Supervisory control and data acquisition (SCADA) is widely used in wind farms as an effective data acquisition system for wind turbines (WTs). However, in practical engineering applications, it is difficult for us to have adequate conditions to collect enough WT blade icing data, which leads to data imbalance and uneven distribution in the feature space. Using the classical synthetic minority oversampling technique (SMOTE) to balance the data may increase the overlap of positive and negative samples, or produce some redundant samples without useful information. A center jumping boosting machine (CJBM) method is proposed that combines an improved clustering-based oversampling (γ mini density peaks clustering SMOTE, γMiniDPC-SMOTE) and light gradient boosting machine (LightGBM) for blade icing prediction. First, to solve the problem of imbalanced and uneven distribution of WT data, a ${\gamma }$ MiniDPC-SMOTE method is proposed, which divides icing samples into multiple clusters, then increases icing samples, and alleviates uneven distribution in feature space. Second, calculating the intercept distance ${d}_{c}$ based on the binary search method and the adaptive selection of DPC parameters based on the step phenomenon of $\gamma $ parameters and verified by $\gamma $ -step of two WT icing data are proposed. Then, for the problem of low operating efficiency of the model under a large amount of imbalanced data, LightGBM is used for model training and icing prediction. Finally, validation was performed on two SCADA datasets. The results showed that the accuracy, precision, recall, F1-measure, and running times increased significantly, proving the superiority of the CJBM.

This paper proposes a novel data-driven adaptive decomposition method named broadband mode decomposition (BMD) for analyzing complex signals containing broadband components, such as square signals and sawtooth signals. For effective broadband signals with “sharp corners”, an unavoidable error occurs when applying former methods, such as variational mode decomposition (VMD) and ensemble empirical mode decomposition (EEMD), due to the Gibbs phenomenon and the interpolation of extreme points. Therefore, we propose BMD for separating the broadband modes, narrowband modes and noise in a complex nonstationary signal. First, an associative dictionary library consisting of typical broadband signals and narrowband signals is established. Second, a smoothness function is employed as the optimization objective function, and the amplitude, frequency and phase of the broadband signals and the filter parameters of the narrowband signals are applied as optimization parameters. Third, the sparsest decomposition results are obtained by searching the dictionary using an artificial chemical reaction optimization algorithm (ACROA). The simulation and experimental signal analyses indicate that BMD is more accurate than EEMD and VMD in extracting broadband components from a noisy signal and that BMD is suitable for quality evaluations of welding inverter power source signals.

The acoustic emission (AE) signal collected by a sensor in the welding process has an overlapping frequency band and weak characteristics under a complex noise background. It is difficult for the wavelet noise reduction method, with single basis function, to effectively match the different characteristic information of the welding crack AE signal. Taking into account the adaptive decomposition characteristics of Empirical Mode Decomposition (EMD), a novel wavelet packet noise reduction method for welding AE signal was proposed. The welding AE signal was adaptively decomposed into several Intrinsic Mode Functions (IMFs) by the EMD. The effective IMFs were selected by the frequency distribution characteristics of the welding crack AE signal. A wavelet packet, with a specific basis function, was subsequently performed on the effective IMFs, which were reconstructed to be the welding crack AE signal. The simulated and experimental results indicated that the proposed method can effectively achieve noise reduction of the welding crack AE signal, which provided a mean for structure crack detection in the welding process.

With the burgeoning development of electric vehicles (EVs), the efficient and economical charging mode, represented by the multi-stage constant current (MsCC) protocol, becomes increasingly popular during daily usage. In this paper, the health indicators (HIs) extracted from the MsCC charging data sets based on the charging behaviour analysis of real-world EVs are discussed, and a novel data-driven strategy for battery state of health (SOH) estimation for the MsCC charging process is proposed. First, the battery HIs are acquired from the MsCC charging datasets and delicately sifted according to the user behaviour of real-world EVs. Next, the correlation between HIs and battery SOH is analyzed through a Pearson correlation coefficients (PCC) matrix to select highly relevant HIs and avoid redundancy. Finally, the mapping relationship is constructed by the long short-term memory (LSTM) model, and its effectiveness is verified by comparison with the other three methods. To emphasize the robustness and generality of the proposed strategy, two training and testing settings and different kinds of MsCC test datasets are applied to test the model performance. The results show that the LSTM-based structure realizes about 32%, 36%, and 49% improvement in battery SOH estimation compared to the benchmark methods, and a promotion of about 18%, 27%, and 10% in prediction accuracy is observed based on another dataset.