Recommendation systems commonly face selection bias from missing-not-at-random (MNAR) collected data. To address this bias, propensity-based methods such as inverse propensity scoring (IPS) and doubly robust (DR) estimators are widely used. In addition, many methods extend the vanilla IPS and DR to further control the bias, variance, propensity mis-calibration, and imbalance, but they only optimize some of the above metrics, limiting the debiasing performance. In this paper, we first empirically find that controlling one metric cannot guarantee the control of other important metrics, then we reveal a fundamental structural commonality among the above four important metrics, and propose a Unified Propensity Optimization (UPO) framework that optimizes all metrics simultaneously by a minimax optimization algorithm. Theoretically, we demonstrate that minimizing the UPO loss effectively controls all metrics, ensuring their simultaneous improvements without incurring additional bias, and achieving reduced variance compared to naively adding up multiple control losses in penalty terms. Empirically, experiments on a semi-synthetic dataset and three real-world datasets validate UPO’s effectiveness, demonstrating superior performance compared to state-of-the-art methods with minor computational overhead. We fully open-source our code.

Electrohydrostatic Actuator (EHA) systems operating under variable conditions suffer from data distribution shifts and feature overlapping of compound faults, leading to degraded diagnostic performance. Existing methods often struggle to align features across different working domains and fail to accurately distinguish coupled fault patterns due to signal entanglement. To address these challenges, this paper proposes a Multi-Condition Feature Aligned Capsule Network (MCFACN) for cross-domain compound fault diagnosis. An Adaptive Integrated Maximum Mean Square Discrepancy (AIMMSD) metric, incorporating higher-order statistics and mean constraints, is proposed to capture distribution differences more comprehensively, Furthermore, an aggregated attention routing mechanism is employed to integrate multi-condition source-domain data with convolutional and capsule features, enhancing domain adaptation and feature separability. Extensive experiments on multi-condition transfer tasks demonstrate the superiority of the proposed method. Specifically, MCFACN achieves an average diagnostic accuracy of 99.4%, validating its effectiveness and robustness for reliable aerospace applications.

The control of the grid-forming (GFM) inverter is crucial, as it provides stable frequency and voltage support to the system, allowing for the integration of more renewable energy sources. This article aims to develop a novel primary control strategy that combines risk-sensitive control and residual compensation control, enhancing flexibility in strategy design to improve GFM inverter performance under disturbances, noise, and parameter uncertainties. The presented risk-sensitive residual generator-based compensation-control strategy includes two risk-sensitive factors$-$one for unbiased state estimation and residual generation, and another for improving closed-loop system performance. These two risk-sensitive factors increase the design freedom of the primary control strategy for the GFM inverter. First, this article establishes a link between risk-sensitive state estimation and residual generation, using the relationship between the risk-sensitive factor and$\tau$-divergence to design a risk-sensitive residual generator. Then, this article introduces a novel connection between risk-sensitive control and residual compensation control. By incorporating a risk-sensitive factor and its relationship with system stability margin, this article solves for the residual compensation controller with only two Riccati equations. Finally, experimental results on the StarSim platform demonstrate the effectiveness of the presented risk-sensitive control strategy for the GFM inverter in managing disturbances, noise, and parameter uncertainties.

Grid-forming (GFM) inverters with inductor–capacitor (LC) filters are used to interface distributed generation with the grid. However, the GFM inverter systems are prone to incipient multiplicative faults due to the inherent parameter drift phenomenon of theLCfilter. Motivated by this scenario, this article presents a primary active fault-tolerant control (PAFTC) strategy based on the robust-residual-observer-based-digital-twin model (RRODTM) to achieve the fault-tolerant control for the incipient multiplicative faults. The RRODTM is composed of a robust residual observer and a residual-driven compensation controller. Specially, the residual-driven compensation controller is designed by the data-driven method based on the multiplicative uncertainty caused by the incipient multiplicative faults. At the same time, this article constructs a data-driven-event-trigger mechanism based on stability margin to achieve the plug-and-play (PnP) function of the residual-driven compensation controller. Based on these, the PAFTC strategy improves the control performance online according to the degree of the incipient multiplicative fault. Comprehensive experiments show that compared to well-known control schemes in the literature, the PAFTC strategy improves the power distribution accuracy and transient response speed as well as enhances the robustness of the system.

This paper presents a primary control strategy that can recover control performance online for the voltage source inverter (VSI) with inductor-capacitor (LC) filter. The control strategy contains a droop-based feedforward controller, an LQR-based state feedback controller and a data-driven residual compensation controller (FFR). An event-triggered mechanism based on stability margin is used to implement the plug-and-play (PnP) function of the residual compensation controller. Simulations show that FFR can realize uncertainty control and disturbance suppression and improve the dynamic performance, robustness and power distribution ability of the VSI system.