The Prediction of Residual Electrical Life in Alternating Current Circuit Breakers Based on Savitzky-Golay-Long Short-Term

In order to improve the accuracy of predicting the remaining electrical life of AC circuit breakers, ensure the safe operation of electrical equipment, and reduce economic losses caused by equipment failures, this paper studies a method based on the Savitzky–Golay convolution smoothing long short-term memory neural network for predicting the electrical life of AC circuit breakers. First, a full lifespan test is conducted to obtain degradation data throughout the entire life cycle of the AC circuit breaker, from which feature parameters that effectively reflect its operational state are extracted. Next, principal component analysis and the maximum information coefficient are used to remove redundancy in the feature parameters and choose the best subset of features. Subsequently, the Savitzky–Golay convolutional smoothing algorithm is employed to smooth the feature sequence, reducing the impact of noise and outliers on the feature sequence while preserving its main trends. Then, a secondary feature extraction is performed on the smoothed feature subset to obtain the optimal secondary feature subset. Finally, the remaining electrical lifespan of the AC circuit breaker is treated as a long-term sequence problem and the long short-term memory neural network method is used for precise time-series forecasting. The proposed model outperforms backpropagation neural networks and the gate recurrent unit in terms of prediction precision, achieving an impressive 97.4% accuracy. This demonstrates the feasibility of using time-series forecasting for predicting the residual electrical lifespan of electrical equipment and provides a reference for optimizing the method of predicting remaining electrical life.

1. Introduction

When it comes to the safety and reliability of power systems, low-voltage AC circuit breakers are considered essential protective devices. They are widely used in transmission, distribution, and generation fields. Their primary function is to rapidly disconnect faulty circuits in the event of a fault, ensuring the safe operation of the power system [1]. The remaining electrical life of an AC circuit breaker reflects its breaking performance in the circuit. Accurately predicting the remaining electrical life of an AC circuit breaker can detect potential faults in a timely manner, thereby improving the reliability of the power system and reducing possible economic losses.

Therefore, analyzing the aging mechanism of circuit breakers and studying the changing trends of characteristic parameters are of significant importance for evaluating the electrical lifespan of circuit breakers, as well as assessing the performance and reliability of the entire power system [2]. By combining experimental data with mathematical models, it is possible to predict the remaining electrical lifespan of circuit breakers, providing a basis for maintenance and ensuring the normal operation of power systems.

For the prediction of the remaining electrical lifespan of low-voltage circuit breakers, commonly used approaches include those based on physical models, experimental data, and intelligent algorithms [3,4]. Physical model-based methods involve an in-depth study of the internal physical mechanisms and failure modes of circuit breakers, establishing mathematical prediction methods. Their advantages lie in detailed system modeling and parameter analysis, along with high interpretability. Experimental data-based methods utilize accumulated operational data of circuit breakers to perform statistical analysis on failure probabilities, thereby obtaining predictions of the remaining lifespan. On the other hand, intelligent algorithm-based methods rely on techniques such as machine learning and neural networks to build prediction models by learning and analyzing large amounts of data, enabling the prediction of the circuit breaker lifespan. It should be noted that the selection of appropriate methods and techniques for feature extraction and the prediction of the remaining electrical lifespan in low-voltage circuit breakers depends on practical situations and application requirements. Continuous optimization and improvement should be carried out during the utilization process. Feature extraction methods and remaining lifespan prediction methods for low-voltage circuit breakers represent important research directions in power systems. Many scholars have already proposed various methods and algorithms in this field.

F. Xing et al. improved the prediction model by considering the influence of arc duration on contact erosion, resulting in increased accuracy [5]. Liu Z et al. studied the impact of arc voltage on the electrical lifespan of circuit breakers, using Monte Carlo methods to simulate the distribution of electrical lifespan under different operations [6]. K. Huang et al. proposed a fuzzy reliability-based lifespan assessment model and an improved gray prediction method based on stepwise data sequences to predict the remaining lifespan of circuit breakers [7]. W. De et al. used gray system theory and LabVIEW programming to develop a prediction model for circuit breaker lifespan [8]. K. Li established a prediction model for electrical remaining lifespan based on the feature parameter of contact loss, assuming a degradation process following the Wiener process [9]. Li K et al. fitted cumulative arc erosion frequency empirical distribution functions and histograms with gamma cumulative curves and probability density curves, demonstrating that the degradation process of AC contactors follows a gamma process, and performed remaining lifespan prediction using the gamma process model [10]. S. Sun et al. selected contact voltage during closing and arc energy during opening as feature parameters, calculated their correlation coefficients with electrical lifespan using Spearman rank correlation, and then combined the contact voltage and arc energy proportionally to establish a univariate regression lifespan prediction model [11]. W Zeng conducted in-depth research on the degradation process of relays and established a relay performance degradation model based on linear regression theory, using the model to predict the lifespan of aerospace relays [12]. Junqiang Liu pointed out that the degradation process exhibits multi-stage characteristics due to uncertainty and proposed a multi-stage performance degradation model based on the Wiener process for aircraft engines, along with a method to predict the remaining lifespan [13].

Overall, these studies have made valuable contributions to the prediction of AC circuit breaker remaining electrical lifespan. However, there are still challenges to address, such as the need for a comprehensive representation of degradation information using a larger set of feature parameters, dealing with the nonlinear and non-smooth characteristics of electrical monitoring signals, and effectively utilizing the relationship between current and historical states to improve prediction accuracy.

With the advancement of artificial intelligence technology, the use of deep learning [14] and machine learning [15] in the field of electrical engineering is becoming increasingly widespread. Deep learning techniques have emerged as a powerful tool for building and analyzing models using deterioration data from electrical equipment. This eliminates the necessity for precise degradation statistical models and physical models, effectively addressing the challenge of estimating the remaining electrical life of complex electrical equipment.

Z.HE and colleagues used the arc voltage and arc current signals gathered during the life testing process to extract features that reflect the degradation process of miniature circuit breakers. These extracted features were used as input parameters to construct a life assessment model for miniature circuit breakers using the backpropagation neural network (BPNN). The research results indicated that the extracted relative closing time and drop-off time could significantly reflect the deterioration process of miniature circuit breakers. The experimental data testing showed that the life assessment model constructed using trend characteristic quantities achieved good evaluation results [16]. S.Sun et al. proposed a deep learning-based method for predicting the remaining life of low-voltage circuit breakers. The method involves extracting vibration segments that represent the mechanical performance of the contact system and constructing a multi-channel convolutional autoencoder long short-term memory network (MCCAE-LSTM) as the prediction model. The MCCAE captures degradation information from the time-series data, while the LSTM component quantitatively predicts the remaining mechanical life. The combination of MCCAE and LSTM enables accurate predictions and reduces the impact of data uncertainty caused by system complexity [17]. Su.JZ explored an algorithm that combines the Savitzky–Golay convolution smoothing technique with the BP neural network. This algorithm uses arc duration, arc energy, and arc phase angle as input parameters for the prediction model, allowing for the creation of a contact residual electrical life prediction model [18]. FU.H used wavelet methods to extract time-frequency domain fault-related features and constructed a CNN-LSTM network prediction model based on the extracted features [19]. Sun.SG proposed a residual life prediction algorithm based on the Modified Multi-scale Permutation Entropy and dual attention long short-term memory mechanism. The feature and time attention mechanisms assigned weights to the features and time steps, respectively, allowing for quantitative prediction of the residual life [20].

This study examined a residual electrical life prediction model for low-voltage AC circuit breakers using the SG-LSTM method. The model considers the circuit breaker’s life status as a long-term degradation sequence and takes into account the correlation between multiple feature parameters over time. First, to effectively capture the degradation information of AC circuit breakers, PCA and MIC methods were employed to select an optimal feature subset. Second, the SG convolution smoothing algorithm was applied to smooth the feature sequence and reduce the influence of noise and outliers. Finally, the LSTM model was utilized to consider the correlation between previous and subsequent states, as well as to memorize long-term historical feature information, enabling the prediction of the time series for circuit breakers. Through case studies, the proposed prediction method showed high precision and offered a novel approach for predicting the residual electrical life of switching devices.