Aiming at the problems of multiple types of power quality composite disturbances, strong feature correlation and high recognition error rate, a method of power quality composite disturbances identification based on multi-resolution S-transform and decision tree was proposed. Firstly, according to IEEE standard, the signal models of seven single power quality disturbances and 17 combined power quality disturbances are given, and the disturbance waveform samples are generated in batches. Then, in order to improve the recognition accuracy, the adjustment factor is introduced to obtain the controllable time-frequency resolution through multi-resolution S-transform time-frequency domain analysis. On this basis, five disturbance time-frequency domain features are extracted, which quantitatively reflect the characteristics of the analyzed power quality disturbance signal, which is less than the traditional method based on S-transform. Finally, three classifiers such as K-nearest neighbor, support vector machine and decision tree algorithm are used to effectively complete the identification of power quality composite disturbances. Simulation results show that the classification accuracy of decision tree algorithm is higher than that of K-nearest neighbor and support vector machine. Finally, the proposed method is compared with other commonly used recognition algorithms. Experimental results show that the proposed method is effective in terms of detection accuracy, especially for combined PQ interference.
With the continuous development of power grid, a variety of nonlinear, impact loads and power electronic equipment are constantly increasing, and the power quality problem is becoming increasingly serious, which has become a highly concerned issue in the world [
In the complex power grid environment, the power quality disturbance signal is non-stationary, and most of the disturbances are represented by the combination of multiple disturbances, that is, the composite power quality disturbance. The characteristic quantity of the composite disturbance is not simply superimposed by a single power quality disturbance, but cross coupling between time-frequency domain characteristics, which requires the signal analysis method to have high efficiency [
Generally speaking, PQD is mainly composed of three parts: disturbance signal detection, feature extraction and disturbance identification as shown in
In order to realize accurate identification of power quality composite disturbances, this paper proposes a power quality disturbance signal identification method based on multi-resolution S-transform (MST) and decision tree. MST algorithm is an improved algorithm of ST, which can achieve controllable time-frequency resolution through parameter adjustment window function. In addition, this method has been successfully applied to the research of seismic signal processing. MST was used to analyze the time-frequency characteristics of PQ signals, and five characteristic statistics of each PQ signal were obtained. On this basis, DT classifier is used to classify PQ signals according to feature statistics. Due to the effectiveness of MST, the efficiency of DT classifier can be guaranteed. And compared with other power quality disturbance identification algorithms, simulation and comparison results show that the proposed algorithm can accurately and quickly classify and identify power quality disturbance signals.
According to IEEE-1159 and literature [
Lable | Disturbance class | Modeling equations | Equations’ parameters |
---|---|---|---|
S1 | Sag | ||
S2 | Swell | ||
S3 | Interrupt | ||
S4 | Flicker | ||
S5 | Impulsive Transient (IT) | ||
S6 | Oscillatory Transient (OT) | ||
S7 | Harmonic |
According to the above power quality disturbance signal model, Matlab 2018a is used to simulate the power quality signal, and the waveform is shown in
From the perspective of single disturbance, the characteristics of voltage sag, rise, interruption and flicker are mainly concentrated in the low-frequency part, and the fundamental frequency distortion characteristics are different; Harmonics are mainly concentrated in the intermediate frequency part; The oscillation is mainly concentrated in the high-frequency part. The composite disturbance has the characteristics of different disturbance types at the same time, which are reflected in different frequency domains. Therefore, features can be extracted from different frequency domains for disturbance recognition. Compared with other methods, feature extraction in frequency domain can obtain more targeted features through the multi-resolution characteristics of MST.
According to Heisenberg uncertainty principle, the time width and frequency width of the signal cannot tend to infinity at the same time, and for a specific given signal, the product of time width and frequency width is a constant. The misjudgment of voltage interruption signal as voltage sag signal in document [
The discrete form of MST is expressed as
The corresponding relationship between
From the perspective of disturbance analysis, low-frequency disturbance analysis includes signal amplitude change and start and end point positioning, which requires MST to have a higher time resolution; The analysis purpose of intermediate frequency disturbances such as harmonics is to determine whether the signal contains harmonic components, which requires higher frequency domain resolution; High frequency features are used to identify oscillations and compound disturbances with oscillations, and the influence of high-frequency energy and noise of other types of disturbance signals should be avoided. Therefore, the setting values of window width adjustment factors in different frequency domains need to be introduced, respectively.
Power quality disturbances can be mainly divided into instantaneous disturbances and steady-state disturbances. Instantaneous disturbances refer to unexpected changes in the amplitude characteristics of disturbance signals, such as voltage sag, voltage sag and voltage interruption, while steady-state disturbances represent significant degradation of the components of disturbance signals, such as voltage flicker and harmonics. Specifically, the characteristics of flicker interference are always included in the low frequency, while the characteristics of sag, surge and interruption signal can be reflected by the fundamental frequency content (50 or 60 Hz). In addition, when analyzing frequency components from 100 to 700 Hz and above 700 Hz, harmonic and transient interference should be detected respectively.
Time frequency analysis of power quality signals the two-dimensional matrix of MST contains a large amount of signal time-frequency information, and the characteristics of each power quality signal should be reflected by its energy and frequency characteristics. In order to effectively extract the time-frequency information of the signal, the following five features are extracted according to the distribution characteristics of power quality disturbances in the time-frequency domain:
Feature 1: minimum amplitude of fundamental frequency component
Feature 2: maximum amplitude of fundamental frequency component
Features 1 and 2 describe the evolution trend of the fundamental frequency component. The energy distribution of power quality signal shows obvious expansion and depression trends, which can accurately describe the characteristics of power quality signal.
Feature 3: correlation coefficient of fundamental frequency component, periodic fluctuation mode is an important feature for detecting flicker events. In addition, scintillation events with different amplitude ranges have similar characteristic fluctuation characteristics. Therefore, flicker events should be effectively identified by checking whether there are obvious periodic fluctuations. On this basis, the correlation coefficient between the scintillation event and the fundamental frequency component is defined as feature 3, and the expression is as follows:
Feature 4: energy of row vector corresponding to frequency
Feature 5: root mean square of row vector corresponding to frequency
The time-frequency diagram of the composite disturbance waveform formed by the superposition of two different kinds of disturbances after multi-resolution S-transform can accurately distinguish the transient component and the steady-state component at the same time, and can visually see the amplitude changes of the disturbance signal in time domain and frequency domain. Taking ten typical PQDS as examples, the time-frequency three-dimensional view of the results obtained by using MST fast algorithm is shown in
In
In order to identify power quality events such as voltage sag, voltage sag, voltage interruption, voltage flicker, voltage harmonics, voltage sag with harmonics, voltage sag with harmonics and voltage interruption with harmonics, five disturbance time-frequency characteristics are extracted through multi-resolution S-transform time-frequency domain analysis. In addition, three classification methods are studied to identify power quality composite disturbance signals. These classifiers are k-nearest neighbor algorithm (KNN), support vector machine algorithm (SVM) and decision tree algorithm (DT).
K-nearest neighbor (KNN) algorithm is one of the simplest methods in data mining classification technology, and has been widely used in many fields. The core idea of KNN algorithm is that if most of the k nearest samples in the feature space belong to a certain category, the sample also belongs to this category and has the characteristics of samples in this category. In determining the classification decision, this method only determines the category of the samples to be divided according to the category of the nearest one or several samples. KNN algorithm is only related to a very small number of adjacent samples in class decision-making. KNN algorithm assigns space to the nearest neighbor training data to classify the new test data and identify single and complex interference [
Parameter
Support vector machine (SVM) is a data mining method based on statistical learning theory, which can successfully deal with many problems such as regression problems (time series analysis) and pattern recognition (classification problems, discriminant analysis) [
In order to find the location of the separation hyperplane, the following data sets must be considered:
The construction idea of decision tree (DT) is as follows: if all the samples in the training sample set are of the same kind, they will be regarded as Leaf nodes; otherwise, according to certain branch division rules, the sample set will be subdivided successively until Leaf nodes. For the same sample set, many decision trees can be generated, and branching rule is crucial to obtain an “optimal” tree [
Gain ratio:
Gini index:
The input data to a decision tree algorithm is a data set that includes symbols and the data entered with their symbolic values. Based on this problem, the decision tree classifier can induce a set of data, which is represented as a tree by distribution. In addition, when generating decision trees, only symbols that are sufficiently dependent on the classification problem are selected. When learning a decision tree, the tree is used to predict new sample results. In addition, the decision tree learning algorithm is called supervised learning method, and each sample in the dataset is classified according to a specific category [
Decision tree (DT) classification is based on decision rules, so binary tree graphs are used to discover correlations between input and output components. The decision tree consists of internal nodes, branches, and terminal nodes. Internal nodes display tests on symbols, branches represent test results, and terminal nodes define class labels. At each node, decisions are made based on the rules obtained from the data. Decision tree learning is a predictive model that describes the observations of a project to arrive at a target number for the project. The method is also used in data mining, machine learning and statistics. Compared with other classification methods, decision tree is relatively fast and more accurate.
In addition, when constructing the training set and test set of disturbance recognition, the whole feature sample set of disturbance recognition was randomly divided into two independent parts with a ratio of 70% to 30%. In this way, random extraction and random validation can ensure that the model trained on the training dataset does not underfit. Due to the relative independence of the validation set and the training set, overfitting of the model to the training data can be avoided to a certain extent. Compared with other two algorithms, the results show that DT algorithm can accurately and quickly classify and identify power quality disturbance signals.
According to IEEE-1459 standard and literature, 24 power quality disturbance databases are automatically generated by using MATLAB 2018a software. The database contains 7 single disturbances and 17 composite disturbances. The sampling frequency of the research signal is 3.2 kHz and the sampling time is 0.4 s. During the simulation, the setting parameters of the disturbance signal are randomly generated within the required range.
Feature extraction method | Classification method | PQDs | Characteristic number | Classification accuracy/% |
---|---|---|---|---|
ST [ |
PNN [ |
11 | 4 | 97.4 |
MST [ |
DT [ |
16 | 5 | 99.97 |
ST [ |
Chaotic integrated decision tree [ |
23 | 9 | 91.9 |
MST | DT | 24 | 5 | 97.5 |
It can be seen from
In this paper, a method for identifying power quality composite disturbances based on MST and DT is proposed. The 24 power quality disturbance signal models and their waveforms were analyzed in the time-frequency domain (MST). According to the characteristics of different types of disturbances, five kinds of disturbance features were extracted. KNN, SVM and DT classifiers were used to effectively identify complex power quality composite disturbances. The following conclusions are obtained:
Compared with the traditional S-transform, MST only needs to extract 5 feature statistics, which is less than most other popular methods, and the calculation cost is relatively small. Three classifiers are used for testing, and it is found that the DT algorithm has the highest accuracy (97.5%). In conclusion, the proposed method can effectively identify power quality disturbances, especially for power quality composite disturbances. The characteristics of the composite disturbances with transient pulse power quality are strongly correlated, and the identification error rate of the related composite disturbances is high, and the identification difficulty is great. The design optimization of the related features still needs to be further studied in the future.
This work was financially supported by National Natural Science Foundation of China (No. 52067013); the Key Natural Science Fund Project of Gansu Provincial Department of Science and Technology (No. 21JR7RA280); the Tianyou Innovation Team Science Foundation of Intelligent Power supply and State Perception for Rail Transit (No. TY202010); and the Natural Science Foundation of Gansu Province (No. 20JR5RA395).
The
The authors declare that they have no conflicts of interest to report regarding the present study.