Author name cluster

Muyao Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

4 papers

1 author row

NeurIPS Conference 2025 Conference Paper

Channel Matters: Estimating Channel Influence for Multivariate Time Series

Muyao Wang
Zeke Xie
Bo Chen
Hongwei Liu
James Kwok

The influence function serves as an efficient post-hoc interpretability tool that quantifies the impact of training data modifications on model parameters, enabling enhanced model performance, improved generalization, and interpretability insights without the need for expensive retraining processes. Recently, Multivariate Time Series (MTS) analysis has become an important yet challenging task, attracting significant attention. While channel extremely matters to MTS tasks, channel-centric methods are still largely under-explored for MTS. Particularly, no previous work studied the effects of channel information of MTS in order to explore counterfactual effects between these channels and model performance. To fill this gap, we propose a novel Channel-wise Influence (ChInf) method that is the first to estimate the influence of different channels in MTS. Based on ChInf, we naturally derived two channel-wise algorithms by incorporating ChInf into classic MTS tasks. Extensive experiments demonstrate the effectiveness of ChInf and ChInf-based methods in critical MTS analysis tasks, such as MTS anomaly detection and MTS data pruning. Specifically, our ChInf-based methods rank top-1 among all methods for comparison, while previous influence functions do not perform well on MTS anomaly detection tasks and MTS data pruning problem. This fully supports the superiority and necessity of ChInf.

PDF Details

EAAI Journal 2025 Journal Article

Real-time dynamic coordinated optimization control with near-global optimal learning for connected plug-in hybrid electric vehicles

Jiaqi Xue
Chao Yang
Jiayi Fang
Xiao Zhang
Muyao Wang

With the application of connected technologies, such as vehicle-to-vehicle and vehicle-to-cloud communication in connected vehicles to obtain various traffic information, it is crucial to balance the optimality and computational burden of the energy management strategy for further improving the fuel economy of the connected plug-in hybrid electric vehicle. Another key factor affecting the improvement of fuel economy and control performance in connected plug-in hybrid electric vehicles is the fluctuation in driving torque resulting from the different response characteristics of the engine and motor during mode switching of the powertrain for the power distribution of the energy management strategy. To address these challenges, this paper proposes a novel real-time dynamic coordinated optimization control scheme that incorporates energy management at the upper layer and adaptive coordination at the lower layer for the connected plug-in hybrid electric vehicle in a vehicle-following scenario. Based on the offline optimal control rules extracted by the extreme learning machine, which possesses good generalization capabilities, the upper-layer guided model predictive control for energy management is implemented by applying the particle swarm optimization algorithm within variable horizons across different road sections. The bottom-layer adaptive fixed-time control scheme, equipped with a coordinated mechanism, is designed to address transient response deviations in the upper-layer results. The effectiveness and advantages of the proposed hierarchical scheme are validated through both the co-simulation platform and a hardware-in-loop test.

Details DOI

EAAI Journal 2024 Journal Article

A physics-informed learning algorithm in dynamic speed prediction method for series hybrid electric powertrain

Wei Liu
Chao Yang
Weida Wang
Liuquan Yang
Muyao Wang
Jie Su

Engine-generator set (EGS) is an important energy supply component of high-voltage microgrid in series hybrid electric powertrain (SHEP). Sustained and steady energy supply from EGS is one of the conditions for the balanced energy between supply and demand. In some high-power processes, the balanced energy would be broken and the dynamic speed of EGS would be out of expectation, which can result in unstable working states of EGS. If the unstable working states of EGS can be known prior, it is significant for the research of unstable state identification and avoidance. Predicting rotational speed of EGS can warn of the previous issue in advance, while the insufficient data of unstable states would encounter overfitting problems in common prediction methods, so it is a challenge to improve the prediction effect of dynamic speed and then accurately predict the unstable states. Base on the above problems, a physics-informed learning algorithm with adaptive mechanism is proposed for EGS rotational speed prediction in this paper. First, a prediction problem related to the stability of SHEP running state is studied, which is found from engineering knowledge. Second, a new mechanism is proposed for physics-informed learning algorithm, and the physical information adopted to learning algorithm is more selective. Third, a professional adaptive function is originally formed according to speed characteristics, which bridge the information between physics and learning algorithm. By importing the experimental data, the prediction accuracy of proposed method in one of the test cycles is better than the results of baseline methods, specifically 27. 11% and 3. 49%, 11. 90% and 7. 94%, 53. 83% and 27. 62%. In summary, the proposed method can have better predictions against other baseline methods.

Details DOI

AAAI Conference 2024 Conference Paper

Considering Nonstationary within Multivariate Time Series with Variational Hierarchical Transformer for Forecasting

Muyao Wang
Wenchao Chen
Bo Chen

The forecasting of Multivariate Time Series (MTS) has long been an important but challenging task. Due to the non-stationary problem across long-distance time steps, previous studies primarily adopt stationarization method to attenuate the non-stationary problem of original series for better predictability. However, existed methods always adopt the stationarized series, which ignore the inherent non-stationarity, and have difficulty in modeling MTS with complex distributions due to the lack of stochasticity. To tackle these problems, we first develop a powerful hierarchical probabilistic generative module to consider the non-stationarity and stochastity characteristics within MTS, and then combine it with transformer for a well-defined variational generative dynamic model named Hierarchical Time series Variational Transformer (HTV-Trans), which recovers the intrinsic non-stationary information into temporal dependencies. Being an powerful probabilistic model, HTV-Trans is utilized to learn expressive representations of MTS and applied to the forecasting tasks. Extensive experiments on diverse datasets show the efficiency of HTV-Trans on MTS forecasting tasks.

PDF Details DOI