Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hybrid Feature Learning with Time Series Embeddings for Equipment Anomaly Prediction

Takato Yasuno

TL;DR

A hybrid approach that integrates 64-dimensional time series embeddings from Granite TinyTimeMixer with 28-dimensional statistical features based on domain knowledge for HVAC equipment anomaly prediction tasks is proposed.

Abstract

In predictive maintenance of equipment, deep learning-based time series anomaly detection has garnered significant attention; however, pure deep learning approaches often fail to achieve sufficient accuracy on real-world data. This study proposes a hybrid approach that integrates 64-dimensional time series embeddings from Granite TinyTimeMixer with 28-dimensional statistical features based on domain knowledge for HVAC equipment anomaly prediction tasks. Specifically, we combine time series embeddings extracted from a Granite TinyTimeMixer encoder fine-tuned with LoRA (Low-Rank Adaptation) and 28 types of statistical features including trend, volatility, and drawdown indicators, which are then learned using a LightGBM gradient boosting classifier. In experiments using 64 equipment units and 51,564 samples, we achieved Precision of 91--95\% and ROC-AUC of 0.995 for anomaly prediction at 30-day, 60-day, and 90-day horizons. Furthermore, we achieved production-ready performance with a false positive rate of 1.1\% or less and a detection rate of 88--94\%, demonstrating the effectiveness of the system for predictive maintenance applications. This work demonstrates that practical anomaly detection systems can be realized by leveraging the complementary strengths between deep learning's representation learning capabilities and statistical feature engineering.

Hybrid Feature Learning with Time Series Embeddings for Equipment Anomaly Prediction

TL;DR

A hybrid approach that integrates 64-dimensional time series embeddings from Granite TinyTimeMixer with 28-dimensional statistical features based on domain knowledge for HVAC equipment anomaly prediction tasks is proposed.

Abstract

In predictive maintenance of equipment, deep learning-based time series anomaly detection has garnered significant attention; however, pure deep learning approaches often fail to achieve sufficient accuracy on real-world data. This study proposes a hybrid approach that integrates 64-dimensional time series embeddings from Granite TinyTimeMixer with 28-dimensional statistical features based on domain knowledge for HVAC equipment anomaly prediction tasks. Specifically, we combine time series embeddings extracted from a Granite TinyTimeMixer encoder fine-tuned with LoRA (Low-Rank Adaptation) and 28 types of statistical features including trend, volatility, and drawdown indicators, which are then learned using a LightGBM gradient boosting classifier. In experiments using 64 equipment units and 51,564 samples, we achieved Precision of 91--95\% and ROC-AUC of 0.995 for anomaly prediction at 30-day, 60-day, and 90-day horizons. Furthermore, we achieved production-ready performance with a false positive rate of 1.1\% or less and a detection rate of 88--94\%, demonstrating the effectiveness of the system for predictive maintenance applications. This work demonstrates that practical anomaly detection systems can be realized by leveraging the complementary strengths between deep learning's representation learning capabilities and statistical feature engineering.
Paper Structure (28 sections, 7 equations, 7 figures, 1 table)

This paper contains 28 sections, 7 equations, 7 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Time Series Anomaly Detection
  4. Foundation Models for Time Series
  5. Hybrid Machine Learning Approaches
  6. Problem Formulation
  7. HVAC Equipment Anomaly Prediction Task
  8. Dataset Characteristics
  9. Proposed Method: Hybrid Learning Framework
  10. System Architecture Overview
  11. Time Series Embedding Path
  12. Statistical Feature Engineering Path
  13. Feature Fusion and Classification
  14. Experimental Setup
  15. Dataset and Preprocessing
  16. ...and 13 more sections

Figures (7)

  • Figure 1: Hybrid learning framework architecture. The system combines time series embeddings from Granite TinyTimeMixer (64-dim) with statistical features (28-dim) to form a 92-dimensional hybrid representation for LightGBM classification.
  • Figure 2: Performance metrics summary across all three prediction horizons (30-day, 60-day, 90-day) for the hybrid model. The radar chart illustrates the balanced achievement of high Precision, Recall, F1-Score, and ROC-AUC across all horizons.
  • Figure 3: ROC curves for all three prediction horizons. The hybrid model achieves ROC-AUC of 0.995 across all horizons, demonstrating near-perfect discrimination between normal and anomalous states.
  • Figure 4: Precision-Recall curves for all three prediction horizons. The high area under the PR curve (0.92--0.94) confirms robust performance even under class imbalance conditions.
  • Figure 5: Confusion matrices for all three prediction horizons. The matrices demonstrate consistently low false positive rates (0.5--1.1%) and high true positive rates (88--94%) across all horizons.
  • ...and 2 more figures