In order to achieve fault diagnosis and prognosis, one needs a sufficient and valid life-cycle data. However, this requirement is difficult for current high-reliable manufacturing system. Good thing is that the technique of accelerated degradation testing can be used to address this issue. Bad thing is that, it needs a reliable testing/measuring technique to build an accurate model for accelerated degradation testing. However, in practical applications, data acquisition is obtained by sensors or measurement devices, which cannot guarantee perfect working condition under the influence of external environment and stresses, resulting in unreliable signals. Furthermore, since traditional models require complex differentiation and cannot obtain analytical expressions when considering unreliable signals, traditional models rarely reflect well this situation. Motivated by these facts, an accurate model for the accelerated degradation testing is proposed in this study with considering the unreliable signals. Based on the proposed model, a closed-form expression for the useful life analysis is derived. The M-H sampling method is used to estimate the unknown parameters used in the proposed model. For illustration, the electrical connectors dataset is analyzed with the proposed model and the traditional models. Comparing the obtained results, the proposed model is more accurate in the useful life analysis than the traditional accelerated degradation testing models by considering the unreliable signals.

In this study, neuro‐fuzzy‐based identification procedure for Hammerstein model with coloured noise is presented. Separable signal is used to realise the decoupling of the identification of dynamic linear part from that of static non‐linear part, and then correlation analysis method is adopted to identify the parameters of the linear part. Furthermore, by combining multi‐innovation and gradient search theory, multi‐innovation‐based extended stochastic gradient approach is derived for improving the parameters estimation accuracy of the non‐linear part and the noise model. In addition, the convergence analysis in the martingale theory illustrates that the parameter estimation error will converge to zero under the persistent excitation condition. Finally, two simulation results demonstrate that the proposed approach has high identification accuracy and good robustness to the disturbance of coloured noise.

Accurate and stable wind power prediction is the basis of wind energy planning, dispatching, and control in the generation and conversion of the wind power generation industry. A new nonlinear hybrid model, singular spectrum analysis (SSA)-temporal neural network (TCN) is proposed to improve the accuracy of wind power sequence prediction. This method introduces SSA, which decomposes the original wind power sequence into four parts: trend, primary detail components, secondary detail components, and noise. The experiment is conducted under the three conditions of no noise removal, only noise removal, and decomposing original sequences without noise into several components to verify that SSA does help improve the model performance. Moreover, this paper adopts a new TCN network instead of the current most advanced Long Short-Term Memory (LSTM) network. It has a longer actual memory range and is better able to cope with the gradient disappearance. In order to study the advancement of the TCN, the experimental part also includes the performance comparison between the TCN and the LSTM network, as well as the ablation experiment of the TCN. Finally, multi-step prediction is conducted to further demonstrate the fitting ability of the TCN, SSA-TCN, and LSTM network. The results show that (a) the SSA-TCN has the best predictive performance among the involved models; (b) SSA can significantly improve the prediction accuracy; (c) the performance of the TCN is better than that of the LSTM network because of the extended causal convolution structure of the TCN and the residual blocks.

Harnessing the advantage of digital twin (DT) technology, smart grid provides tempting prospects for efficient management of energy manufacturing, conservation, demand forecasting, pricing, and scheduling. However, the development of smart grid is challenged by volatility of individual users, cyber attacks, and hesitation about data sharing. In this study, a lightweight blockchain-enhanced federated learning (FL) framework is proposed to address the above issues and promote efficient collaborative DT modeling. Through encrypted blockchain retrieval, data aggregation and outlier detection, a Proof of Distribution consensus algorithm is developed to periodically update a conformal score for each edge user over time. The conformal score measures conformity of one user with patterns of the overall data distribution at current stage and is further utilized as basis for stochastic participant selection and model averaging, which improves stability and pertinence of FL in an uncertain environment with heterogeneity and concept drift. In the meantime, based on average learning loss of users with top-ranked conformal scores and engagement of users in the learning process, a reputation score is updated in real time for each user, which offers timely detection of malicious behaviors in the cooperation. The transmission of models and data are both implemented with blockchain, which guarantees privacy preservation and makes the process auditable. Through simulation of residential load forecasting based on real-world scenarios, double robustness of our framework against data volatility and adversarial attacks is verified.