Traditional fault tree analysis is an effective tool used to evaluate system risk if the required data are sufficient. Unfortunately, the operation and maintenance data of some complex systems are difficult to obtain due to economic or technical reasons. The solution is to invite experts to evaluate some critical aspect of the performance of the system. In this study, the belief degrees of the occurrence of basic events evaluated by experts are measured by an uncertain measure. Then, a system risk assessment method based on an uncertain fault tree is proposed, based on which two general optimization models are established. Furthermore, the genetic algorithm (GA) and the nondominated sorting genetic algorithm II (NSGA-II) are applied to solve the two optimization models, separately. In addition, the proposed risk assessment method is applied for the leakage risk evaluation of a subsea production system, and the two general optimization models are used to optimize the leakage risk and maintenance cost of the subsea production system. The optimization results provide a theoretical basis for practitioners to guarantee the safety of subsea production system.
Remaining useful life (RUL) prediction is a key technology to ensure the reliability and safety of high-end equipment. Deep learning is widely used for RUL prediction due to the excellent feature extraction ability and nonlinear fitting ability. Traditional recurrent neural networks adopt recursive strategy, which easily lead to the problems of error accumulation and low stability. On the other hand, most deep learning methods are used for point prediction and cannot quantify uncertainty in the prediction results. Although some probability prediction methods based on deep learning can provide probability prediction results, it require the assumption that the prediction results follow a specific distribution in advance. However, the distribution of most prediction results does not match a suitable distribution function. To solve above problems, a novel RUL prediction approach based on eXtreme gradient boosting (XGBoost) and multi-quantile recursive neural network (MQ-RNN) is proposed in this article. Specifically, XGBoost is used to select features closely related to RUL, and the selected features are fed into MQ-RNN to train the RUL prediction model. The advantage of MQ-RNN is that it has non-parametric framework, prediction results can be obtained by multi-quantile regression, which does not require prior assumptions about the distribution of predicted results. Furthermore, the proposed framework is verified by C-MAPSS dataset. Finally, comparative experiments are conducted. The experimental results show that the proposed method maintains good predictive performance in both point estimation and interval estimation.
5 December 2025 in International Journal of Green Energy
The degradation of lithium-ion battery (LIB) capacity has both long-term trends and short-term fluctuations, and also exhibits certain stage characteristics. The existing LIB capacity prediction methods are difficult to achieve high accuracy, low computational overhead and high reliability simultaneously in long-term prediction. To address these challenges, this article proposes a novel multi-resolution Mamba network with adaptive degradation perception. Specifically, mechanism features from charge and discharge processes are constructed to characterize the LIB capacity degradation state more precisely. Then an adaptive degradation stage perception soft-gating mechanism is provided to quickly adapt to stage transitions during long-term prediction. Subsequently, a multi-resolution Mamba (MR-Mamba) block is designed for feature extraction, which utilizes a shared-parameter kernel in parallel high-frequency and low-frequency branches to jointly capture global degradation trends and local fluctuations. Finally, a heteroscedastic sliding-window conformal prediction is proposed to generate reliable confidence intervals. Hold-out validation on full lifecycle LIB datasets covering multiple chemical compositions and operating conditions demonstrates that the proposed model achieves higher prediction accuracy, lower computational cost, faster inference speed and superior generalization capability.
State of charge (SOC) serves as a key performance metric for electric vehicle batteries. Existing SOC estimation methods for battery packs often neglect interactions be tween individual cells and pay little attention to the coupled effects of short-term signal fluctuations and long-term degradation processes. To address these issues, this article proposes a multi-scale state perception temporal network (MS-SPTN) for SOC estimation of battery packs during vehicle operation. First, a robust feature set construction method is proposed, which includes packet-level operational features and inconsistency features of individual cells. Furthermore, MS-SPTN model with encoder-decoder architecture is designed to achieve precise SOC estimation. The encoder consists of stacked residual Mamba-FFN blocks to provide global state representation at near-linear complexity. The decoder integrates multi-resolution adaptive sparse attention (MRASA) and cross-scale feature enhancement (CSFE) to capture abrupt fluctuations and slow degradation trends, while further strengthening the complementarity of cross-scale features. Additionally, an online affine calibration layer based on recursive least squares (RLS) is integrated during the inference phase to ensure the accuracy and stability of SOC estimation before MS-SPTN updates. Finally, experiments are conducted using battery pack monitoring data from different vehicles to verify that the proposed method has significant advantages in both accuracy and stability of real-time SOC estimation.