Equipment health monitoring system and method based on Internet of Things technology

Abstract

The invention discloses an equipment health monitoring method based on the Internet of Things technology, and relates to the technical field of the Internet of Things, and the method comprises the steps: collecting aging degree training data of a ZnO valve plate, and training a convolutional neural network; failure probability training data of the ZnO lightning arrester are collected, and an LSTM neural network is trained; inputting the first aging characteristic data into a convolutional neural network to obtain aging parameters reflecting the aging degree of the ZnO valve plate; inputting third failure probability feature data formed after feature level fusion of the second operation feature data and the aging parameters into the LSTM neural network so as to predict the failure probability of the ZnO lightning arrester at the future moment; determining the failure time of the ZnO lightning arrester according to the failure probability, and determining the residual life of the ZnO lightning arrester; according to the invention, the residual life of the lightning arrester can be monitored in real time.

Description

Technical Field

[0001] The present invention belongs to the technical field of the Internet of Things, and specifically relates to a device health monitoring method based on Internet of Things technology. Background Art

[0002] Zinc oxide (ZnO) lightning arresters are important over-voltage protection devices in the power system, and their health status directly affects the safety and stability of the power transmission and distribution system; due to the influence of various factors such as lightning strikes, electro-thermal coupling effects, and environmental humidity changes during the long-term operation of ZnO lightning arresters, the performance of the internal ZnO varistors will gradually deteriorate, resulting in an increase in leakage current, a decrease in non-linear characteristics, a reduction in heat dissipation capacity, and even damage to the material structure, thereby affecting the overall protection ability of the lightning arrester; if the health status of ZnO lightning arresters cannot be effectively monitored and their failure risks accurately predicted, it may lead to the failure of lightning arresters at critical moments, thereby causing damage to electrical equipment and even power grid failures, resulting in significant economic losses and safety hazards.

[0003] Existing ZnO lightning arrester health monitoring methods mainly rely on regular off-line monitoring or infrared thermal imaging detection, etc. Although the above detection methods can reflect the operating status of ZnO lightning arresters to a certain extent, there are still significant limitations; most existing monitoring means rely on single or limited parameters, making it difficult to comprehensively describe the health status of ZnO lightning arresters, resulting in the monitoring results being greatly affected by environmental and operating conditions fluctuations, with insufficient stability and accuracy; in addition, traditional health assessment methods mainly rely on empirical thresholds or simple statistical analysis, and fail to fully utilize the multi-source operation data of ZnO lightning arresters for in-depth mining, making it difficult to build an effective device health status model, thus limiting the prediction ability and unable to meet the needs of intelligent operation and maintenance. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; for this purpose, the present invention proposes a device health monitoring method based on Internet of Things technology.

[0005] To achieve the above object, the present invention provides a device health monitoring method based on Internet of Things technology, including:

[0006] Step 1: Collect the aging degree training data of ZnO varistors and the failure probability training data of ZnO lightning arresters;

[0007] Step 2: Train a convolutional neural network for predicting the aging parameters of ZnO varistors according to the aging degree training data, and train an LSTM neural network for predicting the failure probability of ZnO lightning arresters according to the failure probability training data;

[0008] Step 3: Obtain the first aging characteristic data of the ZnO varistor, and input the first aging characteristic data into a convolutional neural network to obtain the aging parameters reflecting the aging degree of the ZnO varistor;

[0009] Step 4: Obtain the aging parameters within the current time period and the second operating characteristic data of the ZnO lightning arrester, and input the third failure probability characteristic data formed after feature-level fusion of the second operating characteristic data and the aging parameters into an LSTM neural network to predict the failure probability of the ZnO lightning arrester at a future moment;

[0010] Step 5: Determine the failure time of the ZnO lightning arrester according to the failure probability, and obtain the remaining life of the ZnO lightning arrester based on the failure time.

[0011] Preferably, the aging degree training data of the ZnO varistor includes the first aging characteristic data and its corresponding aging parameters. The first aging characteristic data includes breakdown voltage, dielectric loss, non-linear coefficient, conductivity, and thermal conductivity. The aging parameters include leakage current and material deterioration coefficient. The failure probability training data of the ZnO lightning arrester includes the third failure probability characteristic data and its corresponding failure probability. The third failure probability characteristic data includes the second operating characteristic data and the aging parameters. The second operating characteristic data includes the lightning strike times, operating duration, ambient temperature, and ambient humidity during the operation of the ZnO lightning arrester.

[0012] Preferably, the generation method of the material deterioration coefficient in the aging degree training data is as follows:

[0013] Obtain the flaw detection image of the ZnO varistor;

[0014] After graying the flaw detection image, mark it as the image to be matched;

[0015] Obtain the standard reference images of each damage type, set the window step size to 1, and perform cross-correlation calculation on the image to be matched and the standard reference images through a sliding window algorithm to obtain the similarity distance of each window.

[0016] Take the windows with similarity distances less than the preset similarity distance threshold as the damaged parts, and count the number and area of all damaged parts;

[0017] Substitute the number and area of all damaged parts into the mathematical model for calculating the material deterioration coefficient to obtain the material deterioration coefficient;

[0018] Among them, the expression of the mathematical model for calculating the material deterioration coefficient is as follows:

[0019] ;

[0020] In the formula: is the material deterioration coefficient; is the area of the j-th damaged part in the i-th damage type; is the total area of the ZnO varistor; is the weight factor of the i-th damage type, obtained by fitting experimental data; is the total number of damage types; is the number of all damaged parts in the i-th damage type.

[0021] Preferably, the training method of the convolutional neural network is as follows:

[0022] Divide the aging degree training data into an aging degree training set and an aging degree test set;

[0023] Construct a first regression network based on the convolutional neural network, use the first aging feature data in the aging degree training set as the input of the first regression network, and use the aging parameters as the output of the first regression network, and train the first regression network with the goal of minimizing a predefined first loss function to obtain a first regression network to be verified;

[0024] Among them, the expression of the first loss function is as follows:

[0025] ;

[0026] In the formula: is the first loss function; is the number of samples; is the adaptive weight factor; is the predicted value of the th sample; is the true value of the th sample; is the th trainable parameter of the CNN; is the total number of model parameters of the CNN; is the square of the norm; is the maximum value function;

[0027] Use the aging degree test set to test the first regression network to be verified. When it is less than or equal to the preset test error, output the first regression network to be verified as the convolutional neural network for predicting the aging parameters of the ZnO varistor.

[0028] Preferably, the training method of the LSTM neural network is as follows:

[0029] Divide the failure probability training data into a failure probability training set and a failure probability test set;

[0030] Construct a second regression network based on a four-layer LSTM neural network, use the third failure probability feature data in the failure probability training set as the input of the second regression network, and use the failure probability as the output of the second regression network. Then, train the second regression network with the goal of minimizing a predefined second loss function to obtain a second regression network to be verified.

[0031] Among them, the expression of the second loss function is as follows:

[0032] ;

[0033] In the formula: is the first loss function; is the number of samples; is the failure probability of the real h-th sample; is the failure probability of the h-th sample predicted by LSTM; is the time weighting factor; is the regularization coefficient; is the d-th trainable parameter of LSTM; is the total number of parameters of LSTM; is the time decay coefficient; is the time corresponding to sample h; is the exponential function;

[0034] Use the failure probability test set to test the second regression network to be verified. When it is less than or equal to the preset test error, output the second regression network to be verified as the LSTM neural network for predicting the failure probability of ZnO lightning arresters.

[0035] Preferably, the four-layer LSTM neural network consists of an input feature processing layer, a multi-scale time window LSTM layer, a hybrid attention layer, and an output layer. Among them, the input feature processing layer includes a feature normalization module, a feature embedding module, and a dynamic feature fusion module; the multi-scale time window LSTM layer includes a short-term LSTM sub-network, a medium-term LSTM sub-network, a long-term LSTM sub-network, and a time window fusion module; the hybrid attention layer includes a time attention module, a feature attention module, and a fusion module; the output layer includes a fully connected layer, an activation function layer, and a failure probability prediction module.

[0036] Preferably, the determination of the failure time of the ZnO lightning arrester according to the failure probability includes:

[0037] Set a failure probability threshold;

[0038] Compare the failure probability at a future time with the failure probability threshold. If the failure probability is greater than or equal to the failure probability threshold, determine the corresponding future time as the failure time of the ZnO arrester; if the failure probability is less than the failure probability threshold, do not determine the corresponding future time as the failure time of the ZnO arrester.

[0039] Preferably, obtaining the remaining life of the ZnO arrester based on the failure time includes:

[0040] Extract the failure time of the ZnO arrester;

[0041] Calculate the time distance between the failure time and the current time period;

[0042] Take the time distance as the remaining life of the ZnO arrester.

[0043] An equipment health monitoring system based on the Internet of Things technology is implemented based on the above-mentioned equipment health monitoring method based on the Internet of Things technology, and includes:

[0044] A data collection module for collecting aging degree training data of ZnO varistors and failure probability training data of ZnO arresters;

[0045] A model training module for training a convolutional neural network for predicting the aging parameters of ZnO varistors according to the aging degree training data, and training an LSTM neural network for predicting the failure probability of ZnO arresters according to the failure probability training data;

[0046] A first prediction module for obtaining the first aging characteristic data of the ZnO varistor and inputting the first aging characteristic data into the convolutional neural network to obtain the aging parameters reflecting the aging degree of the ZnO varistor;

[0047] A second prediction module for obtaining the aging parameters within the current time period and the second operating characteristic data of the ZnO arrester, and inputting the third failure probability characteristic data formed after feature-level fusion of the second operating characteristic data and the aging parameters into the LSTM neural network to predict the failure probability of the ZnO arrester at a future time;

[0048] A life determination module for determining the failure time of the ZnO arrester according to the failure probability and obtaining the remaining life of the ZnO arrester based on the failure time.

[0049] An electronic device includes a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements the above-mentioned equipment health monitoring method based on the Internet of Things technology.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] The present invention utilizes a long short-term memory (LSTM) network to establish a time series prediction model, and combines a feature-level data fusion strategy to jointly model the aging data of ZnO varistors and the operating status data of ZnO lightning arresters, enabling the prediction method to have stronger adaptability and generalization ability; through intelligent prediction analysis, the present invention can accurately calculate the failure time of ZnO lightning arresters and dynamically deduce their remaining life based on the failure probability, thereby optimizing the maintenance strategy and avoiding the problems of resource waste or insufficient maintenance caused by the traditional fixed inspection mode; through the Internet of Things architecture, remote real-time monitoring of ZnO lightning arresters is achieved, combined with intelligent prediction analysis, enabling the health status assessment to change from the traditional regular inspection mode to continuous monitoring and dynamic prediction, improving the intelligent level of equipment health management; compared with the prior art, the present invention provides a more accurate, efficient, and intelligent ZnO lightning arrester health monitoring solution, which has significant advantages in improving prediction accuracy, optimizing the maintenance strategy, and reducing operation and maintenance costs, providing important technical support for the reliable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0053] Figure 1 is a schematic flow chart of the method of the present invention;

[0054] Figure 2 is a schematic structural diagram of the system of the present invention;

[0055] Figure 3 is a schematic structural diagram of the electronic device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0057] Please refer to Figure 1 , the first aspect embodiment of the present invention provides a device health monitoring method based on Internet of Things technology, including:

[0058] Step 1: Collect the aging degree training data of ZnO varistors and the failure probability training data of ZnO lightning arresters;

[0059] Specifically, the aging degree training data of the ZnO (zinc oxide) varistor includes first aging characteristic data and its corresponding aging parameters. The first aging characteristic data includes, but is not limited to, breakdown voltage, dielectric loss, non-linear coefficient, conductivity, thermal conductivity, etc. The aging parameters include leakage current and material deterioration coefficient. The failure probability training data of the ZnO (zinc oxide) arrester includes third failure probability characteristic data and its corresponding failure probability. The third failure probability characteristic data includes second operating characteristic data and aging parameters. The second operating characteristic data includes, but is not limited to, lightning strike times, operating duration, ambient temperature and humidity during operation of the ZnO arrester, etc. Among them, the non-linear coefficient is calculated by the volt-ampere characteristic test method, and the specific formula is as follows: ; In the formula: is the non-linear coefficient, and are the currents at low voltage and high voltage respectively, and are low voltage and high voltage respectively;

[0060] Among them, the generation method of the material deterioration coefficient in the aging degree training data is as follows:

[0061] Obtain the flaw detection image of the ZnO varistor;

[0062] After graying the flaw detection image, mark it as the image to be matched;

[0063] Obtain the standard reference images of each damage type, set the window step size to 1, and perform cross-correlation calculation on the image to be matched and the standard reference images through the sliding window algorithm to obtain the similarity distance of each window,

[0064] It should be understood that: the standard reference images of each damage type are pre-stored in the system database, where the damage types include, but are not limited to, cracks and holes, etc.;

[0065] Take the windows with similarity distance less than the preset similarity distance threshold as the damaged parts, and count the number and area of all damaged parts;

[0066] It should be noted that: the basic idea of cross-correlation calculation is to take one image as the template and the other image as the matching object, place the matching object on the template, and calculate the similarity of each overlapping part by sliding the matching object in the template window, that is, filter out the same parts. Among them, the similarity distance is calculated by the Euclidean distance algorithm;

[0067] Substitute the number and area of all damaged parts into the mathematical model for calculating the material deterioration coefficient to obtain the material deterioration coefficient;

[0068] Specifically, the expression of the mathematical model for calculating the material deterioration coefficient is as follows:

[0069] ;

[0070] In the formula: is the material deterioration coefficient; is the area of the j-th damaged part in the i-th type of damage; is the total area of the ZnO varistor; is the weight factor of the i-th type of damage, obtained by fitting experimental data; is the total number of damage types; is the number of all damaged parts in the i-th type of damage;

[0071] It should be understood that the aging degree training data of the ZnO varistor and the failure probability training data of the ZnO lightning arrester are actually collected and recorded by technicians according to experiments or historical situations; among them, each item of data is collected by various sensors, and various sensors include but are not limited to microammeter, voltmeter, lightning strike counter, thermocouple, humidity sensor, industrial camera, flaw detector, LCR tester, etc.

[0072] Step 2: Train a convolutional neural network for predicting the aging parameters of the ZnO varistor based on the aging degree training data, and train an LSTM neural network for predicting the failure probability of the ZnO lightning arrester based on the failure probability training data;

[0073] In implementation, the training method of the convolutional neural network is as follows:

[0074] Divide the aging degree training data into an aging degree training set and an aging degree test set;

[0075] Construct a first regression network with a convolutional neural network as the basic architecture, use the first aging feature data in the aging degree training set as the input of the first regression network, and use the aging parameters as the output of the first regression network, and aim to minimize the predefined first loss function to train the first regression network to obtain the first regression network to be verified;

[0076] Among them, the expression of the first loss function is as follows:

[0077] ;

[0078] In the formula: is the first loss function, used to measure the error between the aging parameters of the ZnO varistor predicted by the CNN and the true value; is the number of samples, that is, the number of data samples participating in the training; is the adaptive weight factor; is the predicted value of the th sample; is the true value of the th sample; is the regularization coefficient, used to control the influence of regularization and prevent overfitting of the CNN; is the th trainable parameter of the CNN; is the total number of model parameters of the CNN; is the square of the norm, that is, the sum of the squares of all parameters; is the maximum function;

[0079] Use the aging degree test set to test the first regression network to be verified. When it is less than or equal to the preset test error, output the first regression network to be verified as the convolutional neural network for predicting the aging parameters of ZnO varistors;

[0080] In implementation, the training method of the LSTM neural network is as follows:

[0081] Divide the failure probability training data into a failure probability training set and a failure probability test set;

[0082] Construct a second regression network based on a four-layer LSTM neural network. Use the third failure probability feature data in the failure probability training set as the input of the second regression network, and the failure probability as the output of the second regression network. Train the second regression network with the goal of minimizing the predefined second loss function to obtain the second regression network to be verified;

[0083] Specifically, the four-layer LSTM neural network consists of an input feature processing layer, a multi-scale time window LSTM layer, a hybrid attention layer, and an output layer. Among them, the input feature processing layer includes a feature normalization module, a feature embedding module, and a dynamic feature fusion module (using an attention mechanism or a weighted method for dynamic fusion); the multi-scale time window LSTM layer includes a short-term LSTM sub-network (ST-LSTM), a medium-term LSTM sub-network (MT-LSTM), a long-term LSTM sub-network (LT-LSTM), and a time window fusion module; the hybrid attention layer includes a time attention module, a feature attention module, and a fusion module; the output layer includes a fully connected layer, an activation function layer, and a failure probability prediction module;

[0084] It can be understood that: The four-layer LSTM structure improves the model's prediction ability for the failure probability of ZnO lightning arresters through hierarchical design; the input feature processing layer ensures data consistency and adaptability, and optimizes the representation of different types of inputs through normalization, feature embedding, and dynamic fusion; the multi-scale time window LSTM layer comprehensively captures short-term fluctuations, periodic changes, and long-term trends through parallel modeling of short-term, medium-term, and long-term sub-networks, and adopts a time window fusion strategy to dynamically adjust the contributions of different time scales, improving the robustness of the prediction; the hybrid attention layer combines temporal attention and feature attention, enabling the model to focus on key time steps and important features, improving information utilization efficiency, and reducing interference from redundant data; the output layer ensures reasonable and stable prediction results through fully connected mapping, non-linear transformation, and a failure probability prediction mechanism;

[0085] Among them, the expression of the second loss function is as follows:

[0086] ;

[0087] In the formula: is the first loss function, which is used to measure the error between the failure probability of the ZnO lightning arrester predicted by the LSTM and the true value; is the number of samples; is the failure probability of the h-th true sample; is the failure probability of the h-th sample predicted by the LSTM; is the time weighting factor; is the regularization coefficient, which is used to control the influence of the regularization term and prevent the LSTM from overfitting; is the d-th trainable parameter of the LSTM; is the total number of parameters of the LSTM; is the time decay coefficient, which is used to control the decay rate of the influence of historical data; is the time corresponding to sample h, that is, the time interval between this data point and the current moment; is the exponential function;

[0088] Use the failure probability test set to test the second regression network to be verified until it is less than or equal to the preset test error, and output the second regression network to be verified as the LSTM neural network for predicting the failure probability of ZnO lightning arresters.

[0089] Step 3: Obtain the first aging characteristic data of the ZnO varistor and input the first aging characteristic data into a convolutional neural network to obtain aging parameters reflecting the aging degree of the ZnO varistor;

[0090] Step 4: Obtain the aging parameters and the second operating characteristic data of the ZnO lightning arrester within the current time period, and input the third failure probability characteristic data formed after the feature-level fusion of the second operating characteristic data and the aging parameters into the LSTM neural network to predict the failure probability of the ZnO lightning arrester at a future time;

[0091] It should be noted that: the time length of the current time period is set artificially in advance by technicians. For example, if the current time period is [T - Q, T], then the time length of the current time period is Q, and in the actual acquisition of the third failure probability characteristic data, the aging parameters and the second operating characteristic data at each moment from T - Q to T are collected;

[0092] It should be understood that: the ways of feature-level fusion include but are not limited to feature splicing, statistical normalization fusion, and linear mapping fusion. For example, the third failure probability feature X = [aging parameter X, third failure probability feature X].

[0093] Step 5: Determine the failure time of the ZnO lightning arrester according to the failure probability, and obtain the remaining life of the ZnO lightning arrester based on the failure time;

[0094] In implementation, determining the failure time of the ZnO lightning arrester according to the failure probability includes:

[0095] Set a failure probability threshold;

[0096] Compare the failure probability at a future time with the failure probability threshold. If the failure probability is greater than or equal to the failure probability threshold, then determine the corresponding future time as the failure time of the ZnO lightning arrester; if the failure probability is less than the failure probability threshold, then do not determine the corresponding future time as the failure time of the ZnO lightning arrester;

[0097] In a specific implementation manner, when the failure probability is less than the failure probability threshold, return to Step 4 and perform the acquisition of the third failure probability characteristic data for the next time period;

[0098] In implementation, obtaining the remaining life of the ZnO lightning arrester based on the failure time includes:

[0099] Extract the failure time of the ZnO lightning arrester;

[0100] Calculate the time distance between the failure time and the current time period;

[0101] Take the time distance as the remaining life of the ZnO lightning arrester;

[0102] Exemplarily, assume that the failure time of the ZnO lightning arrester is 9 o'clock tomorrow, and assume that the current time period [T-Q, T] is from 6 o'clock to 9 o'clock today. Then, the time distance between 9 o'clock tomorrow and 9 o'clock today is calculated to be 24 hours, and 24 hours is taken as the remaining life of the ZnO lightning arrester;

[0103] The present invention realizes the accurate evaluation and intelligent prediction of ZnO lightning arresters by constructing a device health monitoring method based on Internet of Things technology, breaking through the limitations of traditional methods in data utilization, feature extraction, and prediction ability; through the introduction of multi-dimensional data fusion and deep learning models, the present invention can effectively mine the complex correlation between the aging characteristics of ZnO varistors and the operating state of lightning arresters, improve the accuracy of aging trend evaluation, and realize the dynamic prediction of the failure probability of lightning arresters.

[0104] Please refer to Figure 2 , based on the same inventive concept, the second aspect embodiment of the present invention provides a device health monitoring system based on Internet of Things technology. For the details not described in this embodiment, please refer to the relevant parts in Embodiment 1. The system includes:

[0105] A data collection module for collecting aging degree training data of ZnO varistors and failure probability training data of ZnO lightning arresters;

[0106] A model training module for training a convolutional neural network for predicting the aging parameters of ZnO varistors according to the aging degree training data, and training an LSTM neural network for predicting the failure probability of ZnO lightning arresters according to the failure probability training data;

[0107] A first prediction module for obtaining the first aging characteristic data of ZnO varistors and inputting the first aging characteristic data into the convolutional neural network to obtain aging parameters reflecting the aging degree of ZnO varistors;

[0108] A second prediction module for obtaining the aging parameters within the current time period and the second operating characteristic data of the ZnO lightning arrester, and inputting the third failure probability characteristic data formed after feature-level fusion of the second operating characteristic data and the aging parameters into the LSTM neural network to predict the failure probability of the ZnO lightning arrester at a future moment;

[0109] A life determination module for determining the failure time of the ZnO lightning arrester according to the failure probability and obtaining the remaining life of the ZnO lightning arrester based on the failure time.

[0110] Please refer to Figure 3, an embodiment of the third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements the device health monitoring method based on the Internet of Things technology described in any one of the above methods.

[0111] Since the electronic device introduced in the content of this embodiment is the electronic device used to implement a device health monitoring method based on the Internet of Things technology in an embodiment of the present application, based on the device health monitoring method based on the Internet of Things technology introduced in the embodiment of the present application, those skilled in the art can understand the specific implementation manners and various forms of change of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiment of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in a device health monitoring method based on the Internet of Things technology in an embodiment of the present application, it falls within the scope of protection of the present application.

[0112] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer, or a data storage device such as a server or data center that includes one or more collections of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0113] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only one way, and in actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the devices or units can be in electrical, mechanical, or other forms.

[0114] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0115] In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0116] Some of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is obtained by software simulation of a large amount of collected data to get a formula closest to the actual situation; the preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0117] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A device health monitoring method based on Internet of Things technology, characterized in that, Including: Step 1: Collect the aging degree training data of ZnO varistors and the failure probability training data of ZnO lightning arresters; Step 2: Train a convolutional neural network for predicting the aging parameters of ZnO varistors based on the aging degree training data, and train an LSTM neural network for predicting the failure probability of ZnO lightning arresters based on the failure probability training data; Step 3: Obtain the first aging characteristic data of the ZnO varistor and input the first aging characteristic data into the convolutional neural network to obtain the aging parameters reflecting the aging degree of the ZnO varistor; Step 4: Obtain the aging parameters within the current time period and the second operating characteristic data of the ZnO lightning arrester, and input the third failure probability characteristic data formed after feature-level fusion of the second operating characteristic data and the aging parameters into the LSTM neural network to predict the failure probability of the ZnO lightning arrester at a future moment; Step 5: Determine the failure time of the ZnO lightning arrester according to the failure probability and obtain the remaining life of the ZnO lightning arrester based on the failure time.

2. The device health monitoring method based on Internet of Things technology according to claim 1, wherein The aging degree training data of the ZnO varistor includes the first aging characteristic data and its corresponding aging parameters. The first aging characteristic data includes breakdown voltage, dielectric loss, nonlinear coefficient, conductivity, and thermal conductivity. The aging parameters include leakage current and material deterioration coefficient. The failure probability training data of the ZnO lightning arrester includes the third failure probability characteristic data and its corresponding failure probability. The third failure probability characteristic data includes the second operating characteristic data and the aging parameters. The second operating characteristic data includes the lightning strike times, operating duration, ambient temperature, and ambient humidity of the ZnO lightning arrester.

3. The device health monitoring method based on Internet of Things technology according to claim 2, characterized in that The generation method of the material deterioration coefficient in the aging degree training data is as follows: Obtain the flaw detection image of the ZnO varistor; After graying the flaw detection image, mark it as the image to be matched; Obtain the standard reference images of each damage type, set the window step size to 1, and perform cross-correlation calculation on the image to be matched and the standard reference images through the sliding window algorithm to obtain the similarity distance of each window; Take the windows with similarity distances less than the preset similarity distance threshold as the damaged parts, and count the number and area of all damaged parts; Substitute the number and area of all damaged parts into the mathematical model for calculating the material deterioration coefficient to obtain the material deterioration coefficient; Among them, the expression of the mathematical model for calculating the material deterioration coefficient is as follows: ; In the formula: is the material deterioration coefficient; is the area of the j-th damaged part in the i-th damage type; is the total area of the ZnO varistor disc; is the weighting factor of the i-th damage type, obtained by fitting experimental data; is the total number of damage types; is the number of all damaged parts in the i-th damage type.

4. The device health monitoring method based on Internet of Things technology according to claim 3, characterized in that, The training method of the convolutional neural network is as follows: Divide the aging degree training data into an aging degree training set and an aging degree test set; Construct a first regression network based on the convolutional neural network, use the first aging characteristic data in the aging degree training set as the input quantity of the first regression network, and use the aging parameters as the output quantity of the first regression network, and train the first regression network with the goal of minimizing the predefined first loss function to obtain the first regression network to be verified; Among them, the expression of the first loss function is as follows: ; Wherein: is the first loss function; is the number of samples; is the adaptive weight factor; is the predicted value of the th sample; is the true value of the th sample; is the regularization coefficient; is the th trainable parameter of the CNN; is the total number of model parameters of the CNN; is the square of the norm; is the maximum value function; Use the aging degree test set to test the first regression network to be verified until it is less than or equal to the preset test error, and then output the first regression network to be verified as a convolutional neural network for predicting the aging parameters of ZnO varistors.

5. The device health monitoring method based on Internet of Things technology according to claim 4, characterized in that The training method of the LSTM neural network is as follows: Divide the failure probability training data into a failure probability training set and a failure probability test set; Construct a second regression network based on a four-layer LSTM neural network. Use the third failure probability feature data in the failure probability training set as the input of the second regression network, and the failure probability as the output of the second regression network. Train the second regression network with the goal of minimizing the predefined second loss function to obtain the second regression network to be verified; Among them, the expression of the second loss function is as follows: ; In the formula: is the first loss function; is the number of samples; is the failure probability of the h-th true sample; is the failure probability of the h-th sample predicted by LSTM; is the time weighting factor; is the regularization coefficient; is the d-th trainable parameter of LSTM; is the total number of parameters of LSTM; is the time decay coefficient; is the time corresponding to sample h; is the exponential function; Use the failure probability test set to test the second regression network to be verified until it is less than or equal to the preset test error, and then output the second regression network to be verified as an LSTM neural network for predicting the failure probability of ZnO arresters.

6. The device health monitoring method based on Internet of Things technology according to claim 5, characterized in that, The four-layer LSTM neural network consists of an input feature processing layer, a multi-scale time window LSTM layer, a hybrid attention layer, and an output layer. Among them, the input feature processing layer includes a feature normalization module, a feature embedding module, and a dynamic feature fusion module; the multi-scale time window LSTM layer includes a short-term LSTM sub-network, a medium-term LSTM sub-network, a long-term LSTM sub-network, and a time window fusion module; the hybrid attention layer includes a time attention module, a feature attention module, and a fusion module; the output layer includes a fully connected layer, an activation function layer, and a failure probability prediction module.

7. The device health monitoring method based on Internet of Things technology according to claim 6, characterized in that, The determination of the failure time of the ZnO arrester according to the failure probability includes: Set a failure probability threshold; Compare the failure probability at a future time with the failure probability threshold. If the failure probability is greater than or equal to the failure probability threshold, determine the corresponding future time as the failure time of the ZnO arrester; if the failure probability is less than the failure probability threshold, do not determine the corresponding future time as the failure time of the ZnO arrester.

8. The device health monitoring method based on Internet of Things technology according to claim 7, characterized in that, The obtaining of the remaining life of the ZnO arrester based on the failure time includes: Extract the failure time of the ZnO arrester; Calculate the time distance between the failure time and the current time period; Take the time distance as the remaining life of the ZnO arrester.

9. An equipment health monitoring system based on Internet of Things technology, which is implemented based on the equipment health monitoring method based on Internet of Things technology described in any one of claims 1-8, and is characterized in that, Includes: A data collection module for collecting the aging degree training data of ZnO varistors and the failure probability training data of ZnO arresters; A model training module for training a convolutional neural network for predicting the aging parameters of ZnO varistors according to the aging degree training data, and training an LSTM neural network for predicting the failure probability of ZnO arresters according to the failure probability training data; A first prediction module for obtaining the first aging feature data of the ZnO varistor and inputting the first aging feature data into the convolutional neural network to obtain the aging parameters reflecting the aging degree of the ZnO varistor; The second prediction module is used to obtain the aging parameters and the second operating characteristic data of the ZnO lightning arrester within the current time period, and input the third failure probability characteristic data formed after feature-level fusion of the second operating characteristic data and the aging parameters into the LSTM neural network to predict the failure probability of the ZnO lightning arrester at a future moment; The life determination module is used to determine the failure time of the ZnO lightning arrester according to the failure probability and obtain the remaining life of the ZnO lightning arrester based on the failure time.

10. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the device health monitoring method based on the Internet of Things technology according to any one of claims 1 to 8.

Images