WO2024020960A1

WO2024020960A1 - Method for predicting remaining useful life of circuit breaker, and electronic device and storage medium

Info

Publication number: WO2024020960A1
Application number: PCT/CN2022/108739
Authority: WO
Inventors: 庞建国; 刘臻; 陈维刚
Original assignee: 西门子股份公司; 西门子（中国）有限公司
Priority date: 2022-07-28
Filing date: 2022-07-28
Publication date: 2024-02-01

Abstract

A method for predicting the remaining useful life of a circuit breaker, and an electronic device and a storage medium. The method comprises: determining parameter values of aging indexes of a circuit breaker, which parameter values are collected within at least two time periods (101); on the basis of the parameter values, executing a training process on a neural network model N times, so as to obtain N aging prediction models, wherein the aging prediction models are used for predicting an aging state of the circuit breaker, and during each training process, the parameter values are taken as training data during the current training process, and are input into the neural network model, such that the neural network model is trained into an aging prediction model obtained during the current training process, N being a positive integer greater than or equal to 2 (102); selecting, from among the N aging prediction models, an aging prediction model that meets a predetermined condition (103); and predicting the remaining useful life of the circuit breaker on the basis of the selected aging prediction model (104). Time costs and resource costs for modeling are saved on.

Description

Method, electronic device and storage medium for predicting remaining service life of circuit breaker

Technical field

The present invention relates to the technical field of circuit breakers, in particular to methods, electronic devices and storage media for predicting the remaining service life (Remaining Useful Life, RUL) of circuit breakers.

Background technique

The circuit breaker is an important switching device in the electrical system of power plants and substations. In the system electrical wiring, isolation switches should be installed on both sides to cooperate with it. Its main functions are: switching operating modes during normal operation, connecting equipment or lines to the power grid or withdrawing from operation, and playing a control role; when equipment and lines fail, quickly remove the faulty loop to ensure normal operation of the non-faulty parts, and play a protective role . RUL refers to the length of time a machine will run before it is repaired or replaced. With RUL, engineers can schedule maintenance, optimize operational efficiency and avoid unplanned downtime.

To ensure circuit breaker reliability, predictive maintenance can be employed, which is a more effective maintenance strategy than reactive and preventive maintenance. Predictive maintenance is scheduled based on diagnostic assessment of equipment life, environmental stress and other factors. The RUL of the circuit breaker is the key to predictive maintenance of the circuit breaker.

Currently, related technologies for predicting the RUL of circuit breakers generally involve complex modeling processes for circuit breakers. However, the modeling process consumes a lot of modeling resources.

Contents of the invention

The embodiment of the present invention provides a method, electronic device and storage medium for predicting the RUL of a circuit breaker.

The first aspect is to provide a method for predicting the RUL of a circuit breaker. Methods include:

Determine the parameter values of the aging indicators of the circuit breaker collected during at least two time periods;

Based on the parameter values, perform N training processes on the neural network model to obtain N aging prediction models. The aging prediction models are used to predict the aging state of the circuit breaker; wherein in each training process: Parameter values are input into the neural network model as training data in this training process, so as to train the neural network model as the aging prediction model obtained in this training process; wherein N is a positive integer of at least 2;

Select an aging prediction model that meets predetermined conditions from the N aging prediction models;

Based on the selected aging prediction model, the RUL of the circuit breaker is predicted.

In a second aspect, an electronic device is provided. Electronic equipment includes:

processor;

memory for storing executable instructions for the processor;

The processor is configured to read the executable instructions from the memory and execute the executable instructions to implement the above method of predicting the RUL of a circuit breaker.

In a third aspect, a computer-readable storage medium is provided with computer instructions stored thereon. The computer instructions, when executed by the processor, implement the above method of predicting the RUL of the circuit breaker.

A fourth aspect provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the above method of predicting the RUL of a circuit breaker.

It can be seen that there is no need to model the circuit breaker. The neural network model is trained multiple times through the parameter values of the aging index to obtain multiple aging prediction models, and the aging of the RUL is predicted from the multiple aging prediction models according to predetermined conditions. Prediction model, and then use the selected aging prediction model to predict RUL, saving the time cost and resource cost of modeling.

Description of drawings

Preferred embodiments of the present invention will be described in detail below to make the above and other features and advantages of the present invention more apparent to those skilled in the art with reference to the accompanying drawings, in which:

Figure 1 is a flow chart of a method for predicting RUL of a circuit breaker according to an embodiment of the present invention.

Figure 2 is a schematic diagram of numerical interpolation according to an embodiment of the present invention.

Figure 3 is a schematic diagram of data preprocessing according to an embodiment of the present invention.

4 is a schematic flowchart of an exemplary process for determining the service life in an iterative manner according to an embodiment of the present invention.

Figure 5 is a schematic diagram of selecting an aging prediction model according to an embodiment of the present invention.

FIG. 6 is a schematic diagram of determining a RUL prediction curve using a time period when a circuit breaker is predicted to fail according to an embodiment of the present invention.

7 is a schematic diagram of an exemplary process for predicting RUL of a circuit breaker according to an embodiment of the present invention.

FIG. 8 is a structural diagram of an electronic device according to an embodiment of the present invention.

Among them, the reference signs are as follows:

标号label	含义meaning
100100	预测断路器的RUL的方法Methods for Predicting RUL of Circuit Breakers
101～104101～104	步骤step
21twenty one	电子脱扣器1的检测点Detection point of electronic trip unit 1
22twenty two	电子脱扣器2的检测点Detection point of electronic trip unit 2
23twenty three	电子脱扣器3的检测点Detection point of electronic trip unit 3
24twenty four	电子脱扣器1的插值曲线Interpolation curve of electronic trip unit 1
2525	电子脱扣器2的插值曲线Interpolation curve of electronic trip unit 2
2626	电子脱扣器3的插值曲线Interpolation curve of electronic trip unit 3
401～405401～405	步骤 step
5151	老化预测曲线 Aging prediction curve
5252	直方图 Histogram
5353	门限线 threshold line
6161	第一测量点 first measuring point
6262	第二测量点 second measuring point
6363	RUL预测曲线 RUL prediction curve
7070	测量 Measurement
7171	数据预处理Data preprocessing
7272	数据 data
7373	训练过程 training process
7474	老化预测模型 Aging prediction model
7575	使用寿命预测过程Service life prediction process
7676	使用寿命 Service life
7777	老化预测模型选择过程Aging Prediction Model Selection Process

7878	选中的老化预测模型Selected aging prediction model
7979	实时测量数据Real-time measurement data
8080	老化预测模型 Aging prediction model
8181	使用寿命预测过程Service life prediction process
8282	RUL预测过程 RUL prediction process
8383	预测的RULPredicted RUL
600600	电子设备 Electronic equipment
601601	处理器 processor
602602	存储器memory

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention clearer, the following examples are given to further describe the present invention in detail.

For the sake of simplicity and intuitiveness in description, the solution of the present invention is explained below by describing several representative embodiments. A large number of details in the embodiments are only used to help understand the solution of the present invention. However, it is obvious that the technical solution of the present invention may not be limited to these details when implemented. In order to avoid unnecessarily obscuring the solutions of the present invention, some embodiments are not described in detail, but only give a framework. Hereinafter, "including" means "including but not limited to", and "based on..." means "at least based on..., but not limited to only based on...". Due to Chinese language habits, when the number of a component is not specified below, it means that the component can be one or more, or it can be understood as at least one.

The applicant found that: in the prior art, in order to predict the RUL of a circuit breaker, it is necessary to perform mathematical modeling or mechanism modeling of the circuit breaker, and then use the modeled circuit breaker and various input parameters to predict the RUL. Specifically, the model of the circuit breaker is an abstraction of the circuit breaker, in which: it can be modeled according to the corresponding mechanism by analyzing the movement rules of the circuit breaker itself; it can also be modeled through experiments on the circuit breaker or Processing of statistical data and modeling based on existing knowledge and experience about circuit breakers. However, regardless of the modeling method, building a circuit breaker model consumes a lot of resources.

In the embodiment of the present invention, there is no need to model the circuit breaker, but the RUL of the circuit breaker is predicted through deep learning.

The concept of deep learning originates from the research of artificial neural networks. A multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning discovers distributed feature representations of data by combining low-level features to form more abstract high-level representation attribute categories or features. The motivation for studying deep learning is to build neural networks that simulate the human brain for analytical learning.

As shown in Figure 1, the method 100 includes:

Step 101: Determine the parameter value of the aging indicator of the circuit breaker collected in at least two time periods.

First, determine the aging index that can characterize the aging degree of the circuit breaker. For example, aging indicators can be implemented as electrical properties of key components related to the circuit breaker that have a critical impact on the life of the circuit breaker. For example, the key components can be the energy storage capacitor in the electronic trip unit (ETU) of the circuit breaker, or the Metal-Oxide-Semiconductor Field-Effect Transistor in the power drive circuit of the circuit breaker. MOSFET), etc. Electrical properties can be implemented as voltage, current, capacitance, or electrical power, among others.

In one embodiment, the aging indicators of the circuit breaker may include: the capacitance value of the energy storage capacitor in the power circuit of the electronic trip unit of the circuit breaker; the on-resistance of the MOSFET in the power drive circuit of the circuit breaker, etc.

The above exemplarily describes the aging indicators that represent the aging degree of the circuit breaker. Those skilled in the art can realize that this description is only exemplary and is not used to limit the protection scope of the embodiments of the present invention.

Collect parameter values of the aging index in at least two time periods to obtain at least two parameter values, where each parameter value corresponds to a respective time period. For example, the parameter values of the aging indicators can be collected periodically according to a predetermined time period sequence. For example, the circuit breaker is arranged in a predetermined environment (such as a high temperature environment), and then data is collected for the circuit breaker once in the first time period (i.e., the first collection moment) to obtain the first parameter value; then, in the second time period, (That is, the second acquisition moment after the first acquisition moment and based on the same time period increment) Collect data again for the circuit breaker to obtain the second parameter value, and so on, you can obtain a sequence corresponding to the time period. Series parameter values. It can be seen that each parameter value corresponds to the time period when the respective data is collected.

Step 102: Based on the parameter values, perform N training processes on the neural network model to obtain N aging prediction models. The aging prediction models are used to predict the aging state of the circuit breaker; in each training process: The parameter values are input into the neural network model as training data in this training process, so that the neural network model is trained as the aging prediction model obtained in this training process; wherein N is a positive integer of at least 2.

The neural network model can be implemented as a feedforward neural network (FFNN) model or a feedback neural network model (Feedback Neural Network). Among them, feedforward neural network models can include: convolutional neural network (CNN) model, fully connected neural network (FCN) model, generative adversarial network (GAN) model, etc. Feedback neural networks can include: recurrent neural network (RNN) models, long short-term memory network (LSTM) models, Hopfield network models and Boltzmann machines, etc.

Preferably, the embodiment of the present invention can use a feedforward neural network model to train an aging prediction model. The input of the aging prediction model is the parameter value that represents the aging state in the previous time period (the current time or before the current time), and the output of the aging prediction model is the parameter value that represents the aging state in the subsequent time period (after the current time). For example, an aging prediction model can predict parameter values for one or more time periods after the previous time period.

Here, since the training process is performed N times separately on the neural network models with the same structure, N aging prediction models can be obtained. At the beginning of each training process, the initial values of network parameters (such as weights) in the neural network model are random, so multiple training processes will produce multiple aging prediction models.

In one embodiment, the parameter values include a first parameter value collected in a first time period (which can be used as a training sample) and a second parameter value collected in a second time period (which can be used as a label for a training sample), where , the second time period is after the first time period, and the second time period is the next time period of the first time period; training the neural network model as the aging prediction model obtained during this training process includes: changing the first parameter value Input to the neural network model, so that the neural network model outputs the prediction parameter value of the second time period; based on the difference between the prediction parameter value of the second time period and the second parameter value, determine the loss function value of the neural network model; Configure the model parameters of the neural network model to perform iteration (for example: based on the configured model parameters, continue to use the subsequently collected parameter values of the third time period to predict the prediction parameter values of the fourth time period, or still use the first time period The parameter value of the second time period is predicted again (predicted parameter value of the second time period) until the loss function value converges to less than the preset threshold, and the iteration is stopped; the configured neural network model is determined as the aging prediction model.

The above training process of the neural network model is executed N times (the initial weight of the neural network model before each training is a random value). That is, for neural network models with the same structure whose initial values are random, N training processes are performed respectively, thereby obtaining N aging prediction models.

The above describes the specific process of training the neural network model using parameter values of two time periods. In fact, the embodiments of the present invention are also applicable to situations where parameter values of more than two time periods are used to train a neural network model. In the process of training a neural network model using parameter values of more than two time periods, the input data of the neural network model can be parameter values of three, four or more time periods, and the output of the neural network model is the input data. Forecast parameter values for the latest time period in . Similarly, the loss function value is determined based on the difference between the prediction parameter value of the latest time period and the parameter value of the latest time period in the input data, and the model parameters of the neural network model are configured based on the loss function value until the loss function Training is completed after the value converges below the preset threshold.

In one embodiment, before performing the training process on the neural network model based on the parameter values, it may also include performing preprocessing on the parameter values. In preprocessing, interpolation, sliding window segmentation and grouping are used to expand the data volume of parameter values and prepare the data in an appropriate format for the next step of the neural network. Data input for preprocessing can come from laboratory testing or from the manufacturer. Due to different data formats from laboratory tests or manufacturers, it is usually not suitable for training neural network models. Interpolation can even out the period of data points, which usually expands the data size and facilitates the training of FFNN.

Preprocessing includes at least one of the following:

(1). Interpolate parameter values.

That is, a continuous function is interpolated based on discrete parameter values so that this continuous curve passes through all given discrete parameter values. For example, interpolation algorithms such as polynomial interpolation, piecewise interpolation, and spline interpolation can be used to interpolate parameter values to obtain more parameter values.

Figure 2 is a schematic diagram of numerical interpolation according to an embodiment of the present invention. Take the detection of the energy storage capacitor of the electronic trip unit as an example to illustrate.

In Figure 2, the detection points of the electronic tripper 1 (each detection point measures the capacitance value of the energy storage capacitor once) include multiple detection points 21; the detection points of the electronic tripper 2 include There are a plurality of detection points, each of which is a plurality of detection points 22; the detection points of the electronic trip device 3 include a plurality of detection points, each of which is a plurality of detection points 23. Various interpolation algorithms can be used: multiple detection points 21 are used to interpolate the interpolation curve 24 of the electronic release 1; multiple detection points 22 are used to interpolate the interpolation curve 25 of the electronic release 2; multiple detection points 23 are used to interpolate. The interpolation curve 26 of the electronic release 3 is obtained. It can be seen that the amount of data before interpolation is small and the measurement intervals are different, but after interpolation, the amount of data is expanded by more than 10 times, and the sample interval is equal to 1 day.

(2) Perform sliding window segmentation on parameter values.

After interpolation, the data of each electronic trip unit is divided into sliding windows respectively, and the data format is prepared according to the input size of the neural network, as shown in the following table:

输入enter	1×L1×L
输出output	[Nsw×Lsw][Nsw×Lsw]

Table 1

Among them, L is the length of the interpolation data, Nsw is the input size of the neural network, Lsw=L-Nsw.

(3) Group parameter values.

Grouping is used to collect data from all electronic trip units.

输入enter	[Nsw×Lsw] _i，i＝1...N_Units [Nsw×Lsw] _i , i＝1...N_Units
输出output	[Nsw×LG][Nsw×LG]

Table 2

in,

N_Units is the number of electronic trip units; i is the serial number of the electronic trip unit; (Lsw) _i is the data length after dividing the i-th electronic trip unit. Figure 3 is a schematic diagram of data preprocessing according to an embodiment of the present invention. Among them, N_Units is 3, and the three electronic trippers are u1, u2 and u3 respectively. First, interpolate the data of each electronic release. For example, for the electronic release u1, the collected data of the i-th time period is m ₁ [i]. You can use the collected m ₁ [1], m ₁ [3], m ₁ [4] interpolate the uncollected m ₁ [2] and m ₁ [5], where the collected m ₁ [6] is the label), and then perform sliding window segmentation on the interpolated data (For example, m ₁ [1], m ₁ [2], m ₁ [3], m ₁ [4], m ₁ [5] are the first window data of the electronic trip unit u1, and m ₁ [6] is The labels of the first window; m ₁ [2], m ₁ [3], m ₁ [4], m ₁ [5] and m ₁ [6] are the second window data of the electronic trip unit u1, m ₁ [ 7] for the label of the second window, etc.). Next, group all electronic trip unit data. Each electronic trip unit has L equal to 340 and Nsw equal to 5. The above data is used to train an artificial neural network model, preferably FFNN, and the trained model is an aging prediction model that uses the previous aging state to predict the next aging state.

The above exemplarily describes typical examples of data preprocessing. Those skilled in the art can realize that this description is only exemplary and is not used to limit the protection scope of the embodiments of the present invention.

Step 103: Select an aging prediction model that meets predetermined conditions from N aging prediction models.

In one embodiment, the predetermined conditions include lifespan filter conditions. For example: the service life screening conditions include: the predicted service life of the circuit breaker is within a predetermined interval; the predicted service life of the circuit breaker is calculated by performing histogram statistics on the N service life obtained by N aging prediction models. The median value within a certain interval containing the most useful lives, etc. Here, the meaning of service life is: the full life cycle time of the circuit breaker (including the past life).

Preferably, the parameter value also includes the third parameter value collected closest to the current time period. Step 103 includes: inputting the third parameter value into N aging prediction models respectively, and each aging prediction model determines the service life of its respective output based on the iterative process performed by each, thereby obtaining N service life; from the N service life Determine the service life that meets the service life screening conditions; output the aging prediction model of the determined service life and determine it as the aging prediction model that meets the predetermined conditions.

In an exemplary embodiment, the service life screening condition includes at least one of the following: the service life is within a predetermined interval (for example, the service life is within an interval of ±10% of the expected life (for example, 20 years)); The service life is the median value within the interval containing the most service life determined based on histogram statistics.

For example, assuming that N is equal to 5, input the third parameter value into 5 aging prediction models respectively to obtain 5 service lives. Then select the service life that satisfies the interval of 20 years ±10% from these five service lives, assuming it to be service life 3, and then output the aging prediction model with service life 3 to determine it as the aging prediction that meets the predetermined conditions. Model.

The specific process for each aging prediction model to determine the service life of its respective output based on the iterative process performed by each one includes: inputting the third parameter value as an input value to the aging prediction model, so that the aging prediction model outputs a time period for the third parameter value. The parameter prediction value of the next time period; when the parameter prediction value of the next time period is greater than the preset parameter threshold value, the parameter prediction value of the next time period is used as the input value to iterate back to the aging prediction model until the aging prediction The parameter prediction value output by the model is less than or equal to the parameter threshold value; the service life is determined based on the time period of the parameter prediction value output by the aging prediction model at the end of the iteration.

After training multiple aging prediction models, each aging prediction model can be used to predict the service life of the circuit breaker. For each aging prediction model, the lifetime can be calculated by propagating the aging state until a parameter threshold value (eg, 80% of the initial capacitance value) is reached. Specifically, it includes: Step (1): Use the parameter value of the aging index obtained from the last measurement as the initial value of the aging prediction model; Step (2): The aging prediction model predicts the parameter value of the aging index in the next time period; Step (3) ): If the predicted parameter value is less than the threshold (for example, less than 80% of the initial capacitance value), it means that the circuit breaker has failed, then go to step (4), otherwise, go to step (2). Step (4), use the number of iterations to estimate the service life of the circuit breaker.

As shown in Figure 4, an exemplary process for iteratively determining service life includes:

Step 401: Enter the recently measured parameter value of the aging index of the circuit breaker, which is called the original parameter value. At this time, the number of iterations N is equal to 0.

Step 402: Input the original parameter value into the trained aging prediction model, and the aging prediction model outputs the parameter value of the next time period corresponding to the original parameter value, which is called the prediction parameter value.

Step 403: Determine whether the prediction parameter value is less than the preset parameter threshold value. If yes, execute step 405 (corresponding to the "Y" branch); otherwise, execute step 404 (corresponding to the "N" branch).

Step 404: Provide the prediction parameter value to the aging prediction model as the next round input value of the aging prediction model, increase N by one, that is, execute N++, and return to step 402.

Step 405: Determine the service life based on the number of iterations N. For example, the number of iterations N times the time period, plus the time period of the original parameter value, is the service life.

Since there are N aging prediction models, N multiple aging prediction curves (including the correspondence between service life and parameter values) can be calculated through the process of Figure 4 above.

As can be seen from FIG. 5 , multiple aging prediction models generate multiple aging prediction curves 51 , where each aging prediction curve 51 corresponds to a respective aging prediction model. Each aging prediction curve 51 includes a corresponding relationship between parameter values and respective predicted service life. In FIG. 5 , the threshold line 53 is a straight line with 80% of the initial parameter value (for example, the initial capacitance value: 50.125 microFarads). For example, the threshold line 53 corresponds to 40.1 microFarads. The intersection of each aging prediction curve 51 and the threshold line 53 is the time point at which the circuit breaker fails (ie, the service life) predicted by the aging prediction model. It can be seen that the multiple aging prediction curves 51 and the threshold lines 53 have multiple intersection points.

The service life filtering conditions include sub-condition 1: 20 years ±10%, and the aging prediction curves that do not meet this sub-condition 1 are filtered out. Furthermore, in filtering out the remaining aging prediction curves that do not satisfy sub-condition 1, a histogram 52 is drawn in a coordinate system with the service life as the abscissa and the number of aging prediction curves as the ordinate. In the histogram 52, the maximum value is 17, and the abscissa interval of the maximum value 17 is [20.9921.22], which means that there are 17 curves passing through the threshold line 53 between the lifetime [20.9921.22], which The median of the 17 time values is 21.1055. The aging prediction model corresponding to the aging prediction curve that meets the median value is determined as the selected aging prediction model.

The above exemplarily describes a typical example of selecting an aging prediction model. Those skilled in the art can realize that this description is only exemplary and is not used to limit the protection scope of the embodiments of the present invention.

Step 104: Predict the remaining service life of the circuit breaker based on the selected aging prediction model.

In one embodiment, based on the selected aging prediction model, predicting the remaining service life of the circuit breaker includes: inputting the parameter values collected during the test time period (for example, the test time period is the current moment) into the aging prediction model to predict the remaining service life of the circuit breaker based on the aging prediction model. The prediction model outputs the parameter prediction value of the next time period of the test time period; when the parameter prediction value of the next time period is greater than the preset parameter threshold value, the parameter prediction value of the next time period is used as the input value to iterate back to the aging Prediction model until the parameter prediction value output by the aging prediction model is less than or equal to the parameter threshold value; based on the difference between the time period of the parameter prediction value output by the aging prediction model and the test time period at the end of the iteration, the remaining service life is determined.

In FIG. 6 , in the coordinate system including the first measurement point 61 and the second measurement point 62 , the abscissa is the measurement time and the ordinate is the parameter value. The measurement time of the first measurement point 61 is the 5.44th year, and the measured parameter value is 50 microfarads. This first measurement point 61 is provided to the selected aging prediction model and the predicted service life is 21.85 years. The measurement time of the second measurement point 62 is the 18.82nd year, and the measured parameter value is 42 microfarads. This second measurement point 62 is fed into the selected aging prediction model, with a predicted service life of 20.46 years. Similarly, multiple measurement points and corresponding predicted service life can be determined, and based on these measurement points and corresponding predicted service life, the RUL prediction curve 63 can be fitted.

In the coordinate system of the RUL prediction curve 63, the abscissa represents measurement time and the ordinate represents RUL. In the RUL prediction curve 63, the RUL corresponding to the 5.44th year of the measurement time is (21.85-5.44)=16.41 (year); the RUL corresponding to the 18.82nd year of the measurement time is (20.46-18.82)=1.64 (year) . It can be seen that through the RUL prediction curve 63, the RUL corresponding to any measurement time can be predicted.

In the embodiment of the present invention, a solution is proposed to perform RUL prediction for smaller data/sample sizes, especially when the tags of the RUL are not available (no complete data during the entire life cycle of the electronic trip unit) . The electronic trip unit is able to implement the algorithm of the embodiment of the invention because it only contains basic mathematical calculations such as addition, multiplication and comparison, etc. Furthermore, the aging prediction model can be continuously upgraded while the circuit breaker is in operation, meaning that as measurement data continues to increase, so does the model accuracy.

In FIG. 7 , the process of determining the selected aging prediction model 78 is performed on the cloud or edge device side. The process includes: first performing continuous measurements 70 on the circuit breaker according to a time period sequence to obtain a series of parameter values corresponding to the aging indicators of the time period sequence, where these parameter values may be derived from laboratory tests or manufacturers. Then, data preprocessing 71 is performed on the parameter values, such as interpolation, sliding window segmentation, and grouping processing, etc., to enrich and normalize the parameter values. Using the parameter values after data preprocessing 71, multiple training processes 73 are performed on the feedforward neural network to obtain multiple aging prediction models 74. During each training process, the predicted value output by the feedforward neural network can be returned to the input end of the feedforward neural network to perform iterations. Among these parameter values, the parameter values closest to those collected in the current time period are input into N aging prediction models to perform the service life prediction process 75 respectively. Each aging prediction model predicts a service life 76 , so N A total of N service lifespan of 76 are obtained from the aging prediction models. Then, the aging prediction model selection process 77 is performed using these N service lives 76 to select the service life that meets the predetermined conditions and its corresponding aging prediction model 78 .

Then, the process of predicting the RUL of the circuit breaker can be performed on the field device side of the circuit breaker. The process includes: obtaining real-time measurement data (parameter values of measured aging indicators) obtained by performing measurements on the circuit breaker 79, and then determining an aging prediction model 80. The aging prediction model 80 is the aging prediction model 78 selected in the aging prediction model selection process 77 . In the service life prediction process 81, the aging prediction model 80 is used to predict the service life of the circuit breaker. In the RUL prediction process 82 , the service life predicted by the aging prediction model 80 minus the time period of the real-time measurement data 79 is the predicted RUL 83 .

The embodiment of the present invention also proposes an electronic device with a processor-memory architecture for predicting the RUL of a circuit breaker. FIG. 8 is a structural diagram of an electronic device according to an embodiment of the present invention.

As shown in Figure 8, the electronic device 600 includes a processor 601, a memory 602, and a computer program stored in the memory 602 and executable on the processor 601. When the computer program is executed by the processor 601, any of the above-mentioned functions are implemented. RUL method for predicting circuit breakers. Among them, the memory 602 can be implemented as various storage media such as electrically erasable programmable read-only memory (EEPROM), flash memory (Flash memory), programmable programmable read-only memory (PROM), etc. The processor 601 may be implemented to include one or more central processing units or one or more field programmable gate arrays, where the field programmable gate array integrates one or more central processing unit cores. Specifically, the central processing unit or central processing unit core may be implemented as a CPU, an MCU, a DSP, or the like.

It should be noted that not all steps and modules in the above-mentioned processes and structure diagrams are necessary, and some steps or modules can be ignored according to actual needs. The execution order of each step is not fixed and can be adjusted as needed. The division of each module is only for the convenience of describing the functional division. In actual implementation, one module can be implemented by multiple modules, and the functions of multiple modules can also be implemented by the same module. These modules can be located on the same device. , or it can be on a different device.

The hardware modules in various embodiments may be implemented mechanically or electronically. For example, a hardware module may include specially designed permanent circuits or logic devices (such as a dedicated processor such as an FPGA or ASIC) to perform specific operations. Hardware modules may also include programmable logic devices or circuits (eg, including general-purpose processors or other programmable processors) temporarily configured by software to perform specific operations. As for the specific use of mechanical means, or the use of dedicated permanent circuits, or the use of temporarily configured circuits (such as configured by software) to implement the hardware modules, it can be decided based on cost and time considerations.

The above are only preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

A method for predicting the remaining service life of a circuit breaker, which is characterized by including:

Determine (101) the parameter value of the aging indicator of the circuit breaker collected in at least two time periods;

Based on the parameter values, perform (102) N training processes on the neural network model to obtain N aging prediction models, where the aging prediction models are used to predict the aging state of the circuit breaker; wherein in each training process: The parameter values are input into the neural network model as training data in this training process, so as to train the neural network model as the aging prediction model obtained in this training process; where N is a positive value of at least 2. integer;

From the N aging prediction models, select (103) an aging prediction model that meets predetermined conditions;

Based on the selected aging prediction model, the remaining service life of the circuit breaker is predicted (104).
The method according to claim 1, characterized in that:

The parameter values include the first parameter value collected in the first time period and the second parameter value collected in the second time period;

The training of the neural network model to the aging prediction model obtained during this training process includes: inputting the first parameter value to the neural network model, so that the neural network model outputs the prediction parameter value of the second time period; based on The difference between the predicted parameter value in the second time period and the second parameter value determines the loss function value of the neural network model; configure the model parameters of the neural network model so that the loss function value is lower than the predicted value. Set a threshold; determine the configured neural network model as the aging prediction model.
The method according to claim 2, wherein the parameter value also includes the third parameter value collected closest to the current time period, and the predetermined condition includes a service life filtering condition;

Selecting (103) an aging prediction model that meets predetermined conditions from the N aging prediction models includes:

The third parameter value is input into the N aging prediction models respectively, and each aging prediction model determines the service life of the respective output circuit breaker based on the iterative process performed by each, thereby obtaining N service life;

Determine the service life that meets the service life screening conditions from the N service life;

The aging prediction model that outputs the determined service life is determined to be an aging prediction model that meets the predetermined conditions.
The method according to claim 3, wherein each aging prediction model determines the service life of its respective output based on an iterative process executed respectively, including:

The third parameter value is input to the aging prediction model as an input value, so that the aging prediction model outputs the parameter prediction value of the next time period; when the parameter prediction value of the next time period is greater than the preset When the parameter threshold value is reached, the parameter prediction value of the next time period is used as the input value to iterate back to the aging prediction model until the parameter prediction value output by the aging prediction model is less than or equal to the parameter threshold value; The service life of the circuit breaker is determined based on the time period of the parameter prediction value output by the aging prediction model at the end of the iteration.
The method according to claim 3, characterized in that:

The service life filtering conditions include at least one of the following:

The service life is within a predetermined interval;

The service life is: the median value within the interval containing the most service life determined by performing histogram statistics on the N service life.
The method according to any one of claims 1 to 5, characterized in that the parameter value of the aging index of the circuit breaker includes at least one of the following:

For the energy storage capacitor in the power circuit of the electronic trip unit of the circuit breaker, the capacitance value collected according to the time period;

For the on-resistance of the metal-oxide semiconductor field-effect transistor in the power drive circuit of the circuit breaker, the resistance value is collected based on the time period.
The method according to any one of claims 1 to 5, characterized in that, before performing a training process on the neural network model based on the parameter values, it further includes performing preprocessing on the parameter values; wherein the preprocessing Processing includes at least one of the following:

Interpolate the parameter values;

Perform sliding window segmentation on the parameter values;

Group the parameter values.
The method according to any one of claims 1-5, characterized in that, based on the selected aging prediction model, predicting (104) the remaining service life of the circuit breaker includes:

The parameter values collected during the test time period are input into the aging prediction model, so that the aging prediction model outputs the parameter prediction value of the next time period of the test time period; when the parameter prediction of the next time period is When the value is greater than the preset parameter threshold value, the parameter prediction value of the next time period is used as the input value to iterate back to the aging prediction model until the parameter prediction value output by the aging prediction model is less than or equal to the Parameter threshold value; determine the remaining service life based on the difference between the time period of the parameter prediction value at the end of the iteration and the test time period.
An electronic device, characterized by including:

processor(601);

Memory (602), used to store executable instructions of the processor (601);

The processor (601) is configured to read the executable instructions from the memory (602) and execute the executable instructions to implement the predictive circuit breaker of any one of claims 1-8 method of remaining useful life.
A computer-readable storage medium with computer instructions stored thereon, characterized in that when the computer instructions are executed by a processor, the method for predicting the remaining service life of a circuit breaker according to any one of claims 1-8 is implemented. .
A computer program product, characterized by comprising a computer program which, when executed by a processor, implements the method for predicting the remaining service life of a circuit breaker according to any one of claims 1-8.