1. Introduction
Recently, with the continuous development of the renewable energy industry, the cumulative installed wind power capacity in this word has greatly increased, and wind power has become a major contributor to power generation [
1]. In order to make better use of wind energy, wind turbines (WTs) are widely built in high altitude areas with cold climates and high humidity. However, in such operating environment, the WT is prone to the phenomenon of blade icing, which may cause many problems [
2]. On the one hand, after the ice accumulates on the blade, the airfoil changes, which reduces the ability to capture wind energy, and leads to increased consumption of energy to drive the blade to rotate, that ultimately reduces the power generation efficiency. On the other hand, icing changes the modal parameters of the corresponding area on the blade, which may cause the blade to break, leading to more serious operating accidents. Therefore, when ice accumulation on the blade is detected, the deicing equipment should be started immediately. Accordingly, timely detection of icing is of great significance to enhance the power generation efficiency and service life of WTs in wind farms.
Due to the convenient access and huge amount of data provided, the research of WT fault detection based on supervisory control and data acquisition (SCADA) data has been extensively studied. Reference [
3] proposed a WT fault detection model based on SCADA data, which used a variety of data mining algorithms. This model can predict failures within 5–60 min before they occur. In a subsequent study, Reference [
4] used principal component analysis (PCA) to decrease the data dimensionality, and then the random forest (RF) is used to identify early faults. Reference [
5] proposed an alarm processing and diagnosis scheme for WT SCADA systems based on Artificial Neutral Network (ANN) for identifying faults in the pitch system. Their simulation results showed that the method can quickly identify faults.
It can be seen from the above references that the WT status monitoring and fault detection based on SCADA data has a good effect. However, many current research focuses on the monitoring and diagnosis of generators, gearboxes, and pitch systems. Few studies have used SCADA data to detect icing on WT blades. Recently, Reference [
6] proposed a blade icing detection method based on random forest algorithm, which contains 29 kinds of characteristic parameters in a SCADA system. Reference [
7] proposed a hybrid fault detection system that integrates multiple intelligent algorithm. This failure detection strategy can accurately detect the early failure of the blade, and can improve maintenance costs and system availability. Reference [
8] uses SCADA data to construct the features of wind power, wind speed, and generator speed. And uses them as the input of support vector machine (SVM), and combined with particle swarm optimization (PSO) algorithm to establish icing detection model. Reference [
9] proposed a SVM fault diagnosis method based on big data analysis, and used multiple wind turbine data for verification, effectively proving the effectiveness and generalization ability of the model.
Since the SCADA data cannot intuitively obtain the health status of wind turbines, feature enhancement is particularly critical when using SCADA data to assess the health status of various components of wind turbines. According to reference [
10,
11,
12], common research is to extract fault features based on expert experience. The lack of expertise and slow manual selection will affect the performance of the model. In recent years, deep learning algorithms have been widely used in many fields [
13,
14,
15], and its performance is also constantly improving. Since deep learning is good at extracting features, recently fully connected neural network (FCNN) algorithm have been widely used in the field of machine health monitoring and fault diagnosis. Reference [
16] used the FCNN method to extract frames from a video and ran a classifier to perform supervised learning and classification of objects in order to obtain different classes of probabilities, thereby classifying the subject matter and detecting any objects in the video. Compared with earlier similar methods, its accuracy has been improved. Reference [
17] studied the generalization ability of FCNNs trained in the context of time series detection, and studied how to control the generalization ability of the network by adjusting control variables. With these hyperparameters, the complexity of the output function can be effectively controlled without imposing explicit constraints. Reference [
18] proposed a blade icing detection model using a deep autoencoders network, and compared with the traditional machine learning (ML) models, demonstrated the high accuracy and generalization ability of the proposed method.
Although deep learning has been successfully applied in unsupervised feature extraction, it can be further improved. The methods proposed in the above studies for selecting feature subsets from the original feature sets all achieve to some extent effective dimension reduction of high-dimensional data. However, the current method to select input features for WT blade icing prediction models still requires effective comprehensive research. Accordingly, in this study an RF model is used to reduce the features of the SCADA data that affect blade icing, and then a K-nearest neighbor (KNN) algorithm is used to enhance the active power feature. The features after the RF reduction and the active power mean square error (MSE) feature enhanced by the KNN algorithm are combined and used as the input of the FCNN. Then, an empirical analysis is performed for the diagnosis of blade icing. The simulation results show that the proposed model has better diagnostic accuracy than the ordinary back propagation (BP) neural network and other methods, and FCNN–MSE method has excellent chronergy and applicability.
The rest of the paper is arranged as follows:
Section 2 introduces the theoretical method of the RF algorithm.
Section 3 introduces the application of the KNN algorithm in this study.
Section 4 describes details of the structure of the FCNN model and optimization methods used in this study.
Section 5 introduces the blade icing detection model and steps proposed in this study.
Section 6 provides a detailed experimental analysis.
Section 7 gives the conclusions drawn from this study.
4. Deep Fully Connected Neural Network Prediction Model for Blade Icing
The traditional BP neural network algorithm has the following drawbacks: 1) the requirements for feature selection are high. The introduction of irrelevant variables will increase noise data and reduce model accuracy. 2) When the number of hidden layers is increased, gradient disappearance and gradient explosion problems occur, thereby resulting in a partial optimal solution. 3) Overfitting problems occur. Various scientists have proposed the concepts of deep learning and deep neural network (DNN) on this basis.
In the blade icing prediction model based on deep FCNN, the structure and related optimization methods adopted by deep learning can overcome the above problems, which can find hidden features of deep-level WT information, and fewer iterations, and have a more powerful nonlinear fitting and self-learning ability.
Deep Neural Network Structure
The structure of the deep FCNN in this paper is basically similar to that of the BP neural network. Each neuron weights and sums the input components and selects the corresponding activation function . Too few hidden neurons and hidden layers result in a model with poor non-linear learning ability, which cannot deeply explore the hidden features of the WT information. Too many neurons and hidden layers result in a model that is highly redundant. Too many parameters are difficult to train, and at the same time may cause overfitting problems. The amount of neurons and hidden layers is mainly determined by experience and cut-and-trial method.
The internal neural network structure of the deep FCNN is shown in
Figure 4, where the relationship between layers is a fully connected relationship. The input layer neurons are set to the determined number of features, and the pre-processed WT information dataset is used as input. After the cut-and-trial method, the amount of neurons in the hidden layers 1–3 are all set to 50, and the output layer neurons are set to 2. The following formula is used to determine whether the blade is frozen.
where
is the number of neurons in layer
;
represents the output of the k-th neuron in layer
as input the output of the j-th neuron in layer
is represented by
;
denotes the weight of the k-th neuron in layer
to the j-th neuron in layer
;
is the bias of j-th neuron in layer
; and
is the activation function.
The final activation function of the output layer for the icing classification of wind power blades is the softmax function. The purpose is to convert the output value of the output layer into a probability value in the interval (0,1), which is expressed as:
where
is the softmax layer output of the i-th sample point;
is the j-th neuron output of the output layer; and
is the number of output neurons of the output layer.