Multi-sensor and Synthetic Data Fusion via a Cortical Gap Network for Time Series Large Data Gap Filling in a Complex Adaptive System

Environmental monitoring with distributed water quality probes in a water filtration system is crucial for public health and environmental protection. However, the long-term large interval gaps of such a data set collected from a suite of water quality probes and media samples have been a big challenge for performance prediction, on-line diagnosis, and process control.

This presentation introduces a novel multivariate time series gap filling technique – a new data science issue crucial across many domains when facing significant challenges with large, contiguous missing data making traditional and even advanced forecasting methods falter. This novel machine learning method - Cortical Gap Network (CGN) - has a nature-based model architecture inspired by the principles of parallel multiscale processing, frequency analysis, and adaptive gating in brain science and neuroscience. CGN uniquely features an adaptive configuration based on dataset-level missingness and integrates parallel pathways (local, global, spectral), including a TimesNet-inspired frequency module, and a gap-aware Mixture-of-Experts (MoE) mechanism within a Transformer framework. Evaluated by a challenging real-world water quality data set with large block missingness, CGN exhibits an ideal solution for multi-sensor and synthetic data fusion and achieves state-of-the-art imputation accuracy (R² ≈ 0.9997) and high computational efficiency (1.2s/epoch vs. 16s/epoch for TimesNet). 

Such a record significantly outperforms baselines based on our real-world dataset. CGN demonstrates the power of brain-inspired, multi-level adaptive architectures for robust and efficient handling of severely incomplete time series that is adaptable for many other applications such as large language models. It ultimately contributes to better sustainable design, strategies, and practices for complex large-scale systems and beyond (e.g., transdisciplinary complex adaptive system).