor fault diagnosis method for gas leakage monitoring based on the naive Bayes and probabilistic neural network classifier. Measurement, [S. l.], v. 194, art. 111037, 2022. DOI: https://doi.org/10.1016/j.measurement.2022.111037. DOI: https://doi.org/10.1016/j.measurement.2022.111037
TSOUKAS, Vasileios et al. A Gas Leakage Detection Device Based on the Technology of TinyML. Technologies, [S. l.], v. 11, n. 2, art. 45, 2023. DOI: https://doi.org/10.3390/technologies11020045. DOI: https://doi.org/10.3390/technologies11020045
YUAN, Hongyong et al. Real-time detection of urban gas pipeline leakage based on machine learning of IoT time-series data. Measurement, [S. l.], v. 242, part B, art. 115937, 2025. DOI: https://doi.org/10.1016/j.measurement.2024.115937. DOI: https://doi.org/10.1016/j.measurement.2024.115937
Publicado
Número
Sección
Licencia
Derechos de autor 2026 João Cláudio Nunes Carvalho, Daniel Alencar Barros Tavares, Ronaldo Tadeu Pontes Milfont

Esta obra está bajo una licencia internacional Creative Commons Atribución 4.0.
Authors who publish in this journal agree to the following terms:
Authors retain copyright and grant the journal the right of first publication, with the work simultaneously licensed under the Creative Commons Attribution License, which permits the sharing of the work with proper acknowledgment of authorship and initial publication in this journal;
Authors are authorized to enter into separate, additional agreements for the non-exclusive distribution of the version of the work published in this journal (e.g., posting in an institutional repository or publishing it as a book chapter), provided that authorship and initial publication in this journal are properly acknowledged, and that the work is adapted to the template of the respective repository;
Authors are permitted and encouraged to post and distribute their work online (e.g., in institutional repositories or on their personal websites) at any point before or during the editorial process, as this may lead to productive exchanges and increase the impact and citation of the published work (see The Effect of Open Access);
Authors are responsible for correctly providing their personal information, including name, keywords, abstracts, and other relevant data, thereby defining how they wish to be cited. The journal’s editorial board is not responsible for any errors or inconsistencies in these records.
PRIVACY POLICY
The names and email addresses provided to this journal will be used exclusively for the purposes of this publication and will not be made available for any other purpose or to third parties.
Note: All content of the work is the sole responsibility of the author and the advisor.
