Relay operation state monitoring system and method

Abstract

The invention relates to the field of relay state monitoring, and particularly discloses a relay operation state monitoring system and method, and the method comprises the steps: firstly obtaining the working vibration information of a relay, the voltage value of a relay measurement point, and the resistance value of the relay measurement point, and then carrying out the feature extraction and correlation analysis of the three values through a deep learning technology, and finally, obtaining a classification result through the classifier, and outputting a relay state label, thereby discovering the abnormal state of the relay as soon as possible, reducing the maintenance cost and the operation risk, and improving the safety and efficiency.

Description

Technical Field

[0001] The present application relates to the field of relay status monitoring, and more specifically, to a relay operating status monitoring system and method. Background Art

[0002] A relay is an electrical switching device that is mainly used to control the flow of current in a circuit. It realizes the switching function through the electromagnetic principle and is usually composed of a coil, contacts, and spring components. When current passes through the coil, a magnetic field is generated, which closes or opens the contacts, thereby controlling the on and off of other circuits. Relays are widely used in fields such as automation control, protection circuits, and signal transmission, and can effectively improve the safety and stability of the system. According to the working voltage and purpose, relays can be divided into low-voltage relays and high-voltage relays to meet different application requirements.

[0003] As the service time of the relay increases, the coil, contacts, and springs in the sealed cavity inside the relay will gradually age, causing the relay to fail to close or stick, or the contact level to deteriorate. Traditional methods of identifying relay failures are through regular inspections and manual inspections. However, both regular inspections and manual inspections have lags, and these inspections usually require shutdowns, which affects production efficiency.

[0004] Therefore, a system and method for monitoring the operating status of a relay are desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a relay operation status monitoring system and method, which first obtains the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally obtains the classification result through a classifier to output a relay status label, so as to detect the abnormal state of the relay early, reduce maintenance costs and operation risks, and improve safety and efficiency.

[0006] According to one aspect of the present application, a relay operation status monitoring system is provided, comprising: The relay operation data acquisition module is used to obtain the relay operation vibration information, the relay measuring point voltage value and the relay measuring point resistance value; A relay operation data extraction module is used to extract a relay operation vibration discrete sample feature vector and a relay measurement point multi-modal global correlation feature vector from the relay operation vibration information, the relay measurement point voltage value and the relay measurement point resistance value; The relay state judgment module is used to obtain the relay state label based on the discrete sample feature vector of the relay working vibration and the multi-modal global correlation feature vector of the relay measuring point mark.

[0007] According to another aspect of the present application, a method for monitoring a relay operation state is provided, comprising: Obtaining relay working vibration information, relay measuring point voltage value and relay measuring point resistance value; Extracting a relay working vibration discrete sample feature vector and a relay measuring point multi-modal global correlation feature vector from the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value; Based on the discrete sample feature vector of the relay working vibration and the multi-modal global correlation feature vector of the relay measuring point mark, a relay state label is obtained.

[0008] Compared with the prior art, the present application provides a relay operation status monitoring system and method, which first obtains the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally obtains the classification result through a classifier to output the relay status label, thereby detecting the abnormal state of the relay early, reducing maintenance costs and operating risks, and improving safety and efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 4 is a block diagram of a relay operation status monitoring system according to an embodiment of the present application.

[0011] Figure 2 It is a block diagram of a relay operation data extraction module in a relay operation status monitoring system according to an embodiment of the present application.

[0012] Figure 3 It is a block diagram of a relay working vibration information feature extraction unit in a relay operating status monitoring system according to an embodiment of the present application.

[0013] Figure 4 4 is a block diagram of a relay measuring point multimodal feature extraction unit in a relay operation status monitoring system according to an embodiment of the present application.

[0014] Figure 54 is a block diagram of a relay state judgment module in a relay operation state monitoring system according to an embodiment of the present application.

[0015] Figure 6 Flow chart of a method for monitoring the operating status of a relay according to an embodiment of the present application. DETAILED DESCRIPTION

[0016] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0017] Figure 1 Schematic diagram of the relay operation status monitoring system of the embodiment of the present application. Figure 1 As shown, according to the embodiment of the present application, the relay operation status monitoring system 100 includes: a relay operation data acquisition module 110, which is used to obtain relay working vibration information, relay measuring point voltage value and relay measuring point resistance value; a relay operation data extraction module 120, which is used to extract the relay working vibration discrete sample feature vector and the relay measuring point mark multimodal global correlation feature vector from the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value; a relay state judgment module 130, which is used to obtain the relay state label based on the relay working vibration discrete sample feature vector and the relay measuring point mark multimodal global correlation feature vector.

[0018] In the above-mentioned relay operation status monitoring system 100, the relay operation data acquisition module 110 is used to obtain the relay operation vibration information, the relay measurement point voltage value and the relay measurement point resistance value. It should be understood that the relay, as an electrical switching device, is mainly used to control the flow of current in the circuit. It uses the electromagnetic principle to realize the switching function, and is usually composed of components such as coils, contacts and springs. When the current passes through the coil, the magnetic field generated will close or disconnect the contacts, thereby controlling the on and off of other circuits. Relays are widely used in fields such as automation control, protection circuits, and signal transmission, and can effectively improve the safety and stability of the system. According to the working voltage and purpose, relays can be divided into low-voltage relays and high-voltage relays to meet different application requirements. However, as the service time of the relay increases, the coils, contacts, springs and other components in its internal sealed cavity will gradually age, resulting in the relay failing to attract, sticking or contact flatness deterioration. Traditional relay fault identification methods mainly rely on regular inspections and manual inspections. Whether it is regular inspections or manual inspections, there is a hysteresis, and these inspections usually require shutdown, which affects production efficiency.

[0019] In the technical solution of the present application, the working vibration information is obtained through vibration sensors, which can be installed on the housing or base of the relay to monitor the vibration frequency and amplitude in real time, helping to identify potential mechanical failures or abnormal conditions. The voltage value of the measuring point is collected by voltage sensors, which can monitor the changes in voltage in the circuit to ensure that the relay works within the predetermined voltage range, and abnormal voltage fluctuations may indicate electrical failures or load problems. The acquisition of the resistance value of the measuring point is obtained by a precision resistance measuring instrument to monitor the contact resistance of the relay contacts to determine whether the contacts are worn or adhered, thereby affecting its switching performance. The acquisition and analysis of these data combined can provide a comprehensive understanding of the operating status of the relay, help timely warning and maintenance, and ensure the safety and stability of the system. Therefore, in the technical solution of the present application, by obtaining the working vibration information of the relay, the voltage value of the relay measuring point, and the resistance value of the relay measuring point, and combining deep learning technology, the relay status label is output to detect the abnormal state of the relay early, thereby reducing maintenance costs and operating risks, and improving safety and efficiency.

[0020] In the above-mentioned relay operation status monitoring system 100, the relay operation data extraction module 120 is used to extract the relay operation vibration discrete sample feature vector and the relay measurement point multimodal global correlation feature vector from the relay operation vibration information, the relay measurement point voltage value and the relay measurement point resistance value. It should be understood that extracting feature vectors from the relay operation vibration information, the measurement point voltage value and the measurement point resistance value involves time series analysis and feature fusion, which can reveal the correlation between electrical characteristics and vibration characteristics, thereby achieving more efficient state judgment and fault diagnosis, and ensuring that the relay operates within a safe range. This feature extraction method not only enhances the accuracy of fault detection, but also improves the overall reliability of the system, helps to warn potential problems in advance, and reduces the risk of downtime.

[0021] Figure 2 FIG. 4 is a block diagram of a relay operation data extraction module in a relay operation status monitoring system according to an embodiment of the present application. Figure 2 As shown, in a specific embodiment of the present application, the relay operation data extraction module 120 includes: a relay working vibration information feature extraction unit 121, used to perform feature extraction on the relay working vibration information to obtain the discrete sample feature vector of the relay working vibration; a relay measuring point multimodal feature extraction unit 122, used to perform feature extraction on the relay measuring point voltage value and the relay measuring point resistance value to obtain the relay measuring point mark multimodal global correlation feature vector.

[0022] It should be understood that feature extraction of relay working vibration information is aimed at obtaining key dynamic characteristics in order to better understand the working status and potential faults of the relay. These vibration data contain information such as frequency, amplitude, phase, etc. Through feature extraction, they can be converted into discrete sample feature vectors for subsequent analysis. The feature extraction process usually includes data preprocessing, signal decomposition and feature calculation. Specifically, first, the original vibration signal is denoised to eliminate environmental interference. Then, the signal is decomposed using time-frequency analysis methods (such as short-time Fourier transform or wavelet transform) to extract meaningful frequency components. Then, features such as mean, variance, kurtosis and frequency distribution are calculated, which reflect the statistical characteristics and regularity of the vibration signal, and finally form discrete sample feature vectors. These vectors provide a basis for subsequent machine learning or fault diagnosis, helping to achieve real-time monitoring and intelligent management.

[0023] Furthermore, feature extraction of the voltage and resistance values ​​of the relay measurement points to obtain multimodal global correlation feature vectors is an important step to improve the accuracy of fault diagnosis and performance evaluation. These measurement point data contain the electrical characteristics of the relay during operation. By extracting features, the intrinsic relationship between them and vibration information can be revealed. The feature extraction process includes data preprocessing, feature calculation and fusion. Specifically, first, the voltage and resistance data are cleaned and normalized to eliminate the influence of noise and different magnitudes. Next, various statistical features such as mean, standard deviation, maximum and minimum values ​​are calculated to describe the basic behavior of voltage and resistance. At the same time, through correlation analysis, the relationship between voltage and resistance is identified, and their derived features such as ratio and rate of change are further calculated. The integration of these features forms a multimodal global correlation feature vector, which can more comprehensively reflect the working status of the relay. By combining these vectors, a more effective fault detection and early warning mechanism can be achieved, thereby ensuring that the relay operates in an efficient and safe state.

[0024] Figure 3 FIG. 4 is a block diagram of a relay operation vibration information feature extraction unit in a relay operation status monitoring system according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the relay working vibration information feature extraction unit 121 includes: a relay working vibration information preprocessing subunit 1211, used to extract vibration sample points from the relay working vibration information and arrange them to obtain a relay working vibration discrete sample input vector; a relay working vibration discrete sample convolution encoding subunit 1212, used to obtain the relay working vibration discrete sample feature vector by using a relay working vibration discrete sample convolution neural network with a one-dimensional convolution kernel.

[0025] It should be understood that extracting and arranging vibration sample points from the relay working vibration information can convert the vibration signal in the time domain into structured data suitable for algorithm processing. It is too simple. First, the original vibration signal needs to be sampled and the appropriate sampling frequency is selected to ensure that the important frequency components in the signal are captured. Then, a series of vibration sample points are extracted from the continuous signal using signal processing techniques, such as the sliding window method. The sample points in each window are regarded as an independent time series. Through this method, the timing characteristics of the signal can be maintained. Then, these sample points are arranged in time order to form discrete sample input vectors. These vectors not only contain the amplitude information of the vibration, but also reflect the changing trend of the vibration. This structured data processing method enables the monitoring system to more accurately identify potential failure modes and operating anomalies, ensuring the reliability and safety of the relay.

[0026] In a specific embodiment of the present application, the relay working vibration information preprocessing subunit 1211 is used to extract a predetermined number of vibration sample points from the relay working vibration information at a predetermined sampling frequency, and arrange the predetermined number of vibration sample points according to the time dimension as a relay working vibration discrete sample input vector. Specifically, first, follow the Nyquist sampling theorem to determine a suitable sampling frequency, that is, the sampling frequency should be at least twice the highest frequency of the signal to prevent aliasing. Subsequently, the original vibration signal is uniformly sampled using the sampling frequency to extract a predetermined number of vibration sample points. These sample points represent the working state of the relay within a specific time period. In actual operation, the sampling process usually involves the recording of timestamps to ensure the time series order of the sample points. After the sampling is completed, the extracted multiple vibration sample points are arranged in time order to form a structured discrete sample input vector. This process ensures the timing of the data and facilitates subsequent signal processing and analysis. The input vector constructed in this way can fully reflect the vibration characteristics of the relay.

[0027] Furthermore, processing the discrete sample input vector of relay working vibration through a convolutional neural network (CNN) with a one-dimensional convolution kernel is an effective method to improve the efficiency and accuracy of vibration signal analysis. Among them, the one-dimensional convolutional network is specially designed to process time series data and can automatically extract local features in the signal. Specifically, first, the input discrete sample vector is divided into multiple small windows that can cover different parts of the vibration signal. Then, the convolution kernel slides on the input vector and extracts local features, such as high-frequency and low-frequency components of the vibration signal, through convolution operations. After each convolution operation, nonlinearity is usually introduced through activation functions (such as ReLU) to enhance the expressive power of the model. In addition, the data dimension is further compressed through the pooling layer to extract more representative features. The combination of these processes enables the network to learn important feature information from complex vibration patterns, thereby generating richer feature vectors, significantly improving the sensitivity and accuracy of fault detection, and ensuring the safe and reliable operation of the relay. Specifically, the discrete sample input vector of the relay working vibration is passed through a discrete sample convolution neural network of the relay working vibration using a one-dimensional convolution kernel to obtain the discrete sample feature vector of the relay working vibration, including: using each layer of the discrete sample convolution neural network of the relay working vibration using the one-dimensional convolution kernel to respectively perform on the input data in the forward transfer of the layer: convolution processing on the input data based on the convolution kernel to generate a convolution feature map; global mean pooling processing based on the feature matrix on the convolution feature map to generate a pooled feature map; and nonlinear activation of the eigenvalues ​​at each position in the pooled feature map to generate an activated feature increase map; wherein the output of the last layer of the discrete sample convolution neural network of the relay working vibration using the one-dimensional convolution kernel is the discrete sample feature vector of the relay working vibration, the input from the second layer to the last layer of the discrete sample convolution neural network of the relay working vibration using the one-dimensional convolution kernel is the output of the previous layer, and the input of the discrete sample convolution neural network of the relay working vibration using the one-dimensional convolution kernel is the discrete sample input vector of the relay working vibration.

[0028] Figure 4 FIG. 4 is a block diagram of a multi-modal feature extraction unit for a relay measuring point in a relay operation status monitoring system according to an embodiment of the present application. Figure 4 As shown, in a specific embodiment of the present application, the relay measuring point multimodal feature extraction unit 122 includes: a relay measuring point data preprocessing subunit 1221, used to perform data preprocessing on the relay measuring point voltage value and the relay measuring point resistance value to obtain a relay measuring point multimodal association matrix; a relay measuring point multimodal convolution encoding subunit 1222, used to pass the relay measuring point multimodal association matrix through a relay measuring point multimodal convolutional neural network based on convolution kernels that are transposed of each other to obtain the relay measuring point multimodal global association feature vector.

[0029] It should be understood that data preprocessing of the voltage and resistance values ​​of the relay measuring points can clean and normalize the raw data to eliminate noise and inconsistency, making the data more suitable for subsequent analysis. Specifically, first, the cleaning stage includes removing outliers and missing data, which may be caused by sensor failure or external interference. If not processed, it will affect the reliability of the final result. Secondly, the voltage and resistance values ​​are normalized to convert data of different magnitudes to the same standard range (such as 0 to 1) to ensure the weight balance between various features in subsequent calculations. After completing data cleaning and normalization, the relationship between voltage and resistance can be calculated through correlation analysis to generate a multimodal association matrix. This matrix presents the relationship between voltage and resistance values ​​in matrix form, which can intuitively reflect their correlation and mutual influence, and facilitate subsequent model training and analysis. Using this multimodal association matrix, the working state of the relay can be more comprehensively understood, thereby improving the accuracy of fault detection and performance evaluation. In summary, data preprocessing not only provides a basis for data analysis, but also lays a solid foundation for the construction of multimodal learning and intelligent monitoring systems.

[0030] Furthermore, the multimodal association matrix combines the relationship between voltage and resistance data and can reflect the complex characteristics of the system, but the original matrix often contains redundant information and noise, so deep learning is needed for feature extraction. In this process, the transposed convolution kernels are designed to capture the interactive information between different modes. Specifically, the network first performs a convolution operation on the association matrix to extract local features. This operation is achieved through the transposed convolution kernels, which can simultaneously consider the interaction between voltage and resistance data, thereby enhancing the learning ability of the model. Subsequently, the feature map processed by the activation function will be further reduced in dimension through the pooling layer to highlight more important feature information and reduce the computational burden. Finally, the extracted features are integrated through the fully connected layer to form a global association feature vector, which summarizes the multimodal characteristics and global relationships of the measurement points, facilitating subsequent classification and regression analysis. Such feature representation not only improves the sensitivity of fault detection, but also provides strong support for performance evaluation and intelligent monitoring of relays. Specifically, the multimodal association matrix of the relay measuring point is passed through a multimodal convolutional neural network of relay measuring points based on convolution kernels that are mutually transposed to obtain a multimodal global association feature vector of the relay measuring point mark, including: each layer of the multimodal convolutional neural network of relay measuring points based on convolution kernels that are mutually transposed performs on the input data in the forward transmission of the layer: performs convolution processing, pooling processing and nonlinear processing based on a first two-dimensional convolution kernel on the input data to obtain a first activation feature map; and performs convolution processing, pooling processing and nonlinear activation processing based on a second two-dimensional convolution kernel on the first activation feature map to obtain a second activation feature map, wherein the first two-dimensional convolution kernel and the second two-dimensional convolution kernel are mutually transposed; wherein the input of the first layer of the multimodal convolutional neural network of relay measuring points based on convolution kernels that are mutually transposed is the multimodal association matrix of the relay measuring point, and the output of the last layer of the multimodal convolutional neural network of relay measuring points based on convolution kernels that are mutually transposed is the multimodal global association feature vector of the relay measuring point mark.

[0031] In a specific embodiment of the present application, the relay measuring point data preprocessing subunit 1221 is used to: arrange the relay measuring point voltage value and the relay measuring point resistance value in time to obtain a relay measuring point voltage input vector and a relay measuring point resistance input vector; calculate the transfer matrix between the relay measuring point voltage input vector and the relay measuring point resistance input vector to obtain the relay measuring point multimodal association matrix.

[0032] It should be understood that arranging the voltage and resistance values ​​of the relay measurement points in time can capture the changes in the electrical performance of the relay at different time points. Specifically, considering that the voltage and resistance values ​​of the relay will change over time during operation, these changes can reflect its operating status and health level. In order to analyze these changes, the voltage and resistance values ​​need to be recorded separately to form time series data. In actual operation, sensors can be used to monitor the voltage and resistance of the relay in real time, and data can be collected regularly (such as every second or every minute). Next, the collected voltage and resistance values ​​are sorted in time order to form two independent input vectors. The voltage input vector records the voltage value at each sampling moment, while the resistance input vector records the corresponding resistance value. This time arrangement can not only intuitively reflect the fluctuation of electrical performance, but also provide an important basis for subsequent feature extraction and pattern recognition. In the generated input vector, each element corresponds to a specific time point, so that the time series characteristics can be fully utilized in subsequent analysis.

[0033] Furthermore, the transfer matrix between the voltage input vector and the resistance input vector of the relay measuring point is calculated to obtain the multimodal association matrix of the relay measuring point, in order to analyze the relationship between the two input data, so as to better understand the overall state and failure mode of the relay. Among them, the transfer matrix can reveal the dynamic correlation between voltage and resistance at different time points, and provide rich information for subsequent feature extraction and state classification. Specifically, the voltage input vector and the resistance input vector need to be time-aligned first to ensure that there are corresponding voltage and resistance values ​​at each time point. This step usually involves data preprocessing and interpolation to ensure that the lengths of the two vectors are the same. Then, by calculating the transfer matrix between the two vectors, the changing relationship between voltage and resistance can be captured. Methods for calculating the transfer matrix usually include linear regression, correlation analysis, or dynamic time warping (DTW). These methods can quantify the relationship between voltage input and resistance input. For example, a multidimensional matrix can be formed by calculating the degree of influence of each voltage value change on the resistance value change, in which each element represents the transfer relationship between voltage and resistance. This transfer matrix reflects the dynamic characteristics and interactions of the system, laying the foundation for subsequent multimodal analysis.

[0034] Among them, calculating the transfer matrix between the relay measuring point voltage input vector and the relay measuring point resistance input vector to obtain the relay measuring point multimodal association matrix includes: deploying the Predicate mechanism into the production environment; using the Predicate mechanism to receive the relay measuring point voltage input vector and the relay measuring point resistance input vector, and returning an electrical measuring point multimodal association matrix; using the calculation logic in the Predicate mechanism to calculate the transfer matrix between the relay measuring point voltage input vector and the relay measuring point resistance input vector, and this transfer matrix will reflect the correlation between the two vectors; using the calculated transfer matrix as output to form the relay measuring point multimodal association matrix.

[0035] Part of the deployment code is shown below.

[0036] import numpy as np import tensorflow as tf from tensorflow.keras.layers import Input, Dense, Dot from tensorflow.keras.models import Model # Define Predicate mechanism def create_predicate_model(input_shape1, input_shape2): # Input layer input_voltage = Input(shape=input_shape1, name='input_voltage') input_resistance = Input(shape=input_shape2, name='input_resistance') # Fully connected layer dense_voltage = Dense(64, activation='relu')(input_voltage) dense_resistance = Dense(64, activation='relu')(input_resistance) # Calculate the transfer matrix transfer_matrix = Dot(axes=1)([dense_voltage, dense_resistance]) # Model definition model = Model(inputs=[input_voltage, input_resistance], outputs=transfer_matrix) return model # Input vector voltage_input = np.random.rand(100, 1) # Voltage values ​​at 100 time points resistance_input = np.random.rand(100, 1) # Resistance values ​​at 100 time points # Create Predicate model predicate_model = create_predicate_model((100, 1), (100, 1)) # Compile the model predicate_model.compile(optimizer='adam', loss='mse') # Calculate the multimodal association matrix multimodal_correlation_matrix = predicate_model.predict([voltage_input, resistance_input]) # Output results print("Multimodal correlation matrix of relay measuring points:") print(multimodal_correlation_matrix) The above deployment code implements a Predicate mechanism for calculating the transfer matrix between the voltage input vector of the relay measuring point and the resistance input vector of the relay measuring point. The core of this mechanism is a neural network model based on TensorFlow / Keras, in which two input layers input_voltage and input_resistance are defined to receive the voltage and resistance input vectors of the relay measuring point respectively. These input vectors are arranged in chronological order. Then, the input vectors are feature extracted through the Dense layer. dense_voltage and dense_resistance process the voltage and resistance input vectors respectively and extract their features. Next, the Dot layer is used to calculate the dot product between the voltage input vector and the resistance input vector to obtain the transfer matrix transfer_matrix. This matrix reflects the correlation between voltage and resistance. Further, the Model class is used to define the structure of the entire Predicate mechanism, with the input as the voltage and resistance vectors and the output as the transfer matrix. Then the compile method is used to compile the model, and the adam optimizer and mse loss function are selected. Finally, the code generates the voltage and resistance input vectors voltage_input and resistance_input, and calculates the multimodal correlation matrix multimodal_correlation_matrix through the predict method. In this way, an efficient multimodal data fusion mechanism is implemented to analyze the working state of the relay. The mechanism can effectively fuse the data (voltage and resistance) from different sensors, and capture the correlation between voltage and resistance by calculating the transfer matrix between them. This fusion method helps to improve the accuracy and reliability of condition monitoring. By extracting the features of the input vector through the fully connected layer and calculating the transfer matrix through the dot product operation, the correlation features between voltage and resistance can be effectively extracted and analyzed. This method performs well in processing time series data and can capture subtle correlation changes. In addition, the Predicate mechanism has high flexibility and scalability. By adjusting the structure and parameters of the fully connected layer, it can adapt to different types of input data and complex correlation relationships. In addition, the mechanism can be easily integrated into the existing deep learning framework for further optimization and expansion. In practical applications, the mechanism can be used to monitor the working state of the relay in real time and detect potential faults and abnormalities in time. By calculating the multimodal association matrix, important feature information can be provided for subsequent state classification and fault diagnosis, thereby improving the reliability and safety of the system.

[0037] In the above-mentioned relay operation status monitoring system 100, the relay state judgment module 130 is used to obtain the relay state label based on the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark. It should be understood that the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark represent the vibration characteristics and electrical characteristics of the relay respectively. By combining these two characteristics, the working state of the relay can be fully reflected. Among them, the vibration feature vector can capture the early signal of mechanical failure, while the multimodal feature vector provides in-depth information on electrical performance. In this way, the accuracy of fault detection can be improved, and the monitoring capability of the relay state in a complex working environment can be enhanced. Ultimately, accurate status labels provide a strong basis for subsequent maintenance decisions and fault predictions, ensuring the safe and reliable operation of relays.

[0038] Figure 5 FIG. 1 is a block diagram of a relay state judgment module in a relay operation state monitoring system according to an embodiment of the present application. Figure 5 As shown, in a specific embodiment of the present application, the relay state judgment module 130 includes: a relay state data feature association unit 131, which is used to optimize the feature manifold smoothness based on the eigenvalue granularity association of the discrete sample feature vector of the relay working vibration and the multi-modal global correlation feature vector of the relay measuring point mark to obtain a relay state classification feature vector; a relay state label classification unit 132, which is used to pass the relay state classification feature vector through a classifier to obtain a classification result, and the classification result is used to represent the relay state label.

[0039] It should be understood that the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark reflect the dynamic mechanical characteristics and static electrical characteristics of the relay respectively. Combining the information of the two can obtain a more comprehensive state description, thereby improving the classification ability of the model.

[0040] In particular, in the technical solution of the present application, the discrete sample feature vector of the relay working vibration is derived from the vibration information of the relay, and is extracted through a one-dimensional convolutional neural network. These features reflect the dynamic mechanical behavior generated by the relay during operation, mainly including vibration amplitude, frequency and phase information, and have strong timing and dynamic characteristics. The multimodal global correlation feature vector of the relay measuring point mark is generated based on measurement data such as voltage and resistance, and is usually arranged in the form of a time series, emphasizing the stability and change trend of electrical performance. Among them, the vibration feature emphasizes the physical motion state, while the electrical feature mainly reflects the working state and health level of the circuit. Since the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark have obvious differences in generation mechanism, data type and information content, they may have different manifold structures in the feature space. This heterogeneity may cause the mismatch of the feature manifold, thereby affecting the effect of feature fusion. The mismatch of the feature manifold means that in the multidimensional space, the relationship between the relay working vibration feature vector and the multimodal global correlation feature vector of the relay measuring point mark is not smooth enough, resulting in the lack of smoothness and consistency of the fused relay state classification feature vector. The smoothness of the feature manifold is an important indicator for evaluating the quality of feature representation. If the feature manifold is not smooth, it may cause uncertainty and misclassification in the classifier when making state judgments, ultimately affecting the accuracy of the relay state label. Therefore, in the technical solution of the present application, the feature manifold smoothness optimization based on the eigenvalue granularity association is performed on the discrete sample feature vectors of the relay working vibration and the multimodal global correlation feature vectors of the relay measuring point marks to obtain the relay state classification feature vector.

[0041] Among them, the characteristic manifold smoothness optimization based on the eigenvalue granularity association is performed on the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark to obtain the relay state classification feature vector, including: constructing a relay distance topological matrix between the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark; constructing a relay eigenvalue granularity association matrix between the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark; performing topological association modulation on the relay eigenvalue granularity association matrix based on the relay distance topological matrix to obtain a relay topological modulation association matrix; using the relay topological modulation association matrix as the perspective co-projection space, projecting the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark to the perspective co-projection space to obtain the perspective modulation relay working vibration discrete sample feature vector and the perspective modulation relay measuring point mark multimodal global correlation feature vector; fusing the perspective modulation relay working vibration discrete sample feature vector and the perspective modulation relay measuring point mark multimodal global correlation feature vector to obtain the relay state classification feature vector.

[0042] The fusion step is specifically expressed as: ; is the relay distance topology matrix between the discrete sample feature vector of the relay working vibration and the multi-modal global correlation feature vector of the relay measuring point, that is, ,and and are column vectors.

[0043] ;in, represents the discrete sample feature vector of the relay working vibration, represents the multi-modal global correlation feature vector of the relay measuring point mark, represents the relay distance topology matrix, represents the transpose of a vector, represents matrix multiplication, represents the relay eigenvalue granularity association matrix, represents the relay topology modulation incidence matrix, represents the discrete sample feature vector of the working vibration of the view modulation relay, represents the multi-modal global correlation feature vector of the view modulation relay measurement point mark, and represents the weighted hyperparameter, Represents the relay state classification feature vector.

[0044] In the technical solution of the present application, the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark have relatively significant feature heterogeneity, which results in that when the feature fusion of the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark is performed, the feature manifold and feature space mismatch caused by the heterogeneity will cause the relay state classification feature vector to fail to have feature manifold smoothness, thereby affecting the certainty of its feature expression.

[0045] Based on this, in the technical solution of the present application, the characteristic manifold smoothness optimization based on the characteristic value granularity association is performed on the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark. First, the relay distance topological matrix between the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark is constructed. This step utilizes the idea of ​​graph theory, takes the eigenvalues ​​of each position in the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark as nodes, and uses the distance metric function to determine the low-dimensional embedded expression of the edges between nodes, so as to construct a feature expression that can quantify the topological correlation relationship between the discrete sample feature vector of the relay working vibration and the multimodal global correlation feature vector of the relay measuring point mark. This step is crucial to understanding the spatial distribution of data points and their mutual connection, and also provides basic structural information for subsequent steps.

[0046] Next, a relay eigenvalue granularity association matrix is ​​constructed between the discrete sample feature vector of the relay working vibration and the multimodal global association feature vector of the relay measuring point mark. This step focuses on exploring the correlation between the internal attributes of the feature vector. In a specific example, the product between the discrete sample feature vector of the relay working vibration and the transposed vector of the multimodal global association feature vector of the relay measuring point mark is calculated to obtain the relay eigenvalue granularity association matrix. Furthermore, the relay eigenvalue granularity association matrix is ​​topologically associated modulated based on the relay distance topology matrix to obtain a relay topology modulation association matrix. That is, the eigenvalue granularity association information of the discrete sample feature vector of the relay working vibration and the multimodal global association feature vector of the relay measuring point mark is driven to wander on the distance topology map to construct a low-dimensional modulation perspective co-projection space domain for mapping the discrete sample feature vector of the relay working vibration and the multimodal global association feature vector of the relay measuring point mark to a continuous high-dimensional regression space attribute.

[0047] Next, the relay topology modulation correlation matrix is ​​used as the perspective co-projection space, and the relay working vibration discrete sample feature vector and the relay measuring point mark multi-modal global correlation feature vector are projected into the perspective co-projection space to obtain the perspective modulation relay working vibration discrete sample feature vector and the perspective modulation relay measuring point mark multi-modal global correlation feature vector. That is, the key to this technical solution is to find an implicit third space with modulation capability, in which as much information of the original high-dimensional data as possible is retained, so that the converted new feature vector not only reduces the dimension, but also maintains the relative position relationship between the original features.

[0048] Finally, the discrete sample feature vector of the working vibration of the perspective modulation relay and the multimodal global correlation feature vector of the perspective modulation relay measuring point mark are fused to obtain the relay state classification feature vector. Fusion can be achieved in various forms, ranging from simple arithmetic average to complex ensemble learning algorithms, depending on the application scenario and the goals pursued. The final relay state classification feature vector combines the main features of the two sets of input data and provides a more comprehensive data representation, which is not only conducive to improving the performance of machine learning models, but also promotes the ability of cross-modal or multi-source data analysis, thereby significantly enhancing the understanding and utilization efficiency of complex data structures.

[0049] Furthermore, the relay state classification feature vector is passed through the classifier to obtain the classification result, which mainly relies on the machine learning algorithm to identify the relationship between the input feature and the relay state, so as to assign the corresponding state label to each sample. The classifier can learn the complex mapping between features and states based on the training data, and then make predictions when faced with new and unseen data. During the training process, the prepared labeled training set is input into the classifier. Through repeated iterations and optimization of model parameters, the classifier can gradually master how to correspond the feature vector with the state label. After the model training is completed, the classifier has the ability to predict new samples. In practical applications, the relay state classification feature vector is input into the trained classifier, and the classifier will generate a classification result for the feature vector according to the learned rules. This result corresponds to a specific relay state label, such as "normal" or "fault". This labeling not only provides maintenance personnel with clear status indications, but also provides a scientific basis for subsequent maintenance and decision-making.

[0050] In summary, the embodiment of the present application first obtains the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally obtains the classification result through the classifier to output the relay status label, so as to detect the abnormal state of the relay early, reduce maintenance costs and operation risks, and improve safety and efficiency.

[0051] As described above, the relay operation status monitoring system 100 according to the embodiment of the present application can be implemented in various terminal devices. In one example, the relay operation status monitoring system 100 can be integrated into the terminal device as a software module and / or a hardware module. For example, the relay operation status monitoring system 100 can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the relay operation status monitoring system 100 can also be one of the many hardware modules of the terminal device.

[0052] Alternatively, in another example, the relay operating state monitoring system 100 and the terminal device may also be separate devices, and the relay operating state monitoring system 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in a predetermined data format.

[0053] Figure 6 Flow chart of a method for monitoring the operating status of a relay according to an embodiment of the present application. Figure 6 As shown, according to the relay operation status monitoring method of the embodiment of the present application, it includes: S110, obtaining relay working vibration information, relay measuring point voltage value and relay measuring point resistance value; S120, extracting relay working vibration discrete sample feature vector and relay measuring point mark multimodal global correlation feature vector from the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value; S130, obtaining a relay state label based on the relay working vibration discrete sample feature vector and the relay measuring point mark multimodal global correlation feature vector.

[0054] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned relay operation status monitoring method have been described in detail above. Figures 1 to 5 The description of the relay operation status monitoring system has been described in detail, and therefore, its repeated description will be omitted.

Claims

1. A relay operation status monitoring system, characterized in that: include: The relay operation data acquisition module is used to obtain the relay operation vibration information, the relay measuring point voltage value and the relay measuring point resistance value; A relay operation data extraction module is used to extract a relay operation vibration discrete sample feature vector and a relay measurement point multi-modal global correlation feature vector from the relay operation vibration information, the relay measurement point voltage value and the relay measurement point resistance value; The relay state judgment module is used to obtain the relay state label based on the discrete sample feature vector of the relay working vibration and the multi-modal global correlation feature vector of the relay measuring point mark.

2. The relay operation status monitoring system according to claim 1, characterized in that: The relay operation data extraction module comprises: A relay working vibration information feature extraction unit, used for performing feature extraction on the relay working vibration information to obtain a discrete sample feature vector of the relay working vibration; The relay measuring point multi-modal feature extraction unit is used to extract features from the relay measuring point voltage value and the relay measuring point resistance value to obtain the relay measuring point multi-modal global correlation feature vector.

3. The relay operation status monitoring system according to claim 2, characterized in that: The relay working vibration information feature extraction unit comprises: A relay working vibration information preprocessing subunit, used for extracting vibration sample points from the relay working vibration information and arranging them to obtain a relay working vibration discrete sample input vector; The relay working vibration discrete sample convolution encoding subunit is used to obtain the relay working vibration discrete sample feature vector by using the relay working vibration discrete sample convolution neural network with a one-dimensional convolution kernel to obtain the relay working vibration discrete sample feature vector.

4. The relay operation status monitoring system according to claim 3, characterized in that: The relay working vibration information preprocessing subunit is used to extract a predetermined number of vibration sample points from the relay working vibration information at a predetermined sampling frequency, and arrange the predetermined number of vibration sample points into a relay working vibration discrete sample input vector according to the time dimension.

5. The relay operation status monitoring system according to claim 4, characterized in that: The relay measuring point multimodal feature extraction unit comprises: A relay measuring point data preprocessing subunit, used for performing data preprocessing on the relay measuring point voltage value and the relay measuring point resistance value to obtain a relay measuring point multimodal association matrix; The relay measuring point multimodal convolutional encoding subunit is used to obtain the relay measuring point multimodal global correlation feature vector by passing the relay measuring point multimodal association matrix through a relay measuring point multimodal convolutional neural network based on mutually transposed convolution kernels.

6. The relay operation status monitoring system according to claim 5, characterized in that: The relay measuring point data preprocessing subunit is used for: Arranging the relay measuring point voltage value and the relay measuring point resistance value in time to obtain a relay measuring point voltage input vector and a relay measuring point resistance input vector; A transfer matrix between the voltage input vector of the relay measuring point and the resistance input vector of the relay measuring point is calculated to obtain a multi-modal association matrix of the relay measuring point.

7. The relay operation status monitoring system according to claim 6, characterized in that: The step of calculating the transfer matrix between the relay measuring point voltage input vector and the relay measuring point resistance input vector to obtain the relay measuring point multimodal association matrix includes: Deploy the Predicate mechanism to the production environment; Using the Predicate mechanism to receive the relay measuring point voltage input vector and the relay measuring point resistance input vector, and return a multi-modal association matrix of electrical measuring points; Using the calculation logic in the Predicate mechanism, a transfer matrix between the voltage input vector of the relay measuring point and the resistance input vector of the relay measuring point is calculated, and this transfer matrix will reflect the correlation between the two vectors; The calculated transfer matrix is ​​used as output to form the multimodal association matrix of the relay measuring points.

8. The relay operation status monitoring system according to claim 7, characterized in that: The relay state judgment module comprises: A relay state data feature association unit is used to optimize the feature manifold smoothness based on the eigenvalue granularity association of the relay working vibration discrete sample feature vector and the relay measuring point multi-modal global association feature vector to obtain the relay state classification feature vector; The relay state label classification unit is used to pass the relay state classification feature vector through a classifier to obtain a classification result, and the classification result is used to represent the relay state label.

9. The relay operation status monitoring system according to claim 8, characterized in that: The relay state data feature association unit comprises: Constructing a relay distance topology matrix between the discrete sample feature vectors of the relay working vibration and the multi-modal global correlation feature vectors of the relay measuring point marks; Constructing a relay eigenvalue granularity correlation matrix between the discrete sample eigenvector of the relay working vibration and the multi-modal global correlation eigenvector of the relay measuring point mark; Performing topological correlation modulation on the relay eigenvalue granularity correlation matrix based on the relay distance topological matrix to obtain a relay topological modulation correlation matrix; Taking the relay topology modulation correlation matrix as the perspective co-projection space, projecting the relay working vibration discrete sample feature vector and the relay measuring point mark multi-modal global correlation feature vector into the perspective co-projection space to obtain the perspective modulation relay working vibration discrete sample feature vector and the perspective modulation relay measuring point mark multi-modal global correlation feature vector; The relay state classification feature vector is obtained by fusing the discrete sample feature vector of the working vibration of the perspective modulation relay and the multi-modal global correlation feature vector of the measuring point mark of the perspective modulation relay.

10. A method for monitoring the operating status of a relay, characterized in that: include: Obtaining relay working vibration information, relay measuring point voltage value and relay measuring point resistance value; Extracting a relay working vibration discrete sample feature vector and a relay measuring point multi-modal global correlation feature vector from the relay working vibration information, the relay measuring point voltage value and the relay measuring point resistance value; Based on the discrete sample feature vector of the relay working vibration and the multi-modal global correlation feature vector of the relay measuring point mark, a relay state label is obtained.

Images