A Multilevel Convolutional Recurrent Neural Network for Blade Icing Detection of Wind Turbine

Abstract

Blade icing detection becomes increasingly significant as it can avoid revenue loss and power degradation. Conventional methods are usually limited by additional costs, and model-driven methods heavily depend on prior domain knowledge. Data-driven methods, especially deep learning approaches without needing the time-consuming handcraft feature engineering, offer a promising solution for blade icing detection. However, the monitoring signals normally have complex and diverse features as wind turbine operates in complex environments, thus effective model is needed for data analyzing. Additionally, the distribution of monitoring data is imbalanced, which causes the abnormal data mining inadequate. In this work, a multilevel convolutional recurrent neural network (MCRNN), is proposed for blade icing detection. Specifically, discrete wavelet decomposition is leveraged to obtain multilevel features both from the time domain and the frequency domain. A parallel structure combining an LSTM branch and a CNN branch is established in each level for feature extraction. To alleviate the severe data imbalance, two mechanisms, including data resampling algorithm and class-rebalanced loss function, are investigated. Furthermore, a multi-step accumulation strategy is proposed to enhance the accuracy of real-time detection. Extensive studies demonstrate that the proposed MCRNN can achieve up 38.8% and 42.9% higher F1-score over the best baseline on the balanced data sets processed by data resampling algorithm and 23.9% and 30.6% higher on imbalanced data sets with MCRNN optimized by the class-rebalanced loss function. The real-time detection verifies the applicability of the proposed method and indicates that the proposed multi-step accumulation strategy can improve the accuracy of icing detection.

A Blockchain-Empowered Cluster-Based Federated Learning Model for Blade Icing Estimation on IoT-Enabled Wind Turbine

“…Instead, data-driven approaches, especially deep learning models, can bypass this problem and map the correlations between operational signals to obtain competitive performances of detecting icing events. Tian et al integrated the discrete wavelet decomposition into a special-designed multilevel convolutional recurrent neural network for blade icing detection [13]. Cheng et al combined CNN with a temporal attention module to automatically determine the important sensors and extract discriminative information from Supervisory Control and Data Acquisition (SCADA) data [14].…”

Multiscale Wavelet-Driven Graph Convolutional Network for Blade Icing Detection of Wind Turbines

“…Yuan et al [33] proposed a wavelet transformation and a fully convolutional neural network (WaveletFCNN) to automatically obtain multiscale wavelet features from the time domain and frequency domains. Tian et al [8] improved WaveletFCNN by introducing a parallel structure consisting of an LSTM and CNN branch network structure. However, although the detail coefficients from Discrete Wavelet Transform (DWT) can be interpreted as an additive decomposition of the signal, referred as multi-resolution [33], the existing approaches still have not specifically considered the scale-specific inter-variate correlations between multiple wavelet scales sufficiently, which could lead to inaccurate representations for real-world scenarios.…”

“…Furthermore, Yuan et al also proposed WaveletAE, an unsupervised model that combines the autoencoder and the discrete wavelet transform [49]. By introducing an LSTM branch network to temporal information, Tian et al improved WaveletFCNN and proposed MCRNN [8].…”

“…These methods directly measure the physical properties of ice and are quite reliable. However, they may have negative effects on the structure of the blade, resulting in costly human and material resources, and they are insensitive to small amounts of ice accretion [8]. As a result, engineers adopt indirect methods to estimate blade icing conditions in practice, such as monitoring the power curve [9], anemometers [10], vibration time wave forms [11], dew point and temperature [12], to determine whether to trigger the deicing procedure.…”

“…Thus, the SCADA system can manage the installed general-purpose sensors that monitor wind turbine operation and provide multivariate time series data of the real-time operating environment to data-driven machine learning models, such as linear discriminant [23], KNN [24], SVM [19] and some hybrid methods [25]- [27]. However, shallow machine learning methods rely heavily on hand-crafted feature designs, which are time consuming and costly, while deep learning methods do not require as much manual feature extraction, which receive a great deal of interest from researchers [8], [28]- [35].…”