Adaptive Outlier Detection for Non-Stationary Stochastic Systems

Cho et al.'s automated modal analysis uses fixed statistical thresholds (e.g., MAC variability) to remove outliers, assuming stationarity. However, real-world systems like EV charging networks (Zhang et al., 2022) or contaminant transport (Muniruzzaman & Rolle, 2023) exhibit non-stationarity where "normal" behavior shifts over time. This research proposes an adaptive outlier detection framework where thresholds are updated in real-time using Bayesian inference or reinforcement learning. For example, in EV charging stations, outlier thresholds for demand spikes could adapt to traffic patterns or weather. Unlike Cho et al.'s pre-cleaning stage, this approach treats outliers as context-dependent signals rather than noise. The novelty lies in merging stochastic process monitoring with online learning, potentially improving robustness in dynamic environments like power grids (Vardhan et al., 2023) or ecological systems.

References:

1. Stochastic Subspace Identification-Based Automated Operational Modal Analysis Considering Modal Uncertainty. Keunhee Cho, Jeong-Rae Cho (2023). Applied Sciences.
2. Stochastic Modeling and Analysis of Public Electric Vehicle Fleet Charging Station Operations. Tianyang Zhang, Xi Chen, Bin Wu, M. Dedeoglu, Junshan Zhang, L. Trajković (2022). IEEE transactions on intelligent transportation systems (Print).
3. A Comparative Analysis of Hyperparameter Tuned Stochastic Short Term Load Forecasting for Power System Operator. B. V. S. Vardhan, M. Khedkar, Ishan Srivastava, P. Thakre, N. Bokde (2023). Energies.
4. Relevance of charge interactions for contaminant transport in heterogeneous formations: a stochastic analysis. M. Muniruzzaman, M. Rolle (2023). Stochastic environmental research and risk assessment (Print).