Method, device, medium and equipment for predicting burnout failure of metering electric energy meter

Abstract

The application discloses a kind of metering electric energy meter burnout fault prediction method, device, medium and equipment, it is related to electric energy meter fault diagnosis technical field.The method comprises: obtaining the material state characterization data of the structural component sample of electric energy meter under different thermal stress parameters and the actual fault grade of electric energy meter;Thermal stress parameters, material state characterization data are input into neural network model to generate predicted fault grade;With the difference between predicted fault grade and actual fault grade as the target to train neural network model, obtain burnout prediction model;The target thermal stress parameter of the structural component of the electric energy meter to be measured is input into the corresponding burnout prediction model, and the burnout fault grade of the electric energy meter to be measured is obtained.The method of the application establishes the mapping relationship between thermal stress parameter and material thermal degradation state and burnout fault grade, can reduce the acquisition demand of material state characterization data in actual detection, realizes the prediction and grading protection of electric energy meter structural component burnout fault.

Description

Technical Field

[0001] This application relates to the field of electricity meter fault diagnosis technology, and in particular to a method, device, medium and equipment for predicting burn-out faults of electricity meters. Background Technology

[0002] Currently, in the field of low-voltage electricity meter operation monitoring, burn-out fault prediction largely relies on electrical parameters such as voltage, current, and power measured by the meter itself, combined with machine learning algorithms to build predictive models. However, electrical parameters are easily affected by external environmental factors such as power grid fluctuations and electromagnetic interference, resulting in insufficient stability of the data source. More importantly, electricity meter burn-out is a gradual process involving changes in the physicochemical properties of materials, directly manifested in the internal physical structural components of the meter. It is difficult to establish a direct and reliable logical relationship between electrical parameters and the state parameters of component materials, which limits the predictive ability of the model to the coverage of training data such as voltage and current. This often results in prediction results with a significant lag, typically only responding when there are drastic changes in electrical performance, failing to achieve true early prevention and accurate warning. Summary of the Invention

[0003] In view of this, this application provides a method, device, medium and equipment for predicting burn-out faults of electricity meters, which realizes burn-out fault prediction based on the physical and chemical nature of materials, and improves the physical interpretability and generalization ability of the prediction results.

[0004] According to a first aspect of this application, a method for predicting burnout faults in an electricity meter is provided, the method comprising: Obtain material state characterization data of structural component samples of the electricity meter under different thermal stress parameters and the actual fault level of the electricity meter; The thermal stress parameters and the material state characterization data are used as training samples and input into the neural network model. Features are extracted in parallel through the neural network model, and the correlation mapping features of the thermal stress parameters are constrained by the material state characterization data in the hidden layer feature space to establish the correlation mapping relationship between the material state characterization data and the thermal stress parameters. Based on the fusion features of the thermal stress parameters and the constrained correlation mapping features, the material quality parameters of the structural component sample are determined, and the predicted fault level is generated. With the goal of minimizing the difference between the predicted fault level and the actual fault level, the neural network model is iteratively trained to obtain the burn-off prediction model corresponding to the structural component sample, wherein the model parameters of the burn-off prediction model are fixed with the correlation mapping relationship; The target thermal stress parameter of the structure component of the to-be-tested electric energy meter is input to the corresponding burnout prediction model, the burnout prediction model reconstructs the corresponding target material state feature in the hidden layer feature space from the target thermal stress parameter based on the association mapping relationship, and determines the material quality parameter of the to-be-tested electric energy meter and outputs a burnout failure grade based on the fusion feature of the feature of the target thermal stress parameter and the target material state feature.

[0005] According to a second aspect of the present application, a burnout failure prediction device of a metering electric energy meter is provided, and the device comprises: An acquisition module is configured to acquire material state representation data of a structure component sample of an electric energy meter under different thermal stress parameters and an actual failure grade of the electric energy meter. A training module is configured to input the thermal stress parameters and the material state representation data as training samples into a neural network model, extract features in parallel through the neural network model, constrain the association mapping feature of the thermal stress parameters in a hidden layer feature space with the material state representation data, establish an association mapping relationship between the material state representation data and the thermal stress parameters, determine a material quality parameter of the structure component sample based on the fusion feature of the feature of the thermal stress parameters and the constrained association mapping feature, and generate a predicted failure grade. The neural network model is iteratively trained to minimize the difference between the predicted failure grade and the actual failure grade, and a burnout prediction model corresponding to the structure component sample is obtained, wherein the model parameters of the burnout prediction model are solidified with the association mapping relationship. A prediction module is configured to input a target thermal stress parameter of a structure component of a to-be-tested electric energy meter to the corresponding burnout prediction model, the burnout prediction model reconstructs the corresponding target material state feature in the hidden layer feature space from the target thermal stress parameter based on the association mapping relationship, and determines the material quality parameter of the to-be-tested electric energy meter and outputs a burnout failure grade based on the fusion feature of the feature of the target thermal stress parameter and the target material state feature.

[0006] According to a third aspect of the present application, a readable storage medium is provided, which stores a program or instructions, and the program or instructions are executed by a processor to implement the steps of the burnout failure prediction method of the metering electric energy meter.

[0007] According to a fourth aspect of the present application, a computer device is provided, which comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the steps of the burnout failure prediction method of the metering electric energy meter when executing the program.

[0008] By the technical scheme, the thermal stress parameter and the material state characterization data are simultaneously taken as heterogeneous feature input neural network models, external environmental stress factors and internal degradation factors of the material are extracted, cross-mapped and multi-modal spliced in the feature space by the model to capture the burnout evolution law under the coupling action of the two, and the fusion features are mapped to material quality parameters that can reflect the flame-retardant residual capacity of the structural components, and then the material quality parameters are used to generate a predicted failure level. The burnout prediction model capable of deducing the material degradation state and the burnout level from the thermal stress parameter and / or the material state characterization data is obtained by iterative training with the minimization of the difference between the predicted failure level and the actual failure level as the target. In actual application, in the face of the technical problem that the internal material state of the electric meter is difficult to obtain in real time and non-destructively, only the target thermal stress parameter of the structural component of the electric meter to be measured needs to be input into the corresponding burnout prediction model, and the model can accurately reconstruct the corresponding target material state feature in the hidden layer space by using the learned correlation mapping relationship, so as to determine the material quality parameter reflecting the flame-retardant residual capacity of the structural component by fusing the original feature of the target thermal stress parameter and the reconstructed target material state feature, and accurately output the burnout failure level of the electric meter to be measured. Thus, by comprehensively modeling the environmental factors and the material degradation state, high-precision and high-robustness evaluation of the progressive burnout evolution state of the internal structural components of the electric meter under strong electromagnetic interference and dynamic load working conditions is realized, so as to early and actively warn the electric meter, effectively avoid functional failure and safety accidents caused by structural burnout, and significantly improve the operation reliability and intelligent maintenance level of the electric meter.

[0009] The above description is only a summary of the technical scheme of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented in accordance with the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0010] The drawings described herein are used to provide further understanding of the present application, and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application, and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of a burnout failure prediction method of a metering electric meter provided by an embodiment of the present application is shown; Figure 2 A red spectrum diagram of a structural component sample after a burnout test is shown; Figure 3 A scanning electron microscope and an energy spectrum diagram of a structural component sample after a 250°C burnout test are shown; Figure 4 A scanning electron microscope and an energy spectrum diagram of a structural component sample after a 450°C burnout test are shown; Figure 5 A schematic diagram of a thermogravimetric curve of a structural component sample after a burn-off test is shown; Figure 6 A comparison line graph of predictions of models trained with different input parameters is shown; Figure 7 A comparison diagram of model predictions and failure levels obtained from burn-off experiments under different thermal stress parameter conditions is shown; Figure 8 A structural block diagram of a burn-off failure prediction device for a metering electric energy meter provided by an embodiment of the present application is shown; Figure 9 A structural diagram of a computer device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0011] Hereinafter, the present application will be described in detail with reference to the accompanying drawings and in conjunction with embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0012] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, in which the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0013] Those skilled in the art can understand that, unless specifically stated, the singular forms "a", "an" and "the" used herein also include the plural forms. It should be further understood that the use of the term "include" in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "joined" to another element, it can be directly connected or joined to the other element, or there can be intermediate elements. In addition, the phrase "and / or" used herein includes all or any single unit and all combinations of the associated listed items.

[0014] Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in various different forms, and should not be interpreted as being limited only to the embodiments set forth herein. It should be understood that these embodiments are provided so that the disclosure of the present application is complete and complete, and the concepts of these exemplary embodiments are sufficiently conveyed to those skilled in the art.

[0015] In the present embodiment, a burn-off failure prediction method for a metering electric energy meter is provided, as shown inFigure 1 The method comprises: In step 101, material state characterization data of a structure component sample of the electric energy meter under different thermal stress parameters and an actual failure level of the electric energy meter are obtained.

[0016] The structure component refers to a key physical carrier in the electric energy meter that directly bears thermal stress and is prone to burning loss, such as a terminal component, a terminal cover component, or an insulating base component. The structure component is usually made of a flame-retardant high polymer material. The flame-retardant high polymer material can be added with a reinforcing phase and a bromine-antimony flame retardant to enhance the flame-retardant performance and support strength of the material, such as a flame-retardant polybutylene terephthalate, which is mixed with a flame retardant as an added component in a PBT (Polybutylene Terephthalate) matrix material.

[0017] It should be understood that the structure component sample is a sample structure component used to collect model training data.

[0018] The thermal stress parameter is used to characterize the external thermal stress condition to characterize the burning loss inducement from the external stress dimension. The thermal stress parameter includes but is not limited to a heating temperature and a continuous heating time. The heating temperature can be collected in real time by a temperature sensor arranged on or near the surface of the terminal component. The continuous heating time can be accumulated and counted by a timer for the duration of the structure component being above a preset temperature threshold.

[0019] The material state characterization data can be obtained by physical burning loss experiments and physicochemical characterization tests on the material of the structure component sample under different temperatures and times, covering the whole life cycle information of the thermal degradation of the high polymer material from the dimensions of molecular chemical bond rupture, macroscopic mass loss kinetics, or flame-retardant element consumption. For example, the infrared spectrum ratio feature of the ratio between the peak intensity or the peak area of the characteristic peaks of different chemical bonds and functional groups is obtained by Fourier transform infrared spectrum testing, the thermogravimetric and heat flow features of the mass change and heat flow change of the structure component sample with temperature are obtained by thermogravimetric analysis and differential scanning calorimetry testing, and the element energy spectrum data that can determine the element content and / or element content change feature of the polymer matrix are obtained by energy dispersive X-ray spectrum testing. It can be understood that the material state characterization data includes but is not limited to at least one of the infrared spectrum ratio feature, the thermogravimetric and heat flow feature, and the element energy spectrum data.

[0020] For example, the infrared spectral ratio feature is obtained by Fourier transform infrared spectroscopy (FTIR) test on the material of the structural component, which can reflect the relative content change of ester carbonyl, ester bond, fatty chain, hydroxyl and other functional groups in the material with temperature evolution. The thermogravimetric and heat flow characteristics are obtained by thermogravimetric analysis (TG / DTG) and differential scanning calorimetry (DSC) tests. The TG curve reflects the change trend of material quality with temperature, and the DSC curve reflects the endothermic and exothermic behavior of the material during heating. The energy spectrum data is obtained by energy dispersive spectroscopy (EDS) test, which can quantify the relative content change of carbon, oxygen, bromine, antimony and other elements in the material, thereby reflecting the degree of volatile loss of the flame retardant during thermal reaction.

[0021] In step 102, the thermal stress parameters and the material state characterization data are input into the neural network model as training samples, the features are extracted in parallel through the neural network model, and the correlation mapping features of the thermal stress parameters are constrained in the hidden layer feature space with the material state characterization data, so as to establish the correlation mapping relationship between the material state characterization data and the thermal stress parameters. The material quality parameters of the structural component sample are determined based on the fusion features of the features of the thermal stress parameters and the constrained correlation mapping features, and the predicted failure level is generated.

[0022] Among them, the material quality parameter is used to reflect the result of thermal degradation or thermal damage of the structural component sample under the action of thermal stress. For example, the material quality parameter can be the mass content change rate, the change rate of molecular structure, the volatile rate of flame retardant, etc.

[0023] It can be understood that the correlation mapping features corresponding to the thermal stress parameters are obtained by neural network model feature transformation, which are used to represent the features of the thermal stress parameters corresponding to the thermal action state. The hidden layer feature space refers to the high-dimensional mathematical representation space inside the neural network model for mapping and aligning heterogeneous features. In this space, different modal input data are converted into unified dimensional feature vectors, thereby providing measurable mathematical basis and algorithm boundary for applying distribution alignment constraint, calculating feature difference and establishing cross-modal topological mapping relationship.

[0024] In this embodiment, the neural network model adopts a multi-modal deep learning architecture, which can simultaneously extract multi-dimensional features of material state characterization data and thermal stress parameters, to extract features from both environmental triggers and material degradation performance dimensions. Thus, the nonlinear coupling between environmental stress and material performance degradation can be accurately captured, and the distribution of the associated mapping features guided by the material state features as a physical supervision scale can be aligned with the real material degradation evolution law. At the algorithm level, the virtual deduction from easily obtainable environmental stress to difficult to measure material intrinsic response is completed. Then, the associated mapping features constrained by the above distribution alignment are spliced in the channel dimension, and the high-dimensional fused features after splicing are reduced and calculated by nonlinear mapping to obtain the material quality parameters and the corresponding predicted failure level of the structural component. In this way, the model can internalize the physical prior mechanism of material microstate evolution in the form of implicit reconstruction of network weights, so that the associated mapping relationship learned by the model is fixed in the model parameters of the burnout prediction model. In the actual prediction process, the material state characterization data of the to-be-tested structural component does not need to be obtained again, and the target material state feature corresponding to the target thermal stress parameter of the to-be-tested structural component can be determined in the hidden layer feature space. Thus, the modeling capability for complex burnout evolution process is improved, the evaluation of micro cumulative damage in the material is more accurate and robust, and fine quantitative perception can be realized in the early stage of burnout. Reliable basis is provided for preventive maintenance and active safety control of internal structural components of the electric energy meter.

[0025] It should be noted that the material quality parameter is positively correlated with the material state characterization data such as TG, DSC, FTIR, and EDS, and is negatively correlated with temperature and duration. For example, the fewer the functional groups in FTIR, the lower the element content in EDS, and the lower the values of TG and DSC, all of which will promote the decrease of the predicted quality retention rate; and the higher the temperature and the longer the time, the material will also gradually degrade, resulting in a decrease in the quality retention rate.

[0026] In actual application scenarios, the neural network model includes multiple parallel multilayer perceptron (MLP) subnetworks, multiple convolutional neural subnetworks (CNN), and feature splicing layers, associated mapping layers, and fully connected regression networks. Step 102 specifically includes the following steps: Step 102-1, calling a multilayer perceptron subnetwork, parallel low-dimensional vector mapping of discrete thermal stress parameters and material state characterization data is performed to obtain low-dimensional mapping features.

[0027] The discrete thermal stress parameters and the material state characterization data are physical quantities without natural time sequence or spatial topology. For example, the thermal stress parameters include the heating temperature, the heating time, and the like; and the material state characterization data includes the relative infrared spectrum ratio, the element mass fraction, and the like.

[0028] In this embodiment, the discrete physical quantities are mapped to low-dimensional vectors by using the multi-layer perception sub-networks with multiple fully connected layers and nonlinear activation functions, and the discrete features of different dimensions are uniformly mapped to a low-dimensional hidden space to obtain low-dimensional mapping features. Thus, a nonlinear reference correlation between the environmental stress intensity and the material component state is established, which not only prevents the introduction of redundant computational overhead, but also avoids feature distortion caused by false local correlation assumptions, thereby providing stable global context information for subsequent feature fusion.

[0029] For example, multiple parallel multi-layer perception sub-networks can process different types of discrete features, such as a first MLP sub-network processing the operating environment parameters (temperature, time), a second MLP sub-network processing the infrared spectrum ratio, and a third MLP sub-network processing the element mass fraction. The low-dimensional mapping features output by the sub-networks are finally aggregated into unified low-dimensional mapping features.

[0030] In step 102-2, multiple convolutional neural sub-networks are called to perform local time sequence feature extraction on the continuous thermal stress parameters and the material state characterization data in parallel to obtain local time sequence features.

[0031] The continuous thermal stress parameters and the material state characterization data are spectral physical quantities. For example, the thermal gravimetric curve (TG), the differential scanning calorimetry curve (DSC), and the infrared spectrum curve are continuous signals distributed along the temperature or wave number axis, and the local waveform features such as the peak position, the peak width, and the baseline drift contained therein are key fingerprint information reflecting the material phase transition, the chemical bond rupture, and the degradation rate.

[0032] In this embodiment, the convolutional neural sub-networks perform sliding window convolution operation on the continuous curve data by using one-dimensional convolution layers to extract the local change mode and gradient features of the curve in different temperature intervals, thereby obtaining local time sequence features. Thus, the microscopic patterns with translation invariance can be automatically captured from the continuous spectra, and the time sequence features representing the material thermal degradation dynamics process can be effectively extracted.

[0033] Exemplarily, the convolutional neural subnetwork can include multiple one-dimensional convolutional layers, pooling layers, and activation function layers, and through layer-by-layer convolution and down-sampling, the original curve data is abstracted into high-order time sequence feature representation. The TG curve appears a cliff-like drop in the interval of 350-450 DEG C, and the convolutional features in this interval can capture the degree of change in mass and the peak value of the rate; the DSC curve appears an endothermic peak in the interval of 220-230 DEG C, and the convolutional features in this interval can reflect the phase change behavior of the material.

[0034] In step 102-3, the correlation mapping layer is called to map the low-dimensional mapping features corresponding to the thermal stress parameters and the local time sequence features to the hidden layer feature space to form the correlation mapping features. In the hidden layer feature space, the correlation mapping features are supervised and constrained by taking the low-dimensional mapping features and the local time sequence features corresponding to the material state representation data as the supervision reference, so that the correlation mapping features and the corresponding features of the material state representation data have consistent feature distribution in the hidden layer feature space, and the correlation mapping relationship is formed.

[0035] In this embodiment, the hidden layer feature transformation is performed on the features corresponding to the thermal stress parameters by the correlation mapping layer, and the correlation mapping features obtained by the feature transformation are constrained by using the supervision information provided by the corresponding features of the material state representation data, so that the features extracted from the thermal stress parameters can form a feature expression corresponding to the material state change in the hidden layer feature space. Thus, a stable feature correspondence is established between the features corresponding to the thermal stress parameters and the features corresponding to the material state representation data. During the training process, with the iterative update of the model parameters, the correlation mapping relationship is gradually solidified in the model parameters of the correlation mapping layer, so that the model deep learning chain from the external environment thermal excitation signal to the internal high polymer material structure degradation physical causal evolution is provided, and a feature basis is provided for subsequent deduction of the material state according to the thermal stress parameters in the absence of material state representation data.

[0036] In step 102-4, the feature splicing layer is called to concatenate and splice the low-dimensional mapping features and the local time sequence features corresponding to the thermal stress parameters, and the supervised and constrained correlation mapping features to generate the fusion features.

[0037] The supervised and constrained correlation mapping features are the features that have been aligned to the distribution of the material state representation data under the supervision signal.

[0038] In step 102-5, the full connection regression network is called to map the fusion features to the material quality parameters, and the predicted failure level is output based on the matching result of the material quality parameters and the preset parameter interval.

[0039] In this embodiment, the thermal stress parameter corresponding basic feature is combined with the material state association feature obtained through association mapping to jointly characterize the fusion feature, so that the fusion feature contains the thermal stress action state and the material state information corresponding to the thermal stress action. The full connection regression network is used to perform nonlinear mapping on the fusion feature, and the association information between the thermal stress action and the material degradation state is converted into a material quality parameter capable of representing the thermal degradation or thermal damage degree of the structure component. Then, the corresponding failure level is determined through the matching relationship between the material quality parameter and the preset parameter interval. Thus, the judgment of the burnout state is avoided according to a single electrical parameter, and the representation ability of the progressive material degradation process of the structure component is improved.

[0040] It should be noted that the corresponding relationship between the preset parameter interval and the burnout failure level is not fixed, but can be adaptively adjusted according to the types of materials used by different models of electric energy meters, structure design and safety margin requirements. The specific values of the preset parameter interval in the embodiments of the present application are not limited. For example, for modified materials with higher heat resistance level, the interval threshold values can be appropriately lowered to prolong the service period; and for old electric meters or high-risk environments, the interval threshold values can be increased to improve the safety sensitivity.

[0041] For example, when the material quality parameter is greater than or equal to the first preset threshold value, it is determined that there is no abnormality, that is, no burnout failure occurs.

[0042] When the material quality parameter is less than the first preset threshold value and greater than or equal to the second preset threshold value, that is, the material quality parameter is located in the first preset parameter interval, it can be determined that the electric energy meter is in the first type of processing level. For example, for typical flame-retardant PBT terminal, when the quality retention rate is between 90% and 98%, the material inside has begun to relax the macromolecular chain or preliminarily precipitate the flame retardant, but has not entered the period of violent chemical bond rupture, and the macroscopic performance is that the mechanical strength decreases but the insulation function is not completely lost, corresponding to the burnout hidden danger state.

[0043] When the material quality parameter is less than the second preset threshold value and greater than or equal to the third preset threshold value, that is, the material quality parameter is located in the second preset parameter interval, it can be determined that the electric energy meter is in the second type of processing level. For example, when the quality retention rate decreases to between 70% and 90%, the ester bond and other key functional groups in the polymer matrix are broken on a large scale, accompanied by the generation of a large number of volatile small molecules and rapid quality loss, and the physicochemical properties of the material have undergone irreversible substantial degradation, corresponding to the burnout failure state.

[0044] When the material quality parameter is less than the third preset threshold, that is, the material quality parameter is located in the third preset parameter interval, it can be determined that the electric energy meter is in the third type of processing level. For example, when the quality retention rate is lower than 40% or even lower, the base resin has been basically decomposed, and the remaining substances are mainly inorganic fillers and carbon residues. The material completely loses the insulation support and flame-retardant oxygen isolation ability, and is extremely easy to cause secondary disasters such as fire or arc short circuit, corresponding to a serious damage state or a serious fault state.

[0045] Step 103, iteratively training the neural network model to obtain a burnout prediction model corresponding to the structural component sample, with the objective of minimizing the difference between the predicted failure level and the actual failure level.

[0046] In the embodiment, the model parameters of the burnout prediction model have learned the correlation mapping relationship between the material state representation data and the thermal stress parameters in the training process.

[0047] In the embodiment, the model parameters of the burnout prediction model have learned the correlation mapping relationship between the material state representation data and the thermal stress parameters in the training process.

[0048] Step 104, inputting the target thermal stress parameter of the structural component of the to-be-tested electric energy meter into the corresponding burnout prediction model. The burnout prediction model reconstructs the corresponding target material state feature from the target thermal stress parameter in the hidden layer feature space based on the correlation mapping relationship, and determines the material quality parameter of the to-be-tested electric energy meter and outputs the burnout failure level based on the fusion feature of the target thermal stress parameter feature and the target material state feature.

[0049] The burnout fault prediction method of the meter provided by the embodiment of the present application takes the thermal stress parameter and the material state characterization data as heterogeneous feature inputs of the neural network model, extracts, cross-maps and multi-modal splices the external environmental stress factors and the intrinsic degradation factors of the material in the feature space in parallel through the model to capture the burnout evolution law under the coupling action of the two, maps the fused deep features to the material quality parameters that can reflect the flame-retardant residual capacity of the structural components, and generates a prediction fault level based on the material quality parameters. The burnout prediction model capable of deducing the material degradation state and the burnout level from the thermal stress parameter is obtained by iterative training with the minimization of the difference between the predicted fault level and the actual fault level as the target. In actual application, in the face of the technical problem that the internal material state of the meter is difficult to obtain in real time and non-destructively, only the target thermal stress parameter of the structural component of the meter to be tested needs to be input into the corresponding burnout prediction model, and the model can accurately reconstruct the corresponding target material state feature in the hidden layer space by using the learned correlation mapping relationship, so as to determine the material quality parameter that can reflect the flame-retardant residual capacity of the structural component by fusing the original features of the target thermal stress parameter and the reconstructed target material state feature, and accurately output the burnout fault level of the meter to be tested. Thus, by comprehensively modeling the environmental factors and the material degradation state, high-precision and high-robustness evaluation of the progressive burnout evolution state of the internal structural components of the meter under strong electromagnetic interference and dynamic load working conditions is realized, so as to early and actively warn the meter, effectively avoid functional failure and safety accidents caused by structural burnout, and significantly improve the operation reliability and intelligent maintenance level of the meter.

[0050] It can be understood that, since the burnout prediction model has solidified the correlation mapping relationship between the material state characterization data and the thermal stress parameter, when predicting the burnout fault level through the burnout prediction model, the input parameter can be the thermal stress parameter, the material state characterization data, or a combination of the two, that is, both can effectively predict the burnout fault level of the meter. However, since the material state characterization data needs to be obtained by disassembling the structural components of the meter and performing corresponding test experiments, the data acquisition process is complicated, the detection period is long, and it is a destructive detection method, which is difficult to meet the real-time requirement of the meter burnout prediction. Therefore, in the present application, the thermal stress parameter is preferred as the input parameter for fault level prediction, so as to realize rapid and real-time evaluation of the burnout fault level of the meter under non-disassembly conditions.

[0051] Further, to ensure the practicability of the burnout prediction model, model A, model B and model C are trained respectively with different modal inputs and fault level labels in Table 1 as training data; wherein, model B and model C are the burnout prediction models trained by the method of the embodiment of the present application. Figure 6The comparison diagram of the prediction results of model A trained with only the single feature group of "temperature + time", model B trained with the three feature groups of "temperature + time", "TG" and "DSC", and model C trained with "temperature + time", "TG", "DSC", "FTIR" and "EDS" under the condition of holding for 60 min is shown. Figure 6 It can be observed that the prediction result of model A is unstable, and there is an abnormal mutation of the prediction value at both high and low temperatures, the fault early warning is lagging, which indicates that the prediction result is poor and basically cannot be put into use. The prediction results of multi-modal models B and C are close, the consistency of the determination results is high, and the broken line is almost free of mutation, the fault can be detected in the low temperature section, and it can be put into use. Although model B has some prediction value deviation in the high temperature interval, the overall still meets the prediction accuracy requirement.

[0052] In order to study the prediction effect of thermal stress parameters on the material burnout behavior and verify the generalization ability and actual engineering applicability of the constructed burnout prediction model, five groups of input data were randomly selected: "230℃-40min", "275℃-20min", "330℃-70min", "390℃-80min" and "460℃-50min" for result verification. The five groups of data were input into the burnout prediction model after training, and the muffle furnace simulation burnout was carried out under the corresponding conditions. The predicted fault level of the model was compared with the fault level obtained after the simulation burnout. As shown in Figure 7 The predicted fault level output by the model is basically consistent with the actual fault level, and only the prediction result under the condition of 230℃-40min deviates, which shows that the burnout prediction model has good performance and accuracy, and can accurately identify the fault level corresponding to different working conditions.

[0053] Table 1 configuration information of different modal input parameters

[0054] In an embodiment, before step 102, the burnout fault prediction method of the metering electric energy meter further comprises: calculating the relative ratio of the functional group absorbance of the structural component based on the infrared spectrum ratio feature; extracting the mass loss value of the local temperature segment of the structural component based on the thermogravimetric and heat flow features; extracting the element mass fraction value of the high polymer matrix of the structural component based on the element energy spectrum data; and performing normalization alignment and splicing processing on the relative ratio, the mass loss value and the element mass fraction value to form a multi-modal feature matrix of the material state representation data.

[0055] The material for making the structural component sample is usually composed of a matrix and a reinforcing phase. The matrix can be resin or plastic (PBT, PC, etc.). The reinforcing phase can be glass fiber, carbon fiber, aramid fiber, carbon nanotube, etc. to improve strength, rigidity, heat resistance or dimensional stability. Taking PBT resin as an example, the polymer matrix can be pure PBT resin or a mixture of PBT resin, flame retardant and other processing aids.

[0056] In this embodiment, by calculating the relative ratio of infrared spectrum, the mass loss value of thermal gravimetric analysis and the element mass fraction value of energy spectrum analysis respectively, and normalizing, aligning and splicing the three modal characteristics, a multi-modal feature matrix covering spectrum, thermal analysis and energy spectrum is formed. Thus, the unified representation and standardization of heterogeneous representation data are realized, the compatibility of multi-source data is improved, the time sequence or spectral structure information of each modal is retained, the material degradation process can be learned in a unified feature space in subsequent model training, and the prediction accuracy of the burnout prediction model is further improved.

[0057] In an embodiment, for the processing of infrared spectrum data, the relative ratio of the functional group absorbance of the structural component is calculated based on the infrared spectrum ratio feature, including: based on the infrared spectrum ratio feature, the ester carbonyl absorbance at the first wavelength, the ester bond absorbance at the second wavelength, the fatty chain absorbance at the third wavelength and the hydroxyl absorbance at the fourth wavelength are extracted; the ester carbonyl absorbance, the ester bond absorbance, the fatty chain absorbance and the hydroxyl absorbance are respectively divided by the benzene ring absorbance to calculate the relative ratio resistant to environmental baseline drift.

[0058] Specifically, taking flame-retardant PBT material as an example, the first wavelength can be 1720 cm -1 , corresponding to the characteristic absorption peak of ester carbonyl (C=O); the second wavelength can be 1270 cm -1 , corresponding to the characteristic absorption peak of ester bond (C-O-C); the third wavelength can be 2950 cm -1 , corresponding to the characteristic absorption peak of fatty chain (-CH2-); and the fourth wavelength can be 3400 cm -1 , corresponding to the characteristic absorption peak of hydroxyl (O-H). The benzene ring absorbance corresponds to the characteristic absorption peak of benzene ring skeleton (C=C) at 1590 cm -1 , since the benzene ring has high thermal stability in the material thermal degradation process, its absorbance changes little, and it can effectively offset the influence of baseline drift as an internal standard reference.

[0059] In this embodiment, the absorbance of each functional group is divided by the absorbance of the benzene ring to obtain relative ratios. These ratios represent the following: the ratio of ester carbonyl absorbance to benzene ring absorbance, reflecting the residual degree of ester carbonyl groups; the ratio of ester bond absorbance to benzene ring absorbance, reflecting the residual degree of ester bonds; the ratio of aliphatic chain absorbance to benzene ring absorbance, reflecting the residual degree of aliphatic chains; and the ratio of hydroxyl absorbance to benzene ring absorbance, reflecting the degree of oxidation of the material. By calculating these ratios, the distortion of absolute absorbance values ​​caused by factors such as inconsistent particle size of the test samples and differences in the testing environment can be effectively offset, improving the repeatability and comparability of infrared spectral characteristics under different testing conditions.

[0060] For example, such as Figure 2 As shown, in the red spectrum at 250℃, the absorbance of the ester carbonyl group and the ester bond is strong, with correspondingly high relative ratios. Multiple peaks and troughs can be observed, indicating that the ester bond structure is intact and many functional groups are still present. When the material is subjected to temperatures above 350℃, the ester bonds break, and the absorbance of the ester carbonyl group and the ester bond decreases significantly, with the corresponding relative ratios also decreasing. Similarly, the absorbance of the aliphatic chain decreases with increasing temperature, and the corresponding decrease in the relative ratio reflects the breaking and volatilization of the butanediol segment. The absorbance of the hydroxyl group increases with the intensification of the material's oxidation, and the increase in the relative ratio reflects the degree of oxidative degradation of the material. The purple spectrum at 450℃ shows an extremely flat curve, indicating that the functional groups have disappeared through chemical reactions. Therefore, the changes in FTIR can reflect the degree of decomposition of the PBT backbone and are positively correlated with the PBT mass retention parameter.

[0061] In one embodiment, the processing of elemental energy spectrum data, which extracts the elemental mass fraction of the polymer matrix of the structural component based on the elemental energy spectrum data, includes: identifying and filtering out the elemental spectral peaks of the reinforcing phase in the elemental energy spectrum data; calculating the relative mass percentage of the constituent elements of the polymer matrix in the elemental energy spectrum data after filtering out the reinforcing phase, and generating the mass fraction of the polymer matrix.

[0062] In this embodiment, by pre-identifying and eliminating the energy spectrum peaks of the reinforcing phase, and only calculating the relative proportions of each element within the polymer matrix, the true residual level of the flame retardant and the degree of carbonization of the matrix can be restored. This makes the extracted mass fraction value a direct evidence of the material's fire resistance degradation, avoiding prediction bias caused by interference from composite material components.

[0063] For example, taking glass fiber as the reinforcing phase, the constituent elements of the glass fiber include elements corresponding to inorganic fillers such as silicon, aluminum, and calcium. The constituent elements of the polymer matrix include base materials and flame retardant components such as carbon, oxygen, bromine, and antimony.

[0064] For the polymer matrix containing bromine antimony flame retardant, the mechanism of action of the flame retardant is: in the low temperature stage, it mainly carries out gas phase flame retardation, Br generates hydrogen bromide (HBr) gas after heating, and hydrogen bromide reacts with antimony oxide (Sb2O3) to generate antimony tribromide (SbBr3) gas, which can capture ·OH and ·H free radicals, thereby inhibiting the combustion reaction. At high temperature, the flame retardant mainly carries out condensed phase flame retardation. SbBr3 promotes the formation of a dense ring aromatic carbon layer in the matrix material, which is manifested as an increase in the C=C peak in the infrared spectrum. The carbon layer formed can isolate oxygen from the outside and hinder heat conduction, and can inhibit the escape of volatile substances from the inside. During the thermal degradation process of the material, the flame retardant decomposes to generate hydrogen bromide gas and antimony tribromide gas and volatilizes, resulting in a continuous decrease in the relative mass percentage of bromine and antimony elements.

[0065] By comparison Figure 3 and Figure 4 It can be found that at 250°C, the sample has not started to heat, and the matrix part composed of PBT and the rod-shaped filler glass fiber can be observed. After the temperature rises to 450°C, the flame retardant appears to be degraded and volatilized, and a large number of independent rod-shaped fibers are fully exposed and interlaced with each other, and almost no continuous matrix phase can be seen, indicating that the thermal decomposition of the matrix material at 450°C is basically complete. Moreover, from the comparison of the EDS spectrum, it can also be found that the element content (wt%) of the matrix material and the proportion of Br element (σ) at 450°C decreases compared to that at 250°C. Therefore, the element mass fraction value can quantify the loss degree of the polymer matrix in the thermal reaction process, and provide important material state information for burnout prediction.

[0066] In an embodiment, for the processing of thermogravimetric and heat flow data, the mass loss value of the local temperature segment of the structure component is extracted based on the thermogravimetric and heat flow feature extraction structure, including: intercepting the local temperature segment in the main thermal decomposition interval from the thermogravimetric curve, and collecting the mass retention rate and its first derivative value corresponding to multiple temperature points in the temperature segment at equal temperature intervals, to form a mass loss time sequence; intercepting the heat flow value sequence in the same temperature segment from the synchronous heat flow curve to form a heat effect time sequence; aligning the mass loss time sequence and the heat effect time sequence along the temperature axis and performing point-by-point outer product or channel splicing to generate a two-dimensional mass loss value.

[0067] Specifically, taking the flame-retardant PBT material as an example, the local temperature segment can be selected as the center of the target temperature 300℃ at which the material undergoes severe weight loss, and the TG curve and DSC curve data in the range of T±50℃ are extracted. Each sampling 50 data points, the instantaneous mass loss rate of the thermogravimetric curve at each point is extracted as the element of the mass loss time sequence vector, and the heat flow value at each point on the differential scanning calorimetry curve is extracted as the element of the thermal effect time sequence vector. By point-by-point outer product, a joint distribution matrix of N*N is constructed, or the two vectors are directly spliced into an N*2 matrix, that is, the mass loss value is constructed. The mass loss value can capture the mass change gradient information and heat flow response characteristics of the material in different temperature intervals, reflecting the stage law of the thermal degradation process.

[0068] As shown in FIG. 8, the orange TG curve is the mass change during continuous heating, and the mass is close to 100% at 250℃, and after the cliff drops, the mass is close to 40% at 450℃, indicating that the PBT is gradually decomposed. Therefore, it can be known that the temperature and time can also reflect the degree of loss of the material in the thermal reaction process, and are negatively correlated with the mass retention parameter. Figure 5

[0069] In this embodiment, by reconstructing the one-dimensional thermogravimetric and heat flow signals in the local temperature segment into a two-dimensional value form, the kinetic process of mass attenuation and the accompanying thermal effect change can be structurally coupled and expressed in the temperature dimension, so that the extracted mass loss value records the mass loss rate, the stage mass residue, and the thermodynamic fingerprint information such as phase change or oxidation exothermic of the material. Therefore, the mass loss value can provide more detailed degradation mode information for the burnout prediction model, and compared with a single feature quantity, the mass loss value can more sensitively distinguish the subtle differences in the early thermal degradation of the material, which helps to enhance the recognition ability and prediction robustness of the model for the burnout precursor of the structural component.

[0070] In actual application scenarios, due to the availability and test cost of the characterization test equipment, the structural components of an electric energy meter can not be subjected to characterization tests of all modalities, resulting in missing data of one or more modalities in the material state characterization data. Therefore, before step 102, the burnout fault prediction method of the metering electric energy meter further includes: when there is missing data in the material state characterization data, performing masking and zero processing on the missing modality data, so as to set the feature values of the missing data to zero.

[0071] Further, in the process of training the prediction fault model, the feature splicing layer determines the missing degree of the modality with missing data based on the feature values set to zero, and reduces the weight of the associated mapping feature corresponding to the modality with missing data based on the missing degree, and increases the relative weight of the low-dimensional mapping feature and the local time sequence feature corresponding to the thermal stress parameter, so as to perform series splicing based on the adjusted weight.

[0072] ​It should be noted that the modal refers to a whole class of data types / data sources. The modal can specifically include FTIR spectral data and EDS element content data, etc. static or quasi-static indicators. For example, thermal analysis data such as TG curve, DSC curve form a dynamic thermal analysis modal, and thermal stress parameters such as temperature and time form a thermal stress parameter modal.

[0073] In this embodiment, when some modalities cannot be obtained, the corresponding input branch is filled with a zero vector, so that the model can automatically increase the weight contribution of the thermal stress parameter feature and reduce the weight contribution of the missing modal feature through the feature weight distribution mechanism learned during the training process, even if the material state data is incomplete or cannot be obtained. The adaptive training of the missing modal is realized to ensure that the model can effectively estimate the material quality parameters according to the thermal stress parameters. Thus, in the case of incomplete data, the degradation reasoning prediction is completed in the running environment, the data acquisition and experimental cost required for material state detection in actual application is reduced, artificial artifacts that violate physical laws are avoided due to filling in the mean value or interpolation, and the continuity and stability of the model output are ensured.

[0074] In an embodiment, in order to convert the damage state of the material layer into a practical safety protection action, after determining the burnout failure level, the corresponding safety control instruction can be triggered to prevent accidents caused by the fault of the electric energy meter. Specifically, the triggering process of the safety control instruction includes the following cases: (1) When the to-be-tested electric energy meter is in the first type of processing level, write the load reduction instruction to the load control register of the target electric energy meter.

[0075] In this embodiment, the triggered load reduction instruction is a kind of active thermal stress relief measure, which directly reduces the joule heat generation rate by limiting the current amplitude flowing through the terminal, so that the component temperature has the opportunity to fall back to the thermal equilibrium safety zone, thereby reversing or delaying the process of cumulative damage. This soft control method avoids false tripping caused by instantaneous temperature rise or slight aging, effectively ensuring the continuous power experience of users.

[0076] (2) When the to-be-tested electric energy meter is in the second type of processing level, send a warning data packet containing the identity of the target electric energy meter and the burnout failure level to the remote operation and maintenance server through the network communication protocol.

[0077] In this embodiment, in the second type of processing level state, simply reducing the load is difficult to curb the trend of material degradation, and manual intervention for component replacement or deep maintenance must be introduced. By uploading the warning data packet containing the unique identity, the operation and maintenance platform can accurately locate the faulty meter and evaluate the risk level, which prevents the fault from further expanding and causing safety accidents, and avoids wasting human resources caused by blind on-site troubleshooting.

[0078] (iii) When the energy meter under test is in the third category of processing level, a high-level pulse signal is output to the trip drive circuit of the target energy meter to control the trip drive circuit to drive the trip mechanism to act, so that the main current loop between the power input side and the load side of the energy meter under test is disconnected.

[0079] The tripping drive circuit may include an electromagnetic trip unit or a thermal trip unit, and the duration and amplitude of the high-level pulse signal are set according to the specifications of the trip unit.

[0080] In this embodiment, the system no longer attempts any adjustments or waits for manual response; instead, it immediately executes the highest-priority physical isolation operation. A high-level pulse signal directly drives the built-in trip unit to forcibly disconnect the main circuit, fundamentally cutting off the energy input path. This rigid protection mechanism based on the ultimate failure state of materials constitutes the last line of defense for the safety of the electricity meter, ensuring that even under extremely harsh operating conditions, it can effectively prevent fire accidents and truly safeguard the lives and property of the people.

[0081] It is understood that the burn-off fault levels predicted by the model can be set to multiple levels with reference to the mapping relationship with the preset parameter range. For example, there may be 10 levels from 0 to 9, or 4 levels from 1 to 4. This application does not specifically limit the presentation format of the burn-off fault levels. The first, second, and third processing levels are used to execute different fault handling strategies and can be associated with at least one burn-off fault level. For example, the first processing level corresponds to burn-off faults of levels 1 to 2 and is used to perform load reduction operation; the second processing level corresponds to burn-off faults of levels 3 to 6 and is used to perform shutdown maintenance; the third processing level corresponds to burn-off faults of levels 7 to 10 and is used to perform circuit disconnection. The division of each processing level can also be set according to actual processing needs, as long as corresponding processing measures can be taken according to the burn-off fault levels predicted by the model.

[0082] In addition, in practical applications, the color change of the meter's terminal surface and the amount of structural deformation can be used as auxiliary indicators to further improve the accuracy of determining the burn-out fault level.

[0083] The method for predicting burnout faults in electricity meters provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0084] It should be noted that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0085] The user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are information and data allowed by the user or fully allowed by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose to allow or refuse.

[0086] Further, as shown in Figure 8 As a specific implementation of the burnout failure prediction method of the metering electric energy meter, the embodiments of the present application provide a burnout failure prediction device 300 of the metering electric energy meter, which comprises an acquisition module 301, a training module 302 and a prediction module 303.

[0087] The acquisition module 301 is configured to acquire material state representation data of a structure component sample of the electric energy meter under different thermal stress parameters and an actual failure level of the electric energy meter. The training module 302 is configured to input the thermal stress parameters and the material state representation data as training samples into a neural network model, extract features in parallel through the neural network model, and constrain the correlation mapping features of the thermal stress parameters in the hidden layer feature space with the material state representation data, so as to establish a correlation mapping relationship between the material state representation data and the thermal stress parameters, determine a material quality parameter of the structure component sample based on the fusion features of the features of the thermal stress parameters and the constrained correlation mapping features, and generate a predicted failure level; and iteratively train the neural network model to obtain a burnout prediction model corresponding to the structure component sample, with the difference between the predicted failure level and the actual failure level being minimized, wherein the model parameters of the burnout prediction model are fixed on the correlation mapping relationship. The prediction module 303 is configured to input a target thermal stress parameter of a structure component of a to-be-tested electric energy meter into the corresponding burnout prediction model, and the burnout prediction model reconstructs a corresponding target material state feature from the target thermal stress parameter in the hidden layer feature space based on the correlation mapping relationship, and determines a material quality parameter of the to-be-tested electric energy meter and outputs a burnout failure level based on the fusion features of the features of the target thermal stress parameter and the target material state feature.

[0088] Further, the structural component includes at least one of a terminal component, a terminal cover component and an insulating base component of the to-be-tested electric energy meter; the material state characterization data includes at least one of infrared spectrum ratio characteristics of the structural component sample, thermogravimetric and heat flow characteristics of the structural component sample with temperature change, and element energy spectrum data of the structural component sample; and the thermal stress parameter includes at least one of a heating temperature and a continuous heating time.

[0089] Further, the burnout failure prediction device 300 of the electric energy meter further includes: a data conversion module (not shown in the figure), configured to calculate a relative ratio of functional group absorbance of the structural component based on the infrared spectrum ratio characteristics; extract a mass loss value of a local temperature segment of the structural component based on the thermogravimetric and heat flow characteristics; extract an element mass fraction value of a polymer matrix of the structural component based on the element energy spectrum data; and perform normalization alignment and splicing processing on the relative ratio, the mass loss value and the element mass fraction value to form a multi-modal feature matrix of the material state characterization data, so as to input the multi-modal feature matrix to the neural network model.

[0090] Further, the data conversion module is specifically configured to extract ester carbonyl absorbance at a first wavelength, ester bond absorbance at a second wavelength, fatty chain absorbance at a third wavelength and hydroxyl absorbance at a fourth wavelength based on the infrared spectrum ratio characteristics; and divide the ester carbonyl absorbance, the ester bond absorbance, the fatty chain absorbance and the hydroxyl absorbance by benzene ring absorbance respectively to calculate a relative ratio resistant to environmental baseline drift.

[0091] Further, the data conversion module is specifically configured to identify and filter out enhanced corresponding element energy spectrum data from the element energy spectrum data of the structural component sample to obtain element energy spectrum data corresponding to the polymer matrix; calculate a relative mass percentage of constituent elements of the polymer matrix based on the element energy spectrum data corresponding to the polymer matrix to generate the mass fraction value.

[0092] Furthermore, the neural network model includes multiple parallel multilayer perceptron subnetworks, multiple convolutional neural subnetworks, as well as a feature concatenation layer, an association mapping layer, and a fully connected regression network. The training module 302 is specifically used to invoke multiple multilayer perceptron subnetworks to perform low-dimensional vector mapping on discrete thermal stress parameters and material state characterization data in parallel, obtaining low-dimensional mapping features; invoke multiple convolutional neural subnetworks to extract local temporal features on continuous thermal stress parameters and material state characterization data in parallel, obtaining local temporal features; and invoke the association mapping layer to map the low-dimensional mapping features corresponding to the thermal stress parameters and the local temporal features to the hidden layer feature space. A correlation mapping feature is formed between the data. In the hidden layer feature space, the low-dimensional mapping feature corresponding to the material state characterization data and the local temporal feature are used as supervision benchmarks to supervise and constrain the correlation mapping feature, so that the correlation mapping feature and the feature corresponding to the material state characterization data have a consistent feature distribution in the hidden layer feature space, forming a correlation mapping relationship. The feature splicing layer is called to concatenate and splice the low-dimensional mapping feature and the local temporal feature corresponding to the thermal stress parameter, as well as the supervised correlation mapping feature, to generate a fused feature. The fully connected regression network is called to map the fused feature to the material quality parameter, and the predicted fault level is output based on the matching result of the material quality parameter and the preset parameter range.

[0093] Furthermore, the burnout fault prediction device 300 for the electricity meter also includes: The missing data handling module (not shown in the figure) is used to set the feature value of the missing data to zero when there is missing data in the material state characterization data. The training module 302 is also used in the process of training the fault prediction model to call the feature splicing layer to determine the degree of missing data of the mode with missing data based on the feature quantity set to zero, and reduce the weight of the associated mapping feature corresponding to the mode with missing data based on the degree of missing data, and increase the relative weight of the low-dimensional mapping feature and local temporal feature corresponding to the thermal stress parameter, so as to perform concatenation splicing based on the adjusted weight.

[0094] Furthermore, the burn-out fault levels include Class I handling level, Class II handling level, and Class III handling level. The burn-out fault prediction device 300 for the electricity meter also includes: The fault handling module (not shown in the figure) is used to write a load reduction command to the load control register of the energy meter under test when the energy meter under test is in the first processing level; when the energy meter under test is in the second processing level, it sends a warning data packet containing the identification of the energy meter under test and the burn-out fault level to the remote operation and maintenance server through the network communication protocol; when the energy meter under test is in the third processing level, it outputs a high-level pulse signal to the trip drive circuit of the energy meter under test to control the trip drive circuit to drive the trip mechanism to disconnect the main current loop between the power input side and the load side of the energy meter under test.

[0095] The specific definition of the burnout fault prediction device of the metering electric energy meter can refer to the definition of the burnout fault prediction method of the metering electric energy meter in the above, which will not be repeated here. Each module in the burnout fault prediction device of the metering electric energy meter described above can be realized by software, hardware and a combination thereof in whole or in part. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0096] Based on the above method as shown in Figure 1 Accordingly, the embodiments of the present application also provide a readable storage medium having a computer program stored thereon, which is executed by a processor to implement the above burnout fault prediction method of the metering electric energy meter as shown in Figure 1

[0097] Based on such understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a U disk, a mobile hard disk, etc.), and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the method of each implementation scenario of the present application.

[0098] Based on the above method as shown in Figure 1 and the device as shown in Figure 8 In order to achieve the above purpose, as shown in Figure 9 The embodiments of the present application also provide a computer device, which includes a processor 401 and a memory 402, and the memory 402 has a program or instructions stored thereon, which can be run on the processor 401, and the program or instructions are executed by the processor 401 to implement the above burnout fault prediction method of the metering electric energy meter as shown in Figure 1

[0099] ​​The memory 402 can be used to store software programs and various data. The memory 402 can mainly include a first storage area storing programs or instructions and a second storage area storing data, wherein the first storage area can store an operating system, application programs or instructions required by at least one function (such as a sound playing function, an image playing function, etc.), and the like. In addition, the memory 402 can include a volatile memory or a non-volatile memory, or the memory 402 can include both a volatile memory and a non-volatile memory. The non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM), or a flash memory. The volatile memory can be a Random Access Memory (RAM), a Static RAM (SRAM), a Dynamic RAM (DRAM), a Synchronous DRAM (SDRAM), a Double Data Rate SDRAM (DDR SDRAM), an Enhanced SDRAM (ESDRAM), a Synch link DRAM (SLDRAM), and a Direct Rambus RAM (DRRAM). The memory 402 in the embodiments of the present application includes but is not limited to these and any other suitable types of memory.

[0100] The processor 401 can include one or more processing units; optionally, the processor 401 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and an application program, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 401.

[0101] The computer device can specifically be a personal computer, a server, a network device, and the like.

[0102] Optionally, the computer device can further include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a WI-FI module, and the like. The user interface can include a display, an input unit such as a keyboard, and the like. Optionally, the user interface can further include a USB interface, a card reader interface, and the like. The network interface can optionally include a standard wired interface, a wireless interface (such as a Bluetooth interface, a WI-FI interface), and the like.

[0103] Those skilled in the art can understand that the computer device structure provided by the embodiment does not constitute a limitation on the computer device, and can include more or fewer components, or combine certain components, or different component arrangements.

[0104] Those skilled in the art can clearly understand from the above description of the embodiments that the present application can be implemented by means of software with a necessary general hardware platform, or by hardware.

[0105] Those skilled in the art can understand that the drawings are only a schematic diagram of a preferred implementation scenario, and the modules or flows in the drawings are not necessarily essential for implementing the present application. Those skilled in the art can understand that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the description of the implementation scenario, or can be changed and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined as one module, or can be further split into multiple sub-modules.

[0106] The above serial number of the present application is only for description, and does not represent the advantages and disadvantages of the implementation scenario. The above disclosure is only several specific implementation scenarios of the present application, but the present application is not limited thereto, and any changes that can be thought of by those skilled in the art should fall within the protection scope of the present application.

Claims

1. A method for predicting burnout faults in an electricity meter, characterized in that, include: Obtain material state characterization data of structural component samples of the electricity meter under different thermal stress parameters and the actual fault level of the electricity meter; The thermal stress parameters and the material state characterization data are used as training samples and input into the neural network model. Features are extracted in parallel through the neural network model, and the correlation mapping features of the thermal stress parameters are constrained by the material state characterization data in the hidden layer feature space to establish the correlation mapping relationship between the material state characterization data and the thermal stress parameters. Based on the fusion features of the thermal stress parameters and the constrained correlation mapping features, the material quality parameters of the structural component sample are determined, and the predicted fault level is generated. With the goal of minimizing the difference between the predicted fault level and the actual fault level, the neural network model is iteratively trained to obtain the burn-off prediction model corresponding to the structural component sample, wherein the model parameters of the burn-off prediction model are fixed with the correlation mapping relationship; The target thermal stress parameters of the structural components of the energy meter under test are input into the corresponding burn-off prediction model. Based on the correlation mapping relationship, the burn-off prediction model reconstructs the corresponding target material state features from the target thermal stress parameters in the hidden layer feature space. Based on the fusion features of the target thermal stress parameters and the target material state features, the model determines the material quality parameters of the energy meter under test and outputs the burn-off fault level.

2. The method for predicting burnout faults in electricity meters according to claim 1, characterized in that, The structural components include at least one of the following: the wiring terminal component of the energy meter, the terminal cover component, and the insulating base component; The material state characterization data includes at least one of the following: infrared spectral ratio characteristics of structural component samples, thermogravimetric and heat flow characteristics of structural component samples as a function of temperature, and elemental energy spectrum data of structural component samples. The thermal stress parameters include at least one of the following: heating temperature and continuous heating time.

3. The method for predicting burnout faults in electricity meters according to claim 2, characterized in that, Before inputting the thermal stress parameters and the material state characterization data as training samples into the neural network model, the method further includes: The relative ratio of the absorbance of functional groups in the structural components is calculated based on the infrared spectral ratio characteristics. Based on the aforementioned thermogravimetric and heat flow characteristics, the mass loss value of local temperature segments of structural components is extracted; The elemental mass fraction values ​​of the polymer matrix of the structural components are extracted based on the elemental energy spectrum data. The relative ratio, the mass loss value, and the element mass fraction are normalized, aligned, and stitched together to form a multimodal feature matrix of the material state characterization data, which is then input into the neural network model.

4. The method for predicting burnout faults in electricity meters according to claim 3, characterized in that, The calculation of the relative ratios of the functional group absorbance of the structural components based on the infrared spectral ratio characteristics includes: Based on the infrared spectral ratio characteristics, the absorbance of the ester carbonyl group at the first wavelength, the absorbance of the ester bond at the second wavelength, the absorbance of the aliphatic chain at the third wavelength, and the absorbance of the hydroxyl group at the fourth wavelength are extracted. The relative ratios that resist environmental baseline drift are calculated by dividing the absorbance of the ester carbonyl group, the absorbance of the ester bond, the absorbance of the aliphatic chain, and the absorbance of the hydroxyl group by the absorbance of the benzene ring, respectively. The extraction of elemental mass fraction values ​​of the polymer matrix of the structural components based on the elemental energy spectrum data includes: The elemental energy spectrum data corresponding to the enhancement in the elemental energy spectrum data of the structural component sample is identified and filtered out to obtain the elemental energy spectrum data corresponding to the polymer matrix. The relative mass percentage of the constituent elements of the polymer matrix is ​​calculated based on the elemental energy spectrum data corresponding to the polymer matrix, and the mass fraction value is generated.

5. The method for predicting burnout faults in electricity meters according to claim 1, characterized in that, The neural network model includes multiple parallel multilayer perceptron subnetworks, multiple convolutional neural subnetworks, and feature splicing layers, correlation mapping layers, and fully connected regression networks. Features are extracted in parallel using the neural network model, and the correlation mapping features of the thermal stress parameters are constrained in the hidden feature space using the material state characterization data to establish a correlation mapping relationship between the material state characterization data and the thermal stress parameters. Based on the fusion features of the thermal stress parameters and the constrained correlation mapping features, the material quality parameters of the structural component sample are determined, and a predicted fault level is generated, including: Multiple multilayer perceptron subnetworks are invoked to perform low-dimensional vector mapping on the discrete thermal stress parameters and the material state characterization data in parallel, thereby obtaining low-dimensional mapping features; Multiple convolutional neural subnetworks are invoked to extract local temporal features from the continuous thermal stress parameters and material state characterization data in parallel, thereby obtaining local temporal features; The association mapping layer is invoked to map the low-dimensional mapping features and local temporal features corresponding to the thermal stress parameters to the hidden layer feature space to form the association mapping features. In the hidden layer feature space, the low-dimensional mapping features and local temporal features corresponding to the material state characterization data are used as supervision benchmarks to supervise and constrain the association mapping features, so that the association mapping features and the features corresponding to the material state characterization data have a consistent feature distribution in the hidden layer feature space, thus forming the association mapping relationship. The feature splicing layer is invoked to concatenate and splice the low-dimensional mapping features and local temporal features corresponding to the thermal stress parameters, as well as the supervised and constrained correlation mapping features, to generate fused features; The fully connected regression network is invoked to map the fused features to the material quality parameters, and the predicted fault level is output based on the matching result of the material quality parameters and the preset parameter range.

6. The method for predicting burnout faults in electricity meters according to claim 5, characterized in that, The method further includes: When there is missing data in the material state characterization data, the feature value of the missing data is set to zero; During the training of the fault prediction model, the feature splicing layer is called to determine the degree of missing data in the modes based on the feature quantity set to zero, and the weight of the associated mapping feature corresponding to the mode with missing data is reduced based on the degree of missing data, while the relative weight of the low-dimensional mapping feature and local temporal feature corresponding to the thermal stress parameter is increased, so as to perform concatenation splicing based on the adjusted weight.

7. The method for predicting burnout faults in electricity meters according to claim 1, characterized in that, The burn-out failure level is divided into three treatment levels, including a first treatment level, a second treatment level, and a third treatment level. The method further includes: When the energy meter under test is in the first processing level, write a load reduction instruction to the load control register of the energy meter under test; When the energy meter under test is in the second processing level, a warning data packet containing the identification of the energy meter under test and the burn-out fault level is sent to the remote operation and maintenance server through the network communication protocol. When the energy meter under test is in the third processing level, a high-level pulse signal is output to the tripping drive circuit of the energy meter under test to control the tripping drive circuit to drive the tripping mechanism to disconnect the main current loop between the power input side and the load side of the energy meter under test.

8. A device for predicting burnout faults in an electricity meter, characterized in that, The device includes: The acquisition module is used to acquire material state characterization data of structural component samples of the electricity meter under different thermal stress parameters and the actual fault level of the electricity meter. The training module is used to input the thermal stress parameters and the material state characterization data as training samples into a neural network model. The neural network model extracts features in parallel and constrains the correlation mapping features of the thermal stress parameters in the hidden layer feature space using the material state characterization data to establish a correlation mapping relationship between the material state characterization data and the thermal stress parameters. Based on the fusion features of the thermal stress parameters and the constrained correlation mapping features, the material quality parameters of the structural component sample are determined, and a predicted fault level is generated. With the goal of minimizing the difference between the predicted fault level and the actual fault level, the neural network model is iteratively trained to obtain the burn-off prediction model corresponding to the structural component sample, wherein the model parameters of the burn-off prediction model are fixed with the correlation mapping relationship; The prediction module is used to input the target thermal stress parameters of the structural components of the energy meter under test into the corresponding burn-off prediction model. Based on the correlation mapping relationship, the burn-off prediction model reconstructs the corresponding target material state features from the target thermal stress parameters in the hidden layer feature space, and determines the material quality parameters of the energy meter under test and outputs the burn-off fault level based on the fusion features of the target thermal stress parameters and the target material state features.

9. A readable storage medium having a program or instructions stored thereon, characterized in that, When the program or instructions are executed by the processor, they implement the method for predicting burnout faults of the metering energy meter as described in any one of claims 1 to 7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for predicting burnout faults of a metering energy meter as described in any one of claims 1 to 7.

Images