Correlated time series (CTS) forecasting plays an essential role in many practical applications, such as traffic management and server load control. Many deep learning models have been proposed to improve the accuracy of CTS forecasting. However, while models have become increasingly complex and computationally intensive, they struggle to improve accuracy. Pursuing a different direction, this study aims instead to enable much more efficient, lightweight models that preserve accuracy while being able to be deployed on resource-constrained devices. To achieve this goal, we characterize popular CTS forecasting models and yield two observations that indicate directions for lightweight CTS forecasting. On this basis, we propose the LightCTS framework that adopts plain stacking of temporal and spatial operators instead of alternate stacking that is much more computationally expensive. Moreover, LightCTS features light temporal and spatial operator modules, called L-TCN and GL-Former, that offer improved computational efficiency without compromising their feature extraction capabilities. LightCTS also encompasses a last-shot compression scheme to reduce redundant temporal features and speed up subsequent computations. Experiments with single-step and multi-step forecasting benchmark datasets show that LightCTS is capable of nearly state-of-the-art accuracy at much reduced computational and storage overheads.

Blade icing detection is critical to maintaining the health of wind turbines, especially in cold climates. Rapid and accurate icing detection allows proper control of wind turbines, including shutting down and clearing the ice, thus ensuring turbine safety. This paper presents a wavelet-driven multiscale graph convolutional network (MWGCN), which is a supervised deep learning model for blade icing detection. The proposed model first uses wavelet decomposition to capture multivariate information in the time and frequency domains, then employs a temporal graph convolutional network to model the intervariable correlations of the decomposed multiscale wavelets, as well as their temporal dynamics. In addition, this paper introduces scale attention to the MWGCN for a further improvement of the model and proposes the method to address the class imbalance problem of the training data sets. Finally, the paper conducts comprehensive experiments to evaluate the proposed model, and the results demonstrate the effectiveness of the model in blade icing detection and its better performance over eight state-of-the-art algorithms, with 17.2% and 11.3% higher F1 scores over the best state-of-the-art baseline on the labeled datasets.

Cyber-physical system sensors emit multivariate time series (MTS) that monitor physical system processes. Such time series generally capture unknown numbers of states, each with a different duration, that correspond to specific conditions, e.g., "walking" or "running" in human-activity monitoring. Unsupervised identification of such states facilitates storage and processing in subsequent data analyses, as well as enhances result interpretability. Existing state-detection proposals face three challenges. First, they introduce substantial computational overhead, rendering them impractical in resourceconstrained or streaming settings. Second, although state-of-the-art (SOTA) proposals employ contrastive learning for representation, insufficient attention to false negatives hampers model convergence and accuracy. Third, SOTA proposals predominantly only emphasize offline non-streaming deployment, we highlight an urgent need to optimize online streaming scenarios. We propose E2Usd that enables efficient-yet-accurate unsupervised MTS state detection. E2Usd exploits a Fast Fourier Transform-based Time Series Compressor (fftCompress) and a Decomposed Dual-view Embedding Module (ddEM) that together encode input MTSs at low computational overhead. Additionally, we propose a False Negative Cancellation Contrastive Learning method (fnccLearning) to counteract the effects of false negatives and to achieve more cluster-friendly embedding spaces. To reduce computational overhead further in streaming settings, we introduce Adaptive Threshold Detection (adaTD). Comprehensive experiments with six baselines and six datasets offer evidence that E2Usd is capable of SOTA accuracy at significantly reduced computational overhead. Our code is available at https://github.com/AI4CTS/E2Usd.

Correlated time series (CTS) forecasting is essential in many practical applications, such as traffic management and server load control. Various deep learning based solutions have been proposed to improve forecasting accuracy. However, while models have become increasingly computationally intensive, they struggle to improve accuracy. This study aims instead to enable more lightweight, accurate models suitable for resource-constrained devices. To achieve this goal, we characterize popular CTS forecasting models, yielding two observations for developing lightweight CTS forecasting. On this basis, we propose the LightCTS framework that adopts plain stacking of temporal and spatial operators instead of alternate stacking which is much more computationally expensive. Moreover, LightCTS features light temporal and spatial operators, L-TCN and GL-Former, offering improved computational efficiency without compromising their feature extraction capabilities. LightCTS also encompasses a last-shot compression scheme to reduce redundant temporal features and speed up subsequent computations. Next, we equip LightCTS with two knowledge distillation modules, Tafd and Caad , that result in LightCTS$^\star$ retaining the original benefits of LightCTS , while also being able to adapt to varying levels of ultra-constrained resources. Experimental studies offer detailed insight into these proposals and provide evidence that both LightCTS and LightCTS$^\star$ are capable of nearly state-of-the-art accuracy at substantially reduced computational costs.

Imputation of Correlated Time Series (CTS) is essential in data preprocessing for many tasks, particularly when sensor data is often incomplete. Deep learning has enabled sophisticated models that improve CTS imputation by capturing temporal and spatial patterns. However, deep models often incur considerable consumption of computational resources and thus cannot be deployed in resource-limited settings. This paper presents ReCTSi (Resource-efficient CTS imputation), a method that adopts a new architecture for decoupled pattern learning in two phases: (1) the Persistent Pattern Extraction phase utilizes a multi-view learnable codebook mechanism to identify and archive persistent patterns common across different time series, enabling rapid pattern retrieval during inference. (2) the Transient Pattern Adaptation phase introduces completeness-aware attention modules that allocate attention to the complete and hence more reliable data segments. Extensive experimental results show that ReCTSi achieves state-of-the-art imputation accuracy while consuming much fewer computational resources than the leading existing model, consuming only 0.004% of the FLOPs for inference compared to its closest competitor. The blend of high accuracy and very low resource consumption makes ReCTSi the currently best method for resource-limited scenarios. The related code is available at https://github.com/ryanlaics/RECTSI.

Icing on wind turbine blades in high-altitude, mountainous, and maritime regions poses significant challenges, leading to power degradation and revenue loss. Existing blade icing detection methods mainly rely on deep learning techniques, such as Graph Neural Networks and Transformers, to analyze Supervisory Control and Data Acquisition (SCADA) data from wind turbine sensors. However, these approaches are computationally intensive, suffer from high latency, and require significant computational resources. Moreover, they often transmit SCADA data to centralized data centers for processing, which introduces delays and compromises data quality due to network instability. Additionally, these methods struggle to capture persistent and transient patterns within SCADA data, resulting in low detection accuracy. To overcome these challenges, we propose FREQICE, an efficient blade icing detection model via frequency-domain learning. FREQICE employs a lightweight frequency-domain MLP combined with a multi-view pretrained codebook, ensuring efficient detection inference. The codebook learns persistent patterns from multiple views and can be retrieved via simple table lookups, while the frequency-domain MLP captures transient temporal features effectively. A feature attention block is used to emphasize the most important feature patterns. Together, these modules effectively capture both persistent and transient patterns, enabling effective and efficient blade icing detection on wind turbines. Experimental results demonstrate that FREQICE achieves state-of-the-art (SOTA) detection accuracy across all baseline models. Compared with the existing SOTA method, FREQICE significantly reduces model complexity while maintaining high accuracy. Specifically, it decreases the total number of parameters from 301510 to 50694, achieving a 5.9× reduction. It also lowers the Multiply-Accumulate (MACC) counts from 8.01 million to 0.04 million, leading to a 200.3× acceleration. Additionally, peak memory usage is reduced from 13.46 MB to 0.30 MB, marking a 44.9× decrease. This substantial reduction in memory footprint facilitates deployment on resource-constrained devices without compromising performance.