Hybrid Meta-Learning Framework for Anomaly Forecasting in Nonlinear Dynamical Systems via Physics-Inspired Simulation and Deep Ensembles

TL;DR

This paper introduces a hybrid meta-learning framework combining physics-inspired simulation, deep learning, and ensemble methods for improved anomaly forecasting in complex nonlinear dynamical systems.

Contribution

It presents a novel integration of physics-based simulation with deep ensemble learning and meta-aggregation for anomaly detection and forecasting in nonstationary systems.

Findings

Hybrid ensemble outperforms individual models in anomaly detection.

Framework enhances robustness to nonlinear deviations.

Effective for early defect detection in complex systems.

Abstract

We propose a hybrid meta-learning framework for forecasting and anomaly detection in nonlinear dynamical systems characterized by nonstationary and stochastic behavior. The approach integrates a physics-inspired simulator that captures nonlinear growth-relaxation dynamics with random perturbations, representative of many complex physical, industrial, and cyber-physical systems. We use CNN-LSTM architectures for spatio-temporal feature extraction, Variational Autoencoders (VAE) for unsupervised anomaly scoring, and Isolation Forests for residual-based outlier detection in addition to a Dual-Stage Attention Recurrent Neural Network (DA-RNN) for one-step forecasting on top of the generated simulation data. To create composite anomaly forecasts, these models are combined using a meta-learner that combines forecasting outputs, reconstruction errors, and residual scores. The hybrid ensemble performs better than standalone models in anomaly localization, generalization, and robustness to nonlinear deviations, according to simulation-based experiments. The framework provides a broad, data-driven approach to early defect identification and predictive monitoring in nonlinear systems, which may be applied to a variety of scenarios where complete physical models might not be accessible.