The industrial control anomaly detection algorithm effectively monitors and identifies anomalies in industrial control systems, ensuring timely issue detection to maintain system reliability, safety, and regular operation. Researchers often integrate a large number of high-precision, high-sampling-frequency sensors into the system to obtain comprehensive real-time equipment operating data. However, deploying consumer-grade high-performance computing devices in large-scale industrial control systems is often impractical, rendering existing anomaly detection systems unusable in practical environments. Most existing sensor feature selection algorithms are unsuitable for industrial control time-series datasets, and some suffer from issues such as a limited reduction in the number of sensors and a decrease in anomaly detection accuracy after reduction. We propose an offline industrial control anomaly detection algorithm based on the maximal information coefficient and hierarchical clustering-deep neural network (MICHC-NN) to address these challenges. We utilize the maximum information coefficient to compute correlations among sensors, addressing the issue of existing feature selection algorithms relying on anomaly labels. In addition, we introduce a key-sensor selection algorithm based on correlation, identifying interrelated sensors in an industrial control system and determining the key sensors most susceptible to system anomalies. These key sensors allow the anomaly detection algorithm to be deployed with a smaller footprint in real industrial control environments. Finally, we employ an autoencoder–decoder model with attention mechanisms to train the anomaly detection network. Experimental results demonstrate that our proposed anomaly detection algorithm achieves performance comparable to using all sensors, with a 75% reduction in the number of sensors employed.

Our work quantifies how much sensor placement affects emission-rate inversion for candidate urban facilities and road segments relative to other design decisions, given that monitoring networks are sited for concentration mapping. We built a paired benchmark covering 30 synthetic cities that places five design decisions on one error scale. Under matched transport, covering traffic in the source space changes normalized source error by 0.74, and dictionary fidelity and sensor calibration by about 0.68 each, against 0.13–0.15 across placement criteria; under an independent stability-dependent dispersion generator, the coverage effect shrinks to the order of the placement effect. Across five error types at matched severity, source identification degrades more slowly than emission rates, with similar patterns for every tested type. Unmodeled mixing height is among the most common errors (0.41–0.53), and an accurate external estimate can recover much of it. A truth-free residual flags severe model errors of every tested type (AUC 0.89–0.95). For seven Chengdu NO₂ stations, poor out-of-sample skill, unstable support, and a large residual led us to withhold estimates. Where major inventory omissions or calibration errors exist, correcting them may reduce error more than switching among the placement criteria tested; identification capacity, rates, and residuals should be reported separately.

Distributed optical fiber sensing (DOFS) based on phase-sensitive optical time-domain reflectometry (ϕ-OTDR) is increasingly used for perimeter and pipeline monitoring, yet most recognizers are evaluated as closed-set classifiers, whereas a deployed fiber also records nuisance sources and event types absent from training. In our experiments, a closed-set recognizer with 0.999 accuracy still gives a false-alarm rate (FAR) of 0.42–0.64 on simulated unknown nuisance classes, and a conformal threshold calibrated only on known negatives does not remove this failure. We therefore propose Physically Consistent Risk Calibration (PCRC), which combines a label-free physical-consistency gate computed from spatial compactness, common-mode ratio, and signal-to-noise ratio (SNR) with a conformal threshold for gate-passing negatives and a three-level alarm/review/discard decision. The guarantee is conditional: conformal calibration controls gate-passing negatives under exchangeability, and the gate removes only physically inadmissible nuisance windows. On three public distributed acoustic sensing (DAS) field datasets, the known-negative FAR follows the target level when calibration and test negatives are exchangeable, held-out threat coverage reaches 0.87 and 0.68 on the two multichannel corpora, and every physically admissible held-out nuisance passes the gate and requires recalibration from verified field negatives. In a controlled simulation whose unknown nuisance classes are constructed to be inadmissible, PCRC produces no observed automatic false alarms while retaining 0.998 known-threat recall.