From Benchmarks to Production: Transferring Time Series Anomaly Detection Methods for Electricity Production Monitoring

Nicolas Vautier; Paul Caron; Nardi Xhepi; Félicie Bizeul; Manel Boumghar; Christophe Degouy; Paul Boniol

cs.LGSep 30, 2026

From Benchmarks to Production: Transferring Time Series Anomaly Detection Methods for Electricity Production Monitoring

Authors: Nicolas Vautier, Paul Caron, Nardi Xhepi, Félicie Bizeul, Manel Boumghar, Christophe Degouy, Paul Boniol

Organizations: EDF Lab Paris Saclay · EDF, DOAAT · Inria, Ecole Normale Sup´erieure (PSL), CNRS

Abstract

Accurate forecasting of electricity production is essential for maintaining the operational efficiency and strategic planning of energy utilities. In industrial settings, such forecasts are generated daily to ensure supply-demand balance and optimal management of production assets. However, the increasing complexity of modern power systems and data flows poses significant challenges for ensuring the reliability and consistency of these forecasts. This paper addresses the problem of anomaly detection in short-term production forecasts at EDF, formulated as identifying atypical intra-day patterns that may signal data quality issues or operational irregularities. We introduce TAMIS, a scalable and interpretable system that analyzes daily production time series to automatically detect anomalous days based on deviations from historical patterns learned from past data. Designed for human-in-the-loop workflows, TAMIS surfaces top-ranked anomalies through an automated daily newsletter, enabling efficient expert review and continuous monitoring. An extensive experimental evaluation on real-world industrial data demonstrates that TAMIS achieves the best accuracy-efficiency trade-off compared to baseline methods. To foster further research and reproducibility, we publicly release the anonymized application datasets used in our study.

Figures & tables

Fig. 1 : Examples of time series in our proposed JO dataset (anomalies are highlighted in red).

Symbol	Description
$\Delta\in\mathbb{N}$	Sampling interval ( $30$ minutes).
$M\in\mathbb{N}$	Number of samples per day ( $M=48$ ).
$\boldsymbol{C}_{d}\in\mathbb{R}^{M}$	Daily window (24-hour profile) for day $d$
$c_{m}\in\mathbb{R}$	$m$ -th value of $\boldsymbol{C}_{d}$
$\boldsymbol{F}_{d}\in\mathbb{R}^{i}$	Feature set for day $d$
$\boldsymbol{f}_{j}\in\mathbb{R}$	$j$ -th value of $\boldsymbol{F}_{d}$

TABLE I : Summary of main notations used in this paper.

Fig. 2 : Raw-based Time series anomaly detection methods in the literature [ 9 ]

Fig. 3 : Example of TSFresh and Catch22 features on IOPS [ 25 ] time series.

Fig. 4 : Time series anomaly detection: from unique detector (a) to supervised ensembling (d).

Fig. 5 : (1) Training and (2) production pipeline of the proposed solution TAMIS . The end-user interface is illustrated in (3).

Feature Description	Formula	Complexity
Expert-based : Top-10 features from energy-production expert-knowledge
amplitude: maximum amplitude for a given window $\boldsymbol{C}_{d}$ .	$\max(\boldsymbol{C}_{d})-\min(\boldsymbol{C}_{d})$	O(n)
val max: Maximum value in $\boldsymbol{C}_{d}$ .	$\max(\boldsymbol{C}_{d})$	O(n)
val min: Minimum value in $\boldsymbol{C}_{d}$ .	$\min(\boldsymbol{C}_{d})$	O(n)
val mean: Mean of $\boldsymbol{C}_{d}$ .	$\mu(\boldsymbol{C}_{d})$	O(n)
val std: Standard deviation of $\boldsymbol{C}_{d}$ .	$\sigma(\boldsymbol{C}_{d})$	O(n)

TABLE II : TAMIS F features. Each feature is computed for an entire window (i.e., day

d

). Note:

\mathbb{I}(p)

is the indicator function for NaN values.

\Delta{C}_{d}

represents first-order differences.

V_{\text{reocc.}}

is the set of reoccurring values in

C_{D}

Characteristics	JO	HYDRAU	THERM
Number of sensors	3198	1740	4515
Total number of measurements	219M	267M	103M
Number of Labeled days	5813	540	578
NaN ratio	20.3%	3.5%	27.9%
Total number of NaN	55334	640	7827
Number of Anomalies	307	57	12

TABLE III : Datasets characteristics. NaN ratio: number of labeled days with at least one NaN value.

Fig. 6 : Precision-Recall curves of TAMIS against (a) detectors on raw data; (b) detectors on TAMIS F feature set; (c) Automatic solutions (Model Selection and Ensembling). (d) Throughput versus AUC-PR for TAMIS against most competitive baselines.

Fig. 7 : Comparision of AUC-PR versus throughput between TAMIS using TAMIS F , Catch22 and TSFresh as feature sets. The throughput corresponds to (a) Features computation, (b) Detectors computation, and (c) weights inference.

Detectors on raw time series
Methods	JO	HYDRAU	THERM
HBOS	0.069	0.297	0.041
LOF	0.228	0.226	0.072
OCSVM	0.138	0.143	0.028
Detectors on (TAMIS F features)
HBOS	0.656	0.596	0.191

TABLE IV : Accuracy (AUC-PR) of TAMIS and baselines applied on raw data and TAMIS F features.

Fig. 8 : Comparison of our proposed approach TAMIS versus baselines on different models in-distribution (trained and tested on JO dataset) and two out-of-distribution scenarios. Note that all f. and r. stands for all features and raw .

Fig. 9 : TAMIS vs Best Detector, Avg Ens, the Best Model Selection strategy and the theoretical Oracle on TSB-UAD [ 11 ] .

Explore similar work

Sep 30, 2026cs.LG

What Streaming Anomaly Detection Finds (and Misses) in Industrial Time Series

EDF relies on continuous monitoring of its power plants to detect anomalies as soon as they occur. Given the absence of a universally optimal streaming method in unsupervised settings, we compare streaming methods with state-of-the-art TSAD models deployed online on a real nuclear power plant dataset. This work also evaluates Automated Anomaly Detection in a streaming context. Results show higher consistency for online TSAD and strong robustness from ensembling strategies.

Magali Parrino, Antoine Ajenjo, Emmanuel Remy +2

EDF R&D, France · Inria, ENS, CNRS, PSL, France

Time-Series Anomaly Detection Streaming Stance

Ask PDF

May 31, 2026cs.LG

ChronosAD: Leveraging Time Series Foundation Models for Accurate Anomaly Detection

Time series anomaly detection is a crucial task in various domains, including finance, healthcare, and industry. However, existing methods often struggle to generalize across different datasets, especially when anomalies are subtle or context-dependent. To solve this issue, we introduce ChronosAD, a novel architecture for anomaly detection that uses a time series foundation model as a feature extractor. Specifically, it employs a two-stage pipeline: first, it uses the foundation model to extract embeddings for each time series in a zero-shot manner. Then, a custom-developed Temporal Block, composed of Bidirectional Long Short-Term Memory (BiLSTM) and Multi-Head Attention, refines these embeddings to capture temporal dependencies and highlight salient patterns. Unlike previous approaches, our model requires minimal task-specific tuning and demonstrates robust generalization across a wide range of domains, including industrial, medical, cyber-physical, and automotive systems. Extensive experiments on 11 benchmarks show that ChronosAD outperforms existing methods by 4.72% in AUC and 6.60% in AP on average. The source code is available at https://github.com/intelligolabs/ChronosAD.

Uzair Khan, Luigi Capogrosso, Francesco Biondani +4

University of Verona · Interdisciplinary Transformation University of Austria · ETH Zurich

Time-Series Anomaly Detection Time Series Foundation Models Time Series Chronos Bidirectional Long Short-Term Memory

Ask PDF

Dec 8, 2025cs.LG

Weighted Contrastive Learning for Anomaly-Aware Time-Series Forecasting

Reliable forecasting of multivariate time series under anomalous conditions is crucial in applications such as ATM cash logistics, where sudden demand shifts can disrupt operations. Modern deep forecasters achieve high accuracy on normal data but often fail when distribution shifts occur. We propose Weighted Contrastive Adaptation (WECA), a Weighted contrastive objective that aligns normal and anomaly-augmented representations, preserving anomaly-relevant information while maintaining consistency under benign variations. Evaluations on a nationwide ATM transaction dataset with domain-informed anomaly injection show that WECA improves SMAPE on anomaly-affected data by 6.1 percentage points compared to a normally trained baseline, with negligible degradation on normal data. These results demonstrate that WECA enhances forecasting reliability under anomalies without sacrificing performance during regular operations.

Joel Ekstrand, Tor Mattsson, Zahra Taghiyarrenani +3

Center for Applied Intelligent Systems Research, Halmstad University, Halmstad, Sweden · Mikael Linden Consulting AB, Stockholm, Sweden

Time-Series Anomaly Detection Contrastive Learning Banking

Ask PDF