A method and system for determining an instrument calibration period

Abstract

The present application relates to the technical field of data analysis, and particularly relates to a kind of instrument verification period determination method and system, comprising the following steps: S1: real-time acquisition instrument working waveform data stream, feature waveform fragment is extracted by edge computing node;S2: feature waveform fragment is input into pre-trained phase state identification model, and current working phase state label is output;S3: based on phase state label, corresponding deterioration rate calculator is activated, and real-time deterioration increment is generated;S4: according to cumulative deterioration increment and phase state transfer path, obtain next verification period.The present application improves the accuracy of remaining life prediction, is particularly suitable for frequently fluctuating working condition or complex operation mode of field instrument, can effectively identify the deterioration acceleration risk caused by frequent working condition fluctuation, and has higher prediction consistency and safety in complex environment than traditional period estimation method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, and particularly relates to a method and system for determining an instrument calibration period. BACKGROUND

[0002] With the continuous development of intelligent metering, industrial automation and urban Internet of Things, various types of field instruments (such as electric energy meters, pressure transmitters, flow meters, etc.) play a key role in energy metering, safety monitoring and process control. In order to ensure the accuracy and reliability of the operation of the instruments, the current management system generally adopts a periodic calibration system, which sets a fixed period according to the type of the instrument and the service life. However, in actual operation, the environment, load condition and use frequency of the instrument are significantly different, and the traditional calibration strategy based on fixed time or human experience has been difficult to meet the periodic management needs in high dynamic and high reliability scenarios.

[0003] Some methods in the prior art attempt to estimate the degree of degradation by running time or cumulative use times, but fail to effectively integrate the operating state characteristics, working condition fluctuations and performance differences between individuals of the instrument, resulting in two typical problems in the calibration period. On the one hand, some instruments are sent for inspection in advance before the degradation is formed, causing waste of resources. On the other hand, some instruments under high frequency state fluctuations exist the risk of overuse due to the failure to identify performance degradation in time, affecting the safety of equipment operation.

[0004] In addition, the existing degradation evaluation methods based on waveform analysis mostly rely on offline large models for processing, which are difficult to realize real-time deployment on edge nodes, and mostly ignore the nonlinear influence of different operating states on the degradation rate, lack dynamic modeling capability between the phase state and the degradation behavior, resulting in poor stability and insufficient interpretability of the prediction results. SUMMARY

[0005] The present application provides a method and system for determining an instrument calibration period, which integrates a new method of operating state recognition, degradation modeling and period dynamic adjustment, constructs an intelligent period management system covering the whole link of perception, calculation and decision, and realizes a high-precision and personalized instrument calibration period determination mechanism.

[0006] A method for determining an instrument calibration period, comprising the following steps:

[0007] S1: Real-time acquisition of instrument working waveform data stream, extraction of feature waveform segments by edge computing nodes;

[0008] S2: Inputting the feature waveform segments into a pre-trained phase state recognition model to output a current working phase state label;

[0009] S3: Activating the corresponding degradation rate calculator based on the phase state label to generate a real-time degradation increment;

[0010] S4: Obtain the next detection period according to the accumulated degradation increment and the phase transition path.

[0011] Optionally, the S1 specifically comprises:

[0012] S11: Deploy multi-channel sampling at the instrument signal output end, synchronously collect voltage waveform, current waveform and timing pulse signal, and form raw waveform data;

[0013] S12: Perform sliding window interception processing on the raw waveform data, and set the window length to 500 ms;

[0014] S13: Implement dynamic baseline calibration processing on the raw waveform data in each window, and extract the following features:

[0015] Waveform distortion degree D H :

[0016] Wherein, W is the waveform point set in the current window, W std is the standard waveform template, D H measures the maximum local difference between the actual waveform and the ideal waveform, x and y represent any waveform points in W and W std , respectively, ‖x-y‖ represents the Euclidean distance between points x and y, which is used to measure the spatial distance of two waveform sampling points, sup represents the upper bound, that is, the "maximum lower bound" among all values, which is used to find the maximum value among the nearest distances from a point to all points in another set, and inf represents the lower bound, that is, the "minimum upper bound" among all values, which is used to find the nearest distance from a point to all points in another set;

[0017] Pulse density ρ: Wherein, N pulse represents the number of pulses with amplitude exceeding the pulse threshold θ per unit time, T is the time window length, and the pulse threshold θ is dynamically updated based on the average amplitude of the previous 10 windows;

[0018] Energy entropy H: Wherein, represents the energy proportion of the i-th wavelet packet frequency band, E i is the energy value of the i-th frequency band, E j is the energy value of the j-th frequency band, and n is the number of frequency bands after wavelet packet decomposition;

[0019] S14: Package the features extracted from each window into a three-dimensional feature vector f = [D H , ρ, H], and channelize and embed each three-dimensional feature vector of each window to form a pseudo-image structure input tensor F inThe segment importance score value is input to a lightweight convolution scoring network of the edge computing node to calculate a segment importance score value s, only feature waveform segments with a score s>τ are retained, and τ is an experience set score threshold.

[0020] Optionally, the segment importance score value is defined as: wherein, represents MobileNetV3 feature transformation after channel pruning, GAP(·) represents global average pooling, W1,b is a linear transformation parameter, and σ(·) represents a normalized score value.

[0021] Optionally, the S2 specifically includes:

[0022] S21: performing time sequence alignment and feature dimension reduction on the feature waveform segments to generate a standardized space-time feature matrix;

[0023] S22: inputting the space-time feature matrix into a double-flow space-time attention network of the phase state recognition model to extract waveform morphology features and pulse time sequence correlation features, respectively;

[0024] S23: fusing the weighted waveform morphology features and the pulse time sequence correlation features through a cross-modal feature fusion gating mechanism, calculating a fusion weight, and generating a hybrid feature vector F 混合 ;

[0025] S24: matching the hybrid feature vector with a predefined phase state prototype cluster based on the hybrid feature vector, and outputting a current working phase state label using a dynamic radius clustering algorithm.

[0026] Optionally, the S24 includes inputting the hybrid feature vector F 混合 to clustering, matching an existing working phase state prototype cluster, and determining a current cluster label using a dynamic radius clustering algorithm:

[0027] Each phase state cluster C i includes a prototype center μ i and a dynamic radius R i , wherein: , wherein R b,i is a basic radius of the cluster i, and is usually a sample standard deviation, is an environment adaptive coefficient, defined as: S 噪声 is a currently measured environmental noise intensity, S 基线 is a noise intensity in a reference environment;

[0028] Matching logic:

[0029] If F 混合 - μ i < R i , then the working phase state cluster C iThat is, if the mixed feature vector falls within the dynamic radius of a certain cluster, directly output the corresponding working phase state label;

[0030] If it falls into the overlapping area of multiple clusters, select the working phase state cluster with the largest correlation to the current degradation increment.

[0031] Optionally, the S3 specifically comprises:

[0032] S31: According to the mapping of the working phase state label to the pre-defined calculator configuration table, load the multi-dimensional degradation model of the corresponding phase state;

[0033] S32: Real-time acquisition of the current environmental parameter set of the instrument, input the basic degradation rate into the degradation model;

[0034] S33: Based on the deviation degree of the historical degradation data of the same phase state and the current basic degradation rate, generate a dynamic compensation factor;

[0035] S34: Fusion of the basic degradation rate, the dynamic compensation factor and the cumulative running time, output the real-time degradation increment.

[0036] Optionally, the degradation model is used to calculate the basic degradation rate, which is represented as:

[0037] Wherein, λ 当前 represents the basic degradation rate under the current environment, λ 基准 represents the reference degradation rate under the calibration environment, ΔT is the temperature difference between the current temperature and the reference temperature, RH 当前 represents the current relative humidity, RH 标定 is the reference humidity at the time of model calibration, ω is the material physical property coefficient, which represents the temperature-stress response sensitivity. The environmental parameters can include vibration spectrum, dust concentration;

[0038] The degradation model also introduces a deviation compensation mechanism based on historical degradation data to obtain a dynamic compensation factor η;

[0039] Synthesis and output of real-time degradation increment: finally, by fusing the basic degradation rate, the dynamic compensation factor and the phase state running time, the real-time degradation increment value at the current time is generated:

[0040]

[0041] Wherein, ΔQ is the cumulative degradation increment, t0, t are the start and end times of the degradation calculation respectively, dt represents the small time increment when performing integral operation on the continuous time domain, which is used to time the real-time degradation rate on the interval [t0, t].

[0042] Optionally, the S4 specifically comprises:

[0043] S41: Calculate the predicted remaining instrument life T based on cumulative degradation increment. 剩余 The formula is:

[0044] Where, ΔQ 平均 Q is the average degradation increment per unit time under the current phase state. 累积 Given the current accumulated total degradation, aggregate the phase states ΔQ from the aforementioned outputs and calculate the time-weighted composite Q. 累积 Q 阈值 To preset the degradation failure threshold, Q is dynamically loaded according to the instrument type. 阈值 Q 阈值 ∈[0.85,0.95],

[0045] S42: Analyze the phase transition path graph and extract the following risk parameters:

[0046] High-frequency transition phase pairs: Count the number F of high-frequency transition paths with an adjacent phase transition frequency > 2 times / hour. 高频 ;

[0047] Critical degradation nodes: Identifying the causes of ΔQ 平均 A sudden increase of more than 50% in phase transition events;

[0048] S43: Construct the two-domain adjustment coefficient K;

[0049] S44: Update cycle using a sliding window mechanism: T 新 =T 基准 ·e -K·γ , among which, T 新 The updated calibration cycle is the one currently calculated, and γ is the environmental acceleration factor, which is dynamically corrected based on real-time data from the temperature and humidity sensors.

[0050] Optionally, the dual-domain adjustment coefficient K is calculated as follows: Where α and β are weighting coefficients, α = 0.6, β = 0.4, F 高频 F represents the number of high-frequency transfer paths. 基准 T represents the historical average number of transfer paths. 标准 The standard verification cycle corresponding to the equipment type.

[0051] An instrument calibration cycle determination system, used to implement the above-mentioned instrument calibration cycle determination method, includes the following modules:

[0052] The data acquisition module is used to acquire the instrument's working waveform data stream in real time and transmit it to the edge computing node;

[0053] The feature extraction module is used to process the waveform data stream in the edge computing node and extract feature waveform segments;

[0054] a phase state recognition module, configured to input the feature waveform segment into a pre-trained phase state recognition model, and output a current working phase state label;

[0055] a degradation evaluation module, configured to call a corresponding degradation rate calculator based on the phase state label, and generate a real-time degradation increment;

[0056] a cycle decision module, configured to output a corresponding next calibration cycle according to the accumulated degradation increment and a phase state transition path.

[0057] The present application has the following advantages:

[0058] The present application realizes fine recognition of the running state of the instrument by constructing a double-flow space-time attention model, which is superior to the traditional coarse-grained state classification method based on waveform amplitude or mean square error; combined with a dynamic radius clustering algorithm for phase state prototype matching, it has high resolution modeling capability for complex state transition process, ensures that the selection of the subsequent degradation model has stronger situational adaptability and robustness, realizes the degradation evolution calculation logic of "state-driven modeling", improves the accuracy of the remaining life prediction, and is especially suitable for field instruments with frequent working condition fluctuations or complex running modes.

[0059] The present application, by constructing a historical degradation sample deviation compensation mechanism based on Mahalanobis distance, enables the degradation increment model to have dynamic correction capability across device individual differences, at the same time, by analyzing the phase state transition path atlas, extracting high-frequency transition path number and key degradation nodes and other risk indicators, and incorporating them into the weight factor calculation of cycle adjustment, realizes the cycle self-adaptive adjustment strategy of "remaining life-behavior frequency" dual-domain fusion, which can effectively identify the degradation acceleration risk caused by frequent working condition fluctuations, and has higher prediction consistency and safety in complex environment than the traditional cycle estimation method.

[0060] The present application, by channel pruning and structure pruning optimization of MobileNetV3 network, enables it to complete waveform segment scoring and feature extraction within millisecond level latency on the edge node, and combined with the cycle dynamic updating mechanism constructed by sliding window and exponential decay model, can realize high frequency and low resource consumption execution of cycle adjustment. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0062] Figure 1A method flowchart of an embodiment of the present application;

[0063] Figure 2 A system function module schematic diagram of an embodiment of the present application. DETAILED DESCRIPTION

[0064] The present application will be described in detail below with reference to the drawings and specific embodiments. It should be noted here that in order to make the embodiments more detailed, the following embodiments are the best, preferred embodiments, and other alternative ways can also be implemented by those skilled in the art for some known technologies; and the drawings are only used to more specifically describe the embodiments, and are not intended to specifically limit the present application.

[0065] It should be noted that in the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiments can include a specific feature, structure or property, but not necessarily every embodiment includes the specific feature, structure or property. In addition, when a specific feature, structure or property is described in combination with an embodiment, it should be within the knowledge of those skilled in the related art to implement such a feature, structure or property in combination with other embodiments (whether or not explicitly described).

[0066] Generally, the terms can be understood at least in part from the context of their use. For example, depending on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular or can be used to describe combinations of features, structures, or characteristics, in the plural, without necessarily dictating whether more than one feature, structure, or characteristic is present. In addition, the term "based on" can be understood as not necessarily of exclusive alternatives, but rather, as allowing for additional implicit factors, whether or not explicitly mentioned.

[0067] As shown in Figure 1 A meter calibration cycle determination method, comprising the following steps:

[0068] S1: Real-time acquisition of instrument working waveform data stream, extraction of characteristic waveform segment through edge computing node;

[0069] S2: Input the characteristic waveform segment into the pre-trained phase state recognition model, and output the current working phase state label;

[0070] S3: Based on the phase state label, activate the corresponding degradation rate calculator to generate real-time degradation increment;

[0071] S4: According to the cumulative degradation increment and the phase state transition path, obtain the next calibration cycle.

[0072] S1 specifically includes:

[0073] S11: Deploy multi-channel sampling at the instrument signal output end to synchronously collect voltage waveform, current waveform and timing pulse signal at a sampling rate of no less than 10 kHz, forming raw waveform data;

[0074] S12: Perform sliding window interception processing on the raw waveform data, with a window length of 500 ms and an overlap rate of no less than 30% between adjacent windows, which ensures complete extraction of transient characteristics and avoids loss of window boundary information; the timing pulse signal is used for time alignment or segment positioning of the waveform data, and the multi-channel signals are synchronized to ensure the timing consistency of the feature analysis;

[0075] S13: Perform dynamic baseline calibration processing on the raw waveform data in each window to extract the following features:

[0076] Waveform distortion degree D H (Calculate the Hausdorff distance between the actual waveform and the standard waveform):

[0077] Where W is the waveform point set in the current window, W std is the standard waveform template, D H measures the maximum local difference between the actual waveform and the ideal waveform, x and y represent any waveform points in W and W std , respectively, ||x-y|| represents the Euclidean distance between points x and y, which is used to measure the spatial distance between two waveform sampling points, sup represents the upper bound, which is the "largest lower bound" among all values, used to find the maximum value among all the nearest distances from a point to all points in another set, and inf represents the lower bound, which is the "smallest upper bound" among all values, used to find the nearest distance from a certain point to all points in another set.

[0078] Pulse density p: Where N pulse represents the number of pulses with an amplitude exceeding the pulse threshold value q per unit time, T is the time window length, and the pulse threshold value q is dynamically updated based on the average amplitude of the previous 10 windows:

[0079] q = m 10 + k * s 10 , where m 10 , s 10 are the average amplitude and standard deviation of the previous 10 windows, respectively, and k is an empirical adjustment factor, k = 2.5.

[0080] Energy entropy H: Where p i represents the energy proportion of the i-th wavelet packet frequency band, E j is the energy value of the i-th frequency band, and the wavelet packet basis function is used to improve the energy resolution of the 1-5 kHz frequency band.j is the energy value of the jth frequency band, n is the number of frequency bands after wavelet packet decomposition;

[0081] S14: encapsulate the features extracted from each window as a three-dimensional feature vector f = [D H , p, H], channelize the three-dimensional feature vector of each window to form a pseudo-image structure input tensor F in , and input it to the lightweight convolution scoring network (MobileNetV3 based on channel pruning) of the edge computing node to calculate the segment importance score value s. Only high-value feature waveform segments with a score s > t are retained, where t is an empirically set score threshold, and the default value is 0.7.

[0082] Feature vector expansion processing: since MobileNetV3 is essentially a convolutional neural network (CNN), it requires input with a tensor structure. First, channelize the three-dimensional feature vector f of each window to form a pseudo-image structure input tensor: where the height is 1, the width is 1, and the number of channels is 3 (corresponding to the three feature dimensions);

[0083] Lightweight scoring network structure design (based on MobileNetV3)

[0084] The MobileNetV3Small architecture after pruning (channel pruning) is used, and only the first 2-3 Bottleneck modules are retained, while the redundant deep classification head is disabled to control the parameter size. The model structure is as follows:

[0085] Input: 1x1x3;

[0086] → ConvBN (1x1, expand = 8);

[0087] → BottleneckBlock x 2 (depth can be pruned);

[0088] → GlobalAveragePooling;

[0089] → 1x1Conv (output single value);

[0090] → Sigmoid activation;

[0091] → Output score value s ∈ (0, 1).

[0092] The segment importance score value is defined as: where represents the MobileNetV3 feature transformation after channel pruning, GAP(·) represents global average pooling, W1, b are linear transformation parameters, and σ(·) represents the normalized score value.

[0093] S2 specifically comprises:

[0094] S21: performing timing alignment and feature dimension reduction on the feature waveform segment to generate a standardized space-time feature matrix; wherein T is the length of the time window, d is the feature dimension after extraction per window (including 3 dimensions or the principal component dimension after compression), and M is the standardized space-time feature matrix, which is input into the phase recognition model;

[0095] S22: inputting the space-time feature matrix into the double-flow space-time attention network of the phase recognition model to extract waveform morphological features and pulse time sequence correlation features respectively;

[0096] The standardized feature matrix M is simultaneously input into two sub-paths in the double-flow structure:

[0097] 1. Morphological flow: local and global structural features of waveform morphology are extracted using dilated convolution, focusing on the evolution trend of distortion degree and energy entropy: F 形态 = DilatedConv(M; r), where r is the dilation rate for capturing a larger receptive field, and DilatedConv represents dilated convolution.

[0098] Specifically: receiving the space-time feature matrix after standardization as input, first using multiple one-dimensional convolution structures with different receptive fields to extract local and global morphological change features, the convolution structure adopts an expansion method to construct multi-scale parallel paths, then the outputs of each path are spliced and fused in the feature dimension to form a unified morphological feature representation, the fusion result is processed through normalization and nonlinear activation operation to improve its feature distribution stability and nonlinear expression ability, and finally the output morphological feature tensor carries the distortion trend and energy change features in the waveform structure, which are used for subsequent fusion processing;

[0099] 2. Time sequence flow: a causal self-attention mechanism is used to model the long-range dependence relationship between pulse density and its time interval: F 时序 = CausalAttention(M), where CausalAttention represents causal sub-attention;

[0100] Specifically: also taking the standardized spatiotemporal feature matrix as input, first, the position coding information in the time dimension is introduced to enhance the model's perception ability of the position of the feature sequence, then, by constructing an attention mask with causal constraints, it is limited that each position in the sequence can only model the correlation based on the information at the current and previous time, on this basis, the long-range temporal dependence relationship in the feature sequence is extracted using the self-attention mechanism, focusing on capturing the temporal patterns reflected by the pulse density change and interval jitter, the attention output result is residual fused with the original input, and then normalized to form a stable temporal feature expression tensor for subsequent feature fusion.

[0101] S23: Fuse the weighted waveform morphology feature and the pulse temporal correlation feature through the cross-modal feature fusion gating mechanism, calculate the fusion weight and generate a hybrid feature vector F 混合 , comprising:

[0102] Fusion weight calculation: where [‖] represents the feature concatenation operation, W2 represents the gating linear transformation matrix, represents the Sigmoid activation function, w ∈ [0, 1] represents the dynamic fusion weight vector;

[0103] Hybrid feature vector generation: F 混合 = w·F 形态 + (1-w)·F 时序 ;

[0104] S24: Based on the hybrid feature vector, match the predefined phase state prototype cluster, and output the current working phase state label using the dynamic radius clustering algorithm.

[0105] S24 includes inputting the hybrid feature vector F 混合 to clustering, matching the existing working phase state prototype cluster, and using the dynamic radius clustering algorithm to determine the current cluster label:

[0106] Each phase state cluster C i includes a prototype center μ i and a dynamic radius R i , wherein: where R b,i is the basic radius of cluster i, which is usually the sample standard deviation, is an environment adaptive coefficient, defined as: S 噪声 is the currently measured environmental noise intensity, S 基线 is the noise intensity in the reference environment;

[0107] Matching logic:

[0108] If F 混合 - μ i < Ri , then match the working phase state cluster C i , that is, if the mixed feature vector falls within the dynamic radius of a certain cluster, directly output the corresponding working phase state label;

[0109] If it falls into the overlapping area of multiple clusters, select the working phase state cluster with the largest correlation with the current degradation increment.

[0110] S3 specifically includes:

[0111] S31: According to the mapping of the working phase state label to the predefined calculator configuration table, load the multi-dimensional degradation model of the corresponding phase state;

[0112] S32: Real-time acquisition of the current environmental parameter set of the instrument, input the degradation model to calculate the basic degradation rate;

[0113] S33: Based on the deviation degree of the historical degradation data of the same phase state and the current basic degradation rate, generate a dynamic compensation factor;

[0114] S34: Fuse the basic degradation rate, the dynamic compensation factor and the cumulative running time to output the real-time degradation increment.

[0115] Environmental parameter injection and basic degradation rate calculation: Real-time acquisition of the parameter set of the current environment of the instrument, and input it as input to the loaded degradation model to calculate the basic degradation rate. The degradation model is used to calculate the basic degradation rate, which is represented as:

[0116] Where λ 当前 represents the basic degradation rate under the current environment, λ 基准 represents the reference degradation rate under the calibration environment, ΔT is the temperature difference between the current temperature and the reference temperature, RH 当前 represents the current relative humidity, RH 标定 is the reference humidity at the time of model calibration, ω is the material property coefficient, which represents the temperature-stress response sensitivity. The environmental parameters can include vibration spectrum, dust concentration;

[0117] To improve the robustness to individual differences of the instrument, environmental drift and measurement errors, the degradation model also introduces a deviation compensation mechanism based on historical degradation data to obtain a dynamic compensation factor η.

[0118] The deviation compensation mechanism includes:

[0119] Extract the historical degradation rate sample set matching the current phase state from the cloud database, and calculate the statistical parameters: mean μ, covariance matrix ∑;

[0120] Based on Mahalanobis distance to evaluate the deviation degree of the current rate:

[0121]

[0122] Generate dynamic compensation factor η:

[0123] Where μ is the historical average of the same rate, ∑ is the historical sample covariance matrix, D M is the Mahalanobis distance, which is used to measure the relative deviation of the current sample from the historical distribution, is the historical sample standard deviation, which is obtained by extracting the main diagonal element from ∑.

[0124] Synthesis and output of real-time degradation increment: finally, by fusing the basic degradation rate, dynamic compensation factor and phase running time, the real-time degradation increment value at the current time is generated:

[0125]

[0126] Where ΔQ is the cumulative degradation increment, t0, t are the start and end times of degradation calculation, respectively, and dt represents the small time increment when integrating on the continuous time domain, which is used to time cumulative real-time degradation rate on the interval [t0, t].

[0127] The system outputs the latest increment value every 5 seconds, and if the growth rate of ΔQ in unit time exceeds the preset threshold (such as 10%), it is marked as abnormal fluctuation and reported to the upper layer scheduling system.

[0128] S4 specifically includes:

[0129] S41: Calculate the instrument remaining life prediction value T 剩余 based on the cumulative degradation increment, the formula is:

[0130] Where ΔQ 平均 is the average degradation increment per unit time under the current phase, Q 累积 is the total amount of degradation accumulated so far, which aggregates each phase ΔQ output before and calculates the comprehensive Q 累积 according to time weighting, Q 阈值 is the preset degradation failure threshold, which dynamically loads Q 阈值 according to the instrument type, Q 阈值 ∈ [0.85, 0.95],

[0131] Pressure transmitter: 0.95;

[0132] Flowmeter: 0.87;

[0133] Electric energy meter: 0.90;

[0134] Intelligent water meter: 0.85;

[0135] Relay protection device: 0.92;

[0136] S42: Analyze the phase transition path map and extract the following risk parameters:

[0137] High-frequency transition phase pair: count the number of high-frequency transition paths f with adjacent phase transition frequency > 2 times / hour 高频 ;

[0138] Critical degradation node: identify phase jump events that cause ΔQ 平均 sudden increase > 50%;

[0139] S43: Construct a dual-domain adjustment coefficient K;

[0140] S44: Update the period using a sliding window mechanism: T 新 = T 基准 · e -K·γ , where T 新 is the updated calibration period currently calculated, and γ is the environmental acceleration factor, which is dynamically corrected according to real-time data from temperature and humidity sensors, and the calculation method is:

[0141] where T 实际 is the current temperature, T 标定 is the calibration reference temperature, and RH is the current relative humidity.

[0142] The final value of the period is limited to: T 新 ∈ [0.3·T 基准 , 1.5·T 基准 ]; Ensure that the period is appropriately contracted under high-frequency state fluctuations and adverse environmental conditions; The period can be extended elastically under stable operation and good environment.

[0143] The dual-domain adjustment coefficient K is calculated as: where α and β are weight coefficients, α = 0.6 and β = 0.4, F 高频 is the number of high-frequency transition paths, F 基准 is the historical average transition path number, and T 标准 is the standard calibration period corresponding to the device type.

[0144] As shown in FIG. Figure 2 , an instrument calibration period determination system for implementing the above method includes the following modules:

[0145] A data acquisition module for real-time acquisition of instrument waveform data stream and transmission to an edge computing node;

[0146] A feature extraction module for processing waveform data stream in the edge computing node and extracting feature waveform segments;

[0147] A phase state recognition module is configured to input the feature waveform segment into a pre-trained phase state recognition model and output a current working phase state label;

[0148] A degradation evaluation module is configured to call a corresponding degradation rate calculator based on the phase state label and generate a real-time degradation increment;

[0149] A cycle decision module is configured to output a corresponding next calibration cycle according to the accumulated degradation increment and the phase state transition path.

[0150] The present application encompasses any alternative, modification, equivalent method and scheme made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details for those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits are not described in detail.

[0151] The above description is only the preferred embodiment of the present application, and it should be pointed out that for those skilled in the art, without departing from the principle of the present application, a number of improvements and refinements can also be made, which should be regarded as the protection scope of the present application.

Claims

1. A method of determining a meter proving period, the method comprising: The method comprises the following steps: S1: real-time acquisition of instrument working waveform data stream, extraction of characteristic waveform segments by edge computing nodes, extraction of waveform distortion, pulse density and energy entropy from each characteristic waveform segment, and encapsulation into a three-dimensional feature vector; S2: input of the characteristic waveform segments into a pre-trained phase state recognition model, generation of a mixed feature vector by the phase state recognition model using a dual-flow spatiotemporal attention network, and output of a current working phase state label by a dynamic radius clustering algorithm; S3: activation of a corresponding degradation rate calculator based on the phase state label, generation of a real-time degradation increment, loading of a multi-dimensional degradation model corresponding to the phase state by the degradation rate calculator, and calculation of a basic degradation rate by the degradation model, which is represented as: ; wherein, represents a base degradation rate under the current environment, represents a reference degradation rate under a calibration environment, is a temperature difference between the current temperature and a reference temperature, represents a current relative humidity, is a reference humidity at the time of model calibration, is a material property coefficient representing temperature-stress response sensitivity; The deterioration model also introduces a deviation compensation mechanism based on historical deterioration data to obtain a dynamic compensation factor ; Synthesis and output of the real-time degradation increment: finally, the basic degradation rate, the dynamic compensation factor and the phase state running time are fused to generate a real-time degradation increment value at the current time: ； wherein, is an accumulated degradation increment, are start and end times of degradation calculation, respectively, denotes a small time increment when performing an integral operation in a continuous time domain, and is used for time accumulation of a real-time degradation rate in an interval ; S4: according to the cumulative degradation increment and the phase state transition path, the next calibration period is obtained, and S4 specifically comprises: S41: calculating the residual life prediction value of the instrument based on the accumulated degradation increment The formula is: wherein, is the mean increment of degradation per unit time in the current phase, is the total amount of degradation that has been accumulated so far, is a preset degradation failure threshold value; S42: analysis of a phase state transition path map to extract the following risk parameters: High frequency transition phase pairs: number of high frequency transition paths with the frequency of adjacent phase transitions > 2 times / hour ; Key degradation nodes: Identify phase jump events that cause a sudden increase of > 50% S43: Constructing the dual-domain adjustment coefficient ; S44: update the period using a sliding window mechanism: wherein, is the current calculated updated calibration period, is an environmental acceleration factor, dynamically corrected according to real-time data from the temperature and humidity sensors; The double-domain adjustment coefficient is calculated as: wherein, is a weight coefficient, , is the number of high-frequency transfer paths, is the historical average number of transfer paths, is the standard calibration period corresponding to the device type.

2. A method of determining a meter proving period according to claim 1, characterized in that, The S1 specifically comprises: S11: deployment of multi-channel sampling at an instrument signal output end, synchronous acquisition of voltage waveform, current waveform and timing pulse signals, and formation of original waveform data; S12: sliding window interception processing of the original waveform data; S13: dynamic baseline calibration processing of the original waveform data in each window to extract the following features: waveform distortion : wherein, is a set of waveform points within a current window, is a standard waveform template, measures the maximum local difference between the actual waveform and the ideal waveform, respectively represent and any waveform point in represents the Euclidean distance between points and measures the spatial distance of two waveform sampling points, represents the supremum, i.e. the "greatest lower bound" among all values, for finding the maximum value among the nearest distances from one point to all points in another set, represents the infimum, i.e. the "least upper bound" among all values, for finding the nearest distance of a certain point to all points in another set; Pulse density : ,in, This indicates that the amplitude exceeds the pulse threshold per unit time. The number of pulses, The time window length and the pulse threshold are given. Dynamically updated based on the average amplitude of the first 10 windows; Energy entropy : , wherein, represents the energy proportion of the th wavelet packet frequency band, is the energy value of the th frequency band, is the energy value of the th frequency band, is the number of frequency bands after wavelet packet decomposition; S14: Encapsulate the features extracted from each window into a three-dimensional feature vector. The three-dimensional feature vectors of each window are channelized and embedded to form a pseudo-image structure input tensor. The lightweight convolutional scoring network, fed into the edge computing nodes, calculates the importance score of the fragment. Only ratings are retained. Characteristic waveform segments, A rating threshold set for experience.

3. A method of determining a meter proving period according to claim 2, wherein, The segment importance score value is defined as: wherein, denotes the MobileNetV3 feature transformation after channel pruning, denotes the global average pooling, is a linear transformation parameter, denotes the normalized score value.

4. The method of claim 1, wherein, The S2 specifically comprises: S21: timing alignment and feature dimension reduction of the characteristic waveform segments to generate a standardized spatiotemporal feature matrix; S22: input of the spatiotemporal feature matrix into a dual-flow spatiotemporal attention network of the phase state recognition model to extract waveform morphology features and pulse timing correlation features, respectively; S23: Fuse the weighted waveform morphology features and the pulse timing correlation features through a cross-modal feature fusion gating mechanism, calculate the fusion weight, and generate a hybrid feature vector ; S24: matching of the mixed feature vector to a pre-defined phase state prototype cluster to output a current working phase state label by a dynamic radius clustering algorithm.

5. A method of determining a meter proving period according to claim 4, wherein, The S24 includes inputting the mixed feature vector to clustering, matching an existing work phase prototype cluster, and determining a current cluster label using a dynamic radius clustering algorithm. Each phase cluster Including a prototype center With a dynamic radius Wherein: Wherein, The basis radius of the cluster The basis radius, The environment adaptive coefficient is defined as: , The current measured environmental noise intensity, The noise intensity in the reference environment; Matching logic: If , then the matching working phase cluster , that is, if the mixed feature vector falls within the dynamic radius of a certain cluster, directly output the corresponding working phase label; If it falls into the overlapping area of multiple clusters, the working phase state cluster with the largest correlation to the current degradation increment is selected.

6. The method of claim 1, wherein, The S3 specifically comprises: S31: mapping of the working phase state label to a pre-defined calculator configuration table to load a multi-dimensional degradation model corresponding to the phase state; S32: real-time acquisition of instrument current environment parameter sets to calculate a basic degradation rate in the degradation model; S33: generation of a dynamic compensation factor based on the deviation degree of historical same-phase degradation data and the current basic degradation rate; S34: fusion of the basic degradation rate, the dynamic compensation factor and the cumulative running time to output a real-time degradation increment.

7. A system for implementing a method of determining a meter calibration period as claimed in any one of claims 1 to 6, characterised in that, The method comprises the following modules: A data acquisition module for real-time acquisition of instrument working waveform data stream and transmission thereof to an edge computing node; A feature extraction module for processing of the waveform data stream in the edge computing node to extract characteristic waveform segments; A phase state recognition module for input of the characteristic waveform segments into a pre-trained phase state recognition model to output a current working phase state label; A degradation evaluation module for calling of a corresponding degradation rate calculator based on the phase state label to generate a real-time degradation increment; A period decision module for output of a corresponding next calibration period according to the cumulative degradation increment and the phase state transition path.

Images