A smart meter upgrade method and system

Abstract

This invention provides a method and system for upgrading smart meters, relating to the field of data processing technology. The method includes: acquiring historical and real-time operating data of the smart meter; calculating the operational capability impact coefficient using a kernel principal component analysis algorithm; predicting the operational capability degradation trajectory of the smart meter in a future period; calculating the operational capability degradation coefficient of the smart meter in the future period and determining whether the operational capability degradation coefficient is greater than a preset operational capability degradation coefficient; if so, determining that the smart meter needs to be upgraded and identifying the target modules to be upgraded in the smart meter; performing reverse reasoning on each target module to obtain the posterior failure probability of each target module; calculating the upgrade priority weight of each target module based on the posterior failure probability, and determining the upgrade path of each target module according to the upgrade priority weight; executing the upgrade path to complete the upgrade of the smart meter; otherwise, determining that the smart meter does not need to be upgraded.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for upgrading smart meters. Background Technology

[0002] With the continuous development of smart grids and electricity information collection systems, smart meters, as important metering and data acquisition terminals in power systems, are widely used in various electricity consumption scenarios. During long-term operation, the internal metering, communication, and control modules of smart meters are susceptible to load fluctuations, environmental changes, and component aging, leading to a gradual degradation in performance and affecting metering accuracy and communication stability.

[0003] In existing technologies, traditional smart meter upgrade methods usually rely on fixed-cycle upgrades or upgrades triggered by a single fault alarm. These methods are difficult to accurately reflect the changing trends of the meter's operating status and potential degradation risks. They often have problems with unreasonable upgrade timing, which can easily lead to upgrade delays or frequent repeated upgrades, affecting equipment operating efficiency and maintenance cost control.

[0004] Furthermore, traditional smart meter upgrade methods often treat the smart meter as a whole during the upgrade process, lacking detailed identification and differential analysis of the operating status of each functional module. They fail to fully consider the interrelationships between modules and the impact of fault propagation, making it difficult to scientifically determine the upgrade targets and upgrade sequence. This results in inaccurate upgrade decisions and unreasonable upgrade paths, making it difficult to meet the needs of intelligent and refined operation and maintenance of smart meters. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a smart meter upgrade method and system, which can solve the problems of traditional smart meter upgrade methods that usually rely on fixed-period upgrades or upgrades triggered by a single fault alarm. These methods are difficult to accurately reflect the changing trends and potential degradation risks of the meter's operating status, often resulting in unreasonable upgrade timing, delayed upgrades, or frequent repeated upgrades, affecting equipment operating efficiency and maintenance cost control. Furthermore, the upgrade process often treats the smart meter as a whole, lacking refined identification and differential analysis of the operating status of each functional module, failing to fully consider the correlation between modules and the impact of fault propagation, and making it difficult to scientifically determine the upgrade targets and upgrade sequence. This leads to inaccurate upgrade decisions, unreasonable upgrade paths, and an inability to meet the technical problems of intelligent and refined operation and maintenance of smart meters.

[0006] A first aspect of this invention provides a method for upgrading a smart meter, comprising: S1: Obtain historical and real-time operating data of the smart meter; S2: Based on historical and real-time operating data, the kernel principal component analysis algorithm is used to calculate the impact coefficient of the smart meter's operating capacity at the current moment; S3: Based on the operational capability impact coefficient, predict the trajectory of smart meter operational capability degradation in the future period; S4: Based on the operational capability degradation trajectory, calculate the operational capability degradation coefficient of the smart meter in the future period, and determine whether the operational capability degradation coefficient is greater than the preset operational capability degradation coefficient; if so, determine that the smart meter needs to be upgraded and proceed to S5; otherwise, determine that the smart meter does not need to be upgraded. S5: Through the smart meter upgrade identification model, identify multiple target modules in the smart meter that need to be upgraded; S6: Perform reverse reasoning on each target module to obtain the posterior failure probability of each target module; S7: Calculate the upgrade priority weight of each target module based on the posterior failure probability, and determine the upgrade path of each target module according to the upgrade priority weight. S8: Execute the upgrade path to complete the upgrade of the smart meter.

[0007] A second aspect of this invention provides a smart meter upgrade system, comprising: a processor and a memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the smart meter upgrade method as described in the first aspect.

[0008] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: In this embodiment of the invention, by calculating the operational capability degradation coefficient of the smart meter in the future time period based on the operational capability degradation trajectory, it is possible to determine whether the smart meter needs to be upgraded. This eliminates reliance on fixed-cycle upgrades or upgrades triggered by a single fault alarm, accurately reflecting the changing trend of the meter's operating status and potential degradation risks. It avoids the problem of unreasonable upgrade timing, reduces the likelihood of delayed or frequent upgrades, improves equipment operating efficiency, and lowers maintenance costs. Through the smart meter upgrade identification model, the target modules to be upgraded in the smart meter are identified. Reverse reasoning is performed on each target module to obtain its posterior fault probability. Based on the posterior fault probability, the upgrade priority weight of each target module is calculated, and the upgrade path for each target module is determined according to the upgrade priority weight. The upgrade process does not treat the smart meter as a whole; it possesses refined identification and differential analysis of the operating status of each functional module, fully considering the correlation between modules and the impact of fault propagation. This allows for the scientific determination of upgrade targets and upgrade order, avoiding inaccurate upgrade decisions and unreasonable upgrade paths, and meeting the needs of intelligent and refined operation and maintenance of smart meters. Attached Figure Description

[0009] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0010] Figure 1 This is a flowchart illustrating a smart meter upgrade method provided in an embodiment of the present invention.

[0011] Figure 2 This is a schematic diagram of the structure of a smart meter upgrade system provided in an embodiment of the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0013] The smart meter upgrade method provided by the present invention will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0014] Reference manual attached Figure 1 The diagram shows a flowchart of a smart meter upgrade method provided by an embodiment of the present invention.

[0015] This invention provides a method for upgrading smart meters, which may include the following steps: S1: Obtain historical and real-time operating data of the smart meter.

[0016] The operational data includes core metering and electrical performance data, meter internal status and self-test data, communication and interaction performance data, environmental and operating condition data, and time and load mode data. Specifically, the core metering and electrical performance data includes voltage, current, power, energy, power factor, frequency, and metering error data; the meter internal status and self-test data includes module operating status, internal temperature, event records, and internal error logs; the communication and interaction performance data includes communication success rate, communication signal strength, communication latency, and online / offline records; the environmental and operating condition data includes external ambient temperature, external ambient humidity, geographical location information, and installation location attributes; and the time and load mode data includes timestamps corresponding to each data point, load curve data, and calendar information.

[0017] In one possible implementation, after S1 and before S2, steps S1A and S1B are also included: S1A: Preprocessing of historical operational data, including data cleaning, data alignment, data noise reduction, data labeling, and data normalization.

[0018] S1B: Based on the preprocessed historical operating data, determine the core operating status features used to describe the operating status of the smart meter.

[0019] In one possible implementation, S1B specifically includes sub-steps S1B1 to S1B6: S1B1: Extract the basic operational features from the preprocessed historical operational data, generate derived operational features based on the basic operational features, combine the basic operational features and derived operational features, and construct a candidate core operational status feature pool.

[0020] It should be noted that the preprocessed historical operating data is categorized and organized according to data type and time order, and basic operating features that can directly characterize the smart meter's operating status are extracted based on a preset statistical window. These basic operating features include: for core metering and electrical performance data, the following are extracted: average voltage, average current, average power, energy increment, frequency deviation, power factor fluctuation, and metering error statistics. For internal meter status and self-test data, the following are extracted: module online status percentage, average internal temperature, number of error log occurrences, event record trigger frequency, and abnormal duration. For communication and interaction performance data, the following are extracted: communication success rate, average communication latency, average signal strength, number of offline occurrences, and online rate. For environmental and operating condition data, the following are extracted: average environmental temperature and humidity, temperature and humidity fluctuation range, geographic area code, and installation location category. Finally, for time and load pattern data, the following are extracted: time period identifier, weekday or holiday identifier, average load, peak load, valley load, and peak-valley difference. Derived operating features are generated based on the combination relationships, temporal variation relationships, and state coupling relationships among features. These derived operating features include voltage change rate, current fluctuation coefficient, power mutation rate, metering error drift rate, internal temperature rise rate, error event growth rate, communication delay jitter coefficient, signal strength attenuation rate, load slope, load periodic fluctuation coefficient, and coupling offset index between electrical performance and environmental conditions. Subsequently, the basic operating features and derived operating features are combined and summarized to form a candidate core operating state feature pool.

[0021] S1B2: Perform redundancy filtering and saliency filtering on the candidate core operating state feature pool in sequence to obtain the filtered candidate core operating state feature pool.

[0022] Specifically, the redundancy filtering of each operational feature in the candidate core operational status feature pool is performed first to reduce the impact of highly correlated, repetitive, or information-overlapping operational features on subsequent modeling. In practice, the redundancy between features can be assessed based on the value sequences of each operational feature in historical samples, combined with correlation indices, variance inflation factors, mutual information values, or distance correlation coefficients. When the correlation exceeds a preset redundancy threshold, operational features with clearer physical meaning, higher data integrity, or better fluctuation stability are retained, while other redundant operational features are eliminated, resulting in a pre-screened feature set. Based on this, the pre-screened feature set undergoes significance filtering to select operational features with strong discriminative and interpretative capabilities compared to the smart meter operational status labels or status evaluation results. In practice, significance indices can be calculated using analysis of variance, chi-square test, mutual information analysis, F-test, correlation test, or evaluation methods based on single-feature classification performance. Operational features with significance indices below a preset significance threshold are eliminated, ultimately resulting in the filtered candidate core operational status feature pool.

[0023] S1B3: Calculate the entropy weight of each running feature in the filtered candidate core running state feature pool using the entropy weight method.

[0024] Specifically, after obtaining the filtered candidate core operational status feature pool, the values ​​of each operational feature in historical samples are first normalized to eliminate the influence of dimensional differences between different features, and the relative distribution of each operational feature in different samples is obtained. Based on this, the information entropy of each operational feature is calculated according to the dispersion of its distribution in all historical samples, which is used to characterize the information content and discriminative ability of the corresponding operational feature. Specifically, the more uniform the distribution of an operational feature, the higher its information entropy and the weaker its discriminative ability; conversely, the more significant the distribution differences, the lower its information entropy and the stronger its discriminative ability. Subsequently, the degree of difference is calculated based on the information entropy of each operational feature, and the degree of difference is normalized to obtain the entropy weight of each operational feature.

[0025] S1B4: Using various operational features as input features and the smart meter operational status label as output, a random forest model is trained, and the feature importance score of each operational feature is calculated using the Gini index decrease. in, This represents the feature importance score of the j-th running feature. This represents the total number of decision trees in the random forest. This represents the index of a split node in the t-th decision tree that uses the j-th running feature as the splitting feature. Let represent the set of nodes in the t-th decision tree that use the j-th running feature as the splitting feature. In the t-th decision tree, the i-th... Gini index of each node before splitting In the t-th decision tree, the i-th... The number of samples in the left child node after a node splits. In the t-th decision tree, the i-th... The number of samples from nodes before the split. In the t-th decision tree, the i-th... The Gini index of the left child node after a node splits. In the t-th decision tree, the i-th... The number of samples in the right child node after a node splits In the t-th decision tree, the i-th... The Gini index of the right child node after a node splits. This indicates the number of categories of tags representing the operating status of the smart meter. In the t-th decision tree, the i-th... The proportion of class c samples within each node.

[0026] S1B5: Normalize the entropy weight method weights and feature importance scores, and then perform weighted fusion of the normalized entropy weight method weights and feature importance scores to obtain the final importance score of each running feature.

[0027] S1B6: Based on the final importance score, sort the various operating features in the filtered candidate core operating status feature pool in descending order, and determine the top-ranked preset number of operating features as the core operating status features used to describe the operating status of the smart meter.

[0028] In this embodiment of the invention, by first constructing a candidate core operating state feature pool containing basic operating features and derived operating features, and then sequentially performing redundancy filtering, saliency filtering, entropy weight analysis, and feature importance evaluation based on random forest, and normalizing and fusing the multi-source evaluation results, it is possible to screen out the core operating state features that are most representative, have stronger distinguishing ability, and are more stable for the smart meter operating state from a large amount of historical operating data. This avoids redundant features, weakly correlated features, and noisy features from interfering with subsequent modeling, improves the comprehensiveness, objectivity, and accuracy of feature selection, and provides a more reliable data foundation for subsequent calculation of operating capability influence coefficients, degradation trajectory prediction, and upgrade decisions, thereby improving the accuracy of smart meter operating state representation and the credibility of overall upgrade judgment.

[0029] S2: Based on historical and real-time operating data, the kernel principal component analysis algorithm is used to calculate the impact coefficient of the smart meter's operating capacity at the current moment.

[0030] In one possible implementation, S2 specifically includes sub-steps S201 to S209: S201: Perform correlation analysis on the preprocessed historical operating data and core operating status characteristics, select target operating data that affect the core operating status characteristics from the preprocessed historical operating data, and construct an operating data input matrix.

[0031] Specifically, a correlation analysis is first performed on the preprocessed historical operational data and core operational status characteristics to identify operational data dimensions that have a significant impact on these characteristics. The correlation analysis comprehensively assesses the linear and nonlinear correlation strength by calculating the Pearson correlation coefficient, Spearman rank correlation coefficient, or maximum information coefficient between each operational data dimension and each core operational status characteristic. Operational data dimensions whose absolute correlation values ​​exceed a preset threshold are identified as target operational data. Subsequently, the target operational data selected at each historical time point are organized into a two-dimensional matrix in chronological order, where each row represents a data sample at a historical time point, and each column represents an operational data dimension, thus constructing the operational data input matrix.

[0032] S202: Standardize the running data input matrix to obtain a standardized running data input matrix, and map each data sample in the standardized running data input matrix to a high-dimensional feature space.

[0033] Specifically, the operational data input matrix is ​​standardized to eliminate dimensional differences between different operational data dimensions. The Z-score standardization method is preferred, transforming each operational data dimension into standardized data with a mean of 0 and a standard deviation of 1, thus obtaining a standardized operational data input matrix. Based on this, a pre-defined nonlinear mapping function is used to map each data sample in the standardized operational data input matrix to a high-dimensional feature space to improve data separability.

[0034] S203: Calculate the covariance matrix of the standardized running data input matrix mapped to the high-dimensional feature space, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors.

[0035] Specifically, after completing the high-dimensional mapping, the covariance matrix of each high-dimensional data sample in the high-dimensional feature space is calculated, and the covariance matrix is ​​decomposed into eigenvalues ​​and eigenvectors. The eigenvalues ​​characterize the variance of the data in the corresponding direction, and the eigenvectors represent the principal component directions in the high-dimensional feature space.

[0036] S204: Represent the feature vector as a linear combination of each data sample in a high-dimensional feature space, and obtain the sample coefficient representation corresponding to the feature vector, so as to transform the feature decomposition problem in the high-dimensional feature space into a sample coefficient solving problem.

[0037] Specifically, since the high-dimensional feature space has a high dimension, directly solving the feature vector is computationally complex. Therefore, the feature vector is represented as a linear combination of each high-dimensional data sample, and the weight of each sample in the feature vector is represented by the sample coefficients, thereby transforming the feature vector solving problem into the sample coefficient solving problem.

[0038] S205: Based on the sample coefficient representation and the inner product relationship of each data sample in the high-dimensional feature space, construct the kernel matrix, and transform the eigenvalue decomposition process corresponding to the covariance matrix into the feature solving process on the kernel matrix.

[0039] Specifically, a kernel matrix is ​​constructed based on the sample coefficient representation and the inner product relationship of samples in the high-dimensional feature space. Each element of the kernel matrix is ​​calculated from the corresponding sample using a kernel function, which is equivalent to the inner product of the samples in the high-dimensional feature space. Based on the kernel matrix, the eigenvalue decomposition problem of the covariance matrix is ​​transformed into the eigenvalue solving problem on the kernel matrix, thereby avoiding explicit computation of the high-dimensional mapping.

[0040] S206: Based on the eigenvalues ​​of the kernel matrix, extract multiple uncorrelated kernel principal components and calculate the variance contribution rate of each kernel principal component.

[0041] Specifically, the eigenvalues ​​of the kernel matrix are calculated and sorted from largest to smallest. Each eigenvalue corresponds to a kernel principal component, and its magnitude reflects the variance contribution of the data in that direction. The variance contribution rate of each kernel principal component is further calculated to measure its contribution to the overall data information.

[0042] S207: Calculate the cumulative variance contribution rate based on the contribution rates of each variance, and select the target kernel principal component from the kernel principal components based on the cumulative variance contribution rate.

[0043] Specifically, the cumulative variance contribution rate is calculated based on the variance contribution rate of each kernel principal component, and a preset threshold is set. Kernel principal components are selected sequentially from front to back, and the target kernel principal components are determined when their cumulative variance contribution rate first reaches or exceeds the threshold.

[0044] S208: Take the data sample corresponding to the real-time running data as the new data sample, calculate the projection value of the new data sample on the direction of each target kernel principal component, and obtain the kernel principal element corresponding to the new data sample.

[0045] Specifically, the real-time running data collected at the current moment is standardized and used as new data samples, and its projection value is calculated in the direction of each target kernel principal component, thereby obtaining the kernel principal component corresponding to the new data sample.

[0046] S209: Using the variance contribution rate of each target principal component as a weighting coefficient, the principal components corresponding to the newly added data samples are weighted and fused to obtain the influence coefficient of the smart meter's operating capability at the current moment.

[0047] In this embodiment of the invention, by filtering the correlation between historical operating data and core operating status features, combining kernel principal component analysis (KPC) algorithm to perform nonlinear mapping and dimensionality reduction on multidimensional operating data, and weighted fusion of real-time data based on the variance contribution rate of the kernel principal components, redundant information and noise interference can be effectively eliminated while retaining key operating information. This constructs an operating capability influence coefficient with high expressive power and strong robustness for the current operating status of smart meters. This approach not only improves the ability to characterize complex nonlinear operating features but also avoids the computational complexity problem caused by directly processing high-dimensional data, making the operating status assessment more accurate and stable, and providing a reliable basis for subsequent degradation trend prediction and upgrade decisions.

[0048] S3: Based on the operational capability impact coefficient, predict the trajectory of smart meter operational capability degradation in the future.

[0049] In one possible implementation, S3 specifically includes sub-steps S301 to S306: S301: Construct a time series of core operating status features based on the core operating status features corresponding to historical and real-time operating data.

[0050] S302: By using conventional differencing and seasonal differencing, the core operating state characteristic time series is stabilized to obtain a stable core operating state characteristic time series.

[0051] S303: Based on the time series characteristics of stable core operating status and the operating capacity influence coefficient, a predictive model for the degradation of smart meter operating capacity is constructed, with the operating capacity influence coefficient as an exogenous variable. in, Indicates the result after stabilization. The core operational state characteristics at any given time. Indicates satisfaction The lag operator, express The core operational state characteristics at any given time. express The core operational state characteristics at any given time. Indicates the ordinary difference order. Indicates satisfaction The seasonal lag operator with a delay of s time units. Indicates the order of seasonal difference. This represents a seasonal autoregressive polynomial, where P represents the number of historical periods involved in the modeling. Let p represent the non-seasonal autoregressive polynomial, and p represent the number of historical moments involved in the modeling. Let Q represent the seasonal moving average polynomial, and let Q represent the number of historical periodic disturbance terms involved in the modeling. Let q represent the non-seasonal moving average polynomial, and q represent the number of historical disturbance terms involved in the modeling. express The random disturbance term at time t, Represents the regression coefficients of exogenous variables. express The impact coefficient of operational capability at any given time.

[0052] S304: Construct an adaptive function for the prediction model of smart meter operational capability degradation, and determine the optimal combination of orders for the prediction model of smart meter operational capability degradation by optimizing the algorithm with the goal of minimizing the function value of the adaptive function.

[0053] Specifically, after the initial construction of the smart meter operational capability degradation prediction model, historical sample data is divided into model input data and corresponding real observation results. The prediction model is then fitted and calculated based on the historical sample data to obtain prediction results for each historical moment. The prediction results are compared with the corresponding real observation results to calculate the prediction deviation. The prediction deviations are then statistically summarized to construct an adaptive function characterizing the model's prediction error level. With the minimum value of the adaptive function as the optimization objective, an optimization algorithm iteratively searches for the structural parameters of the prediction model. The model fitting and error calculation process is repeated under different parameter combinations, and the corresponding adaptive function values ​​are compared. When the preset convergence condition is met or the maximum number of iterations is reached, the structural parameter combination that minimizes the adaptive function value is selected as the optimal order combination.

[0054] S305: Based on the optimal order combination, initialize the smart meter operational capability degradation prediction model, and update the model parameters of the smart meter operational capability degradation prediction model online using the least squares method: in, express Parameter update gain at time t, express The covariance matrix of the parameter estimation error at time t. express The model input vector at time t, express Time-varying forgetting factor at any given moment, subscript This indicates the transpose operation. express The model parameter vector at time t, express The model parameter vector at time t, express The covariance matrix of the parameter estimation error at time t. Represents the identity matrix.

[0055] S306: The average value of the historical operating capacity influence coefficient is used as the operating capacity influence coefficient for the next future moment and input into the updated smart meter operating capacity degradation prediction model. By performing backward multi-step rolling prediction, the operating capacity prediction value sequence for multiple future moments is obtained, and the operating capacity prediction value sequence is determined as the operating capacity degradation trajectory of the smart meter in the future period.

[0056] Specifically, after updating the parameters of the smart meter operational capability degradation prediction model, its statistical characteristics are calculated based on the historical operational capability influence coefficient sequence. The average value within a preset time window is used as the estimated value of the operational capability influence coefficient for the next future time, thus obtaining exogenous input data for future time. The average value is used to characterize the overall level of the current operating environment and conditions, thereby reflecting the continuity of operational capability influencing factors in short-term prediction. The operational capability influence coefficient for the next future time is input into the updated prediction model, and combined with the current operating status characteristics and model parameters, the predicted value of the operational capability for the next future time is calculated. Subsequently, this prediction result is used as the input for the next prediction, and the prediction is progressively pushed forward in chronological order, performing multi-step rolling prediction to obtain a sequence of predicted operational capability values ​​for multiple consecutive future time periods. The operational capability prediction value sequence is combined in chronological order to form an operational capability evolution sequence, which is then determined as the operational capability degradation trajectory of the smart meter in future time periods.

[0057] In this embodiment of the invention, by constructing a time series of core operating state features and performing differential stabilization processing, combined with a degradation prediction model using the operating capability influence coefficient as an exogenous variable, dynamic modeling of the evolution process of smart meter operating capability is achieved. Furthermore, by optimizing the model order through an adaptive function and employing a least-squares online update mechanism, the model's fitting accuracy and time-varying adaptability to actual operating data are improved. Simultaneously, a rolling prediction method is used to obtain continuous future time-based operating capability prediction results. This allows for a more accurate depiction of the degradation trend and trajectory of smart meters in future periods, enhancing the ability to detect potential performance degradation in advance. This provides a forward-looking and continuous predictive basis for subsequent upgrade decisions, improving the overall stability and reliability of the prediction.

[0058] S4: Based on the operational capability degradation trajectory, calculate the operational capability degradation coefficient of the smart meter in the future period, and determine whether the operational capability degradation coefficient is greater than the preset operational capability degradation coefficient. If so, determine that the smart meter needs to be upgraded and proceed to S5. Otherwise, determine that the smart meter does not need to be upgraded.

[0059] Optionally, the calculation method for the operational capability degradation coefficient specifically includes sub-steps S401 to S407: S401: Based on the operational capability degradation trajectory, extract the predicted operational capability value and the first arrival time of the error limit for each core operational status feature at a single future moment.

[0060] S402: Calculate the individual degradation coefficient of each core operating status characteristic based on the predicted operating capacity value.

[0061] Optionally, the calculation method for the individual degradation coefficient is divided into two types: positive drift and negative drift, where the i-th core operating state characteristic is manifested. in, Represents the single-item degradation coefficient of the i-th core operating state characteristic. This represents the predicted operational capability of the i-th core operational status feature at a single future time t0. This represents the minimum value of the i-th core operating state feature within the reference interval. This represents the failure threshold of the i-th core operating state characteristic. The degradation regulation index represents the characteristic of the i-th core operating state. This represents the maximum value of the i-th core operating status feature within the reference interval.

[0062] S403: Calculate the weight coefficients of each core operating status feature based on historical observation data of each core operating status feature.

[0063] Specifically, firstly, the values ​​of each core operational status feature in historical observation data are standardized to eliminate the influence of dimensional differences on weight calculation and to obtain the relative distribution of each core operational status feature in different observation samples. Based on this, the information entropy of each core operational status feature is calculated according to the dispersion of its distribution across all observation samples. This entropy characterizes the information uncertainty and discriminative power of the corresponding core operational status feature. Specifically, core operational status features with more uniform distribution have higher information entropy and weaker discriminative power, while core operational status features with more significant distribution differences have lower information entropy and stronger discriminative power. Subsequently, the degree of difference is calculated based on the information entropy of each core operational status feature, and the degree of difference is normalized to obtain the weight coefficient of each core operational status feature.

[0064] S404: Weighted fusion of each individual degradation coefficient and weight coefficient to obtain the initial operational degradation coefficient at a future single moment.

[0065] S405: Calculate the time degradation correction coefficient based on the first arrival time of the error limit of each core operating state characteristic.

[0066] S406: Using a time degradation correction factor, the initial operational degradation factor is corrected to obtain the operational capability degradation factor for a future single moment: in, This represents the performance degradation coefficient at a single future time t0. This represents the individual degradation coefficient of the i-th core operating state feature at a single future time t0. This represents the time degradation correction factor at a single future time t0. Represents an exponential function. Indicates from the current moment The time span up to a single future moment t0, Indicates the overall failure time. Indicates the number of core operational status features. The weight coefficient represents the i-th core operating state feature. This represents the first arrival time of the error limit for the i-th core operating state characteristic, where min indicates taking the minimum value. This represents the predicted operational capability of the i-th core operational status feature at a future time τ. This indicates the upper limit of the error exceeding the standard for the i-th core operating state feature. This indicates that the error of the i-th core operating state feature exceeds the lower limit of the standard. Indicates a future time period. Indicates the future window length.

[0067] S407: Repeat S401 to S406 to calculate the operational capability degradation coefficient for each future time period, and select the maximum operational capability degradation coefficient as the operational capability degradation coefficient of the smart meter for the future time period.

[0068] In this embodiment of the invention, by refining the operational capability degradation trajectory into individual degradation coefficients for each core operational state feature, and combining them with feature weights calculated based on historical data for weighted fusion, and introducing an error limit first-arrival time to construct a time degradation correction mechanism, the future operational risks of smart meters can be comprehensively quantified from two dimensions: "degree of degradation" and "degradation speed." This avoids misjudgments caused by relying solely on a single predicted value or a single threshold, ensuring that the operational capability degradation coefficient reflects both the degree of collaborative degradation of multiple features and the urgency of potential failures. This improves the accuracy and foresight of upgrade decisions, enabling more scientific, precise, and reliable upgrade triggering decisions.

[0069] S5: Through the smart meter upgrade identification model, identify multiple target modules in the smart meter that need to be upgraded.

[0070] Optionally, the smart meter upgrade identification model is specifically: a smart meter upgrade identification model based on a long short-term memory neural network, wherein the long short-term memory neural network includes a fusion memory gate and an output gate, and the fusion memory gate is formed by fusing the forget gate and the input gate of the original long short-term memory neural network.

[0071] Optionally, the update formula for the fused memory gate is as follows: in, This represents the fusion memory gating signal at time z. This represents the tanh activation function. The intermediate variable representing the fused memory gate at time z. This represents the Sigmoid activation function. This represents the weight matrix of the fused memory gates. This represents the matrix multiplication operator. This indicates a splicing operation. This indicates that the hidden state output at time z-1 is the same as the hidden state output at the previous time. This represents the input data at time z. This represents the cell state at time z-1, which is the cell state at the previous time. This represents the bias vector of the fused memory gate.

[0072] Furthermore, the candidate cell state and cell state update formulas are specifically as follows: in, This represents the candidate cell state at time z, which is the updated candidate cell state. The weight matrix representing the state of the candidate unit. The bias vector representing the state of the candidate unit. This represents the cell state at time z, which is the updated cell state. This represents the cell state at time z-1.

[0073] Furthermore, the update formula for the output gate is specifically as follows: in, This represents the hidden state output at time z. This indicates that the output gate at time z is the updated output gate. This represents the weight matrix of the output gate. This represents the control coefficient (which can be set by those skilled in the art according to actual needs). The size is not limited in this invention; optionally, Set to 0.5). This represents the bias vector of the output gate.

[0074] Specifically, based on the data content corresponding to each module of the smart meter from historical and real-time operational data, module state features that characterize the operational health of each module are extracted. A sequence of module state features reflecting the evolution of module states over time is constructed according to a preset time window. This sequence of module state features is then input into the smart meter upgrade identification model for training. The smart meter upgrade identification model is a time-series prediction model built on a Long Short-Term Memory (LSTM) neural network. The LSM network fuses the original forget gate and input gate through a memory fusion gate to achieve coordinated regulation of historical information retention and new input information updates. It also generates the hidden state output for each time step through an output gate, thereby enhancing the modeling ability of the module state evolution process. After model training, the module state feature sequence within the preset time window before the current moment is input into the upgrade identification model for forward inference to obtain the probability distribution of each module being in a normal state, an abnormally degraded state, or a fault state at the predicted future time. The status of each module is determined based on the probability distribution. When the probability of abnormal degradation or fault status of a module reaches a preset threshold, or when its predicted status is abnormal degradation or fault status, the module is identified as the target module to be upgraded, thereby realizing the identification of modules in the smart meter that need to be upgraded first.

[0075] In this embodiment of the invention, by constructing a smart meter upgrade identification model based on a long short-term memory neural network and introducing a fusion memory gate to coordinate the retention of historical information and the updating of current input information, the model can effectively capture the temporal evolution characteristics and potential change trends of the operating status of each module, thereby improving the predictive ability for abnormal degradation and fault states of modules. Simultaneously, by determining the probability distribution of the future state of modules, the model achieves accurate identification of the target module to be upgraded, avoiding blindly upgrading all modules or misjudging the upgrade target, improving the pertinence and efficiency of upgrade decisions, reducing unnecessary system resource consumption, and enhancing the intelligence and refinement of the smart meter upgrade process.

[0076] S6: Perform reverse reasoning on each target module to obtain the posterior failure probability of each target module.

[0077] In one possible implementation, S6 specifically includes sub-steps S601 to S605: S601: Construct a dynamic Bayesian network model based on the functional relationships between the target modules and the influence of each target module on the overall operating status of the smart meter.

[0078] Specifically, firstly, based on the physical connection relationships, control call relationships, data transmission relationships, and power supply dependencies of each target module in the smart meter, the functional relationships between the target modules are analyzed. On this basis, each target module is used as a module state node in a dynamic Bayesian network, and the overall operating state of the smart meter is used as the system state node. Based on the impact of abnormalities or faults in each target module on the overall operating state, connections are established from the module state nodes to the system state nodes. For target modules with dependencies between preceding and subsequent levels, collaborative work, or fault propagation relationships, directed connections are established between the corresponding module state nodes to represent the conditional dependencies and fault propagation relationships between modules. Furthermore, the network is temporally expanded using a preset time slice length, forming a dynamic Bayesian network model that includes an initial time slice and a state transition time slice.

[0079] S602: Discretize the operating status of each target module, and determine the prior probability of each target module being in an abnormal degradation state or a fault state based on historical operating data and real-time operating data.

[0080] Specifically, firstly, based on the operating parameters, anomaly records, communication status, temperature rise status, self-test results, and event logs corresponding to each target module in historical and real-time operating data, state discrimination indicators that can characterize the health status of the modules are extracted. Then, according to preset state division rules, the continuous operating states of each target module are divided into normal state, abnormal degradation state, and fault state, and the discretized state results are mapped to corresponding state values. Based on this, the frequency, duration, and state transition of each target module in the historical operating data are statistically analyzed, and the current state is corrected in conjunction with real-time operating data, thereby obtaining the prior probability of each target module being in an abnormal degradation state or a fault state.

[0081] S603: Calculate the influence weight of each target module on the overall operating status of the smart meter, and construct a conditional probability relationship model based on the influence weight: in, This represents the probability that the overall operating state of the smart meter is in an abnormal degradation state or a fault state, given that the states of each target module are known. This represents the overall operating status variable of the smart meter. This indicates that the smart meter's overall operating status is in an abnormal degradation state or a faulty state. Indicates the first The state variables of each target module Indicates the first The target module is in an abnormal degradation state or a fault state. Indicates the first All target modules are in normal condition. Indicates the total number of target modules. Indicates the first The weights of the impact of each target module on the overall operating status of the smart meter (those skilled in the art can set them according to actual needs). The size is not limited in this invention.

[0082] S604: Based on the comparison between the operational capability degradation coefficient and the preset operational capability degradation coefficient, the overall operational status of the smart meter is taken as an abnormal degradation state or a fault state and input into the dynamic Bayesian network model as an observation condition, and the initial posterior fault probability of each target module is output.

[0083] Specifically, the operational capability degradation coefficient is first compared with a preset operational capability degradation coefficient. When the operational capability degradation coefficient reaches or exceeds a preset threshold, the overall operating state of the smart meter is determined to be in an abnormal degradation state or a fault state, and this state is input as an observation condition to the system state node in the dynamic Bayesian network. Under the constraint of the observation condition, the state probability of each target module is updated in reverse through the dynamic Bayesian network, transforming the prior probability of each target module based on historical and real-time data into a conditional probability under the condition of overall system abnormality. Finally, the updated probability of each target module being in an abnormal degradation state or a fault state is output as the initial posterior fault probability.

[0084] S605: Based on the coupling relationship between each target module, the initial posterior failure probability is weighted and corrected to obtain the posterior failure probability of each target module.

[0085] Specifically, taking any target module as the current correction object, the coupling weight coefficients between this target module and the other target modules are obtained. The initial posterior fault probabilities of each associated target module are multiplied by their corresponding coupling weight coefficients and then summed using a weighted method. The weighted sum is then divided by the sum of the coupling weight coefficients between this target module and the other target modules to obtain the posterior fault probability of this target module under the known abnormal degradation or fault state of the overall smart meter operation. Thus, the posterior fault probability of the target module not only reflects its own initial fault probability under overall abnormal observation conditions but also comprehensively reflects the influence of other target modules coupled with it on its fault state.

[0086] Optionally, the coupling weight coefficient is determined as follows: Within a preset time window, the occurrence of faults in each target module is statistically analyzed, and associated events where anomalies occur simultaneously or sequentially between different target modules are identified. Using a specific target module as a benchmark, the frequency of synchronous or delayed anomalies in other target modules when the module fails is statistically analyzed to characterize the correlation strength between modules. The correlation strengths corresponding to the same target module are normalized to obtain the coupling weight coefficients between each target module, thus making the coupling relationships between different modules comparable. The larger the coupling weight coefficient, the stronger the fault correlation between the corresponding modules, and the more significant the impact on the results when correcting the initial posterior fault probability.

[0087] In this embodiment of the invention, a dynamic Bayesian network model is constructed that includes functional relationships between modules and system state dependencies. Under the observation condition of an abnormal overall operating state, the state probabilities of each target module are inferred backwards. Simultaneously, the posterior probabilities are weighted and corrected based on the coupling relationships between modules. This allows for tracing the potential fault contribution of each module downwards from the system level, achieving a refined decomposition of responsibility from overall anomalies to local modules. This approach not only improves the accuracy and interpretability of fault location but also fully considers the correlation effects between modules and the fault propagation effect, thereby obtaining a posterior fault probability that better reflects the actual operating mechanism. This provides a more reliable decision-making basis for subsequent upgrade priority ranking and path planning.

[0088] S7: Calculate the upgrade priority weight of each target module based on the posterior failure probability, and determine the upgrade path of each target module according to the upgrade priority weight.

[0089] Optionally, the calculation method for upgrade priority weight specifically includes sub-steps S701 to S705: S701: Normalize the probability of each posterior fault to obtain the basic risk proportion of each target module.

[0090] Specifically, firstly, the posterior failure probabilities of each target module are summarized, and the sum of the posterior failure probabilities of all target modules is calculated. Then, the posterior failure probabilities corresponding to each target module are divided by the sum to convert the original posterior failure probabilities into a relative proportion form under a unified scale, thereby obtaining the basic risk proportion of each target module, so that the risk representation of each target module is transformed from an absolute probability into a proportional relationship relative to the overall failure risk distribution.

[0091] S702: Calculate the mean and dispersion of the posterior fault probability distribution, and based on the mean and dispersion, calculate the relative deviation coefficient of each target module: in, Indicates the first The relative deviation coefficient of each target module Indicates the deviation adjustment coefficient. Indicates the first The posterior failure probability of each target module This represents the mean of the distribution of posterior failure probabilities for all target modules. The standard deviation represents the dispersion of the posterior failure probability distribution of all target modules. It represents a very small positive number.

[0092] S703: Calculate the priority score for each target module based on the basic risk ratio and the relative deviation coefficient.

[0093] Specifically, after obtaining the basic risk proportion and relative deviation coefficient of each target module, the basic risk proportion is used as the fundamental quantity reflecting the failure risk level of the target module, and the relative deviation coefficient is used as the adjustment quantity reflecting the degree of deviation of the target module from the overall risk distribution. Based on this, a nonlinear mapping relationship is constructed to amplify the relative deviation coefficient, and then it is fused with the basic risk proportion to obtain the priority score of each target module.

[0094] S704: Normalize each priority score to obtain the upgrade priority weight of each target module.

[0095] S705: Determine the upgrade path for each target module according to upgrade priority weights.

[0096] Specifically, after obtaining the upgrade priority weights of each target module, the target modules are first sorted in descending order according to their upgrade priority weights to obtain a priority sequence. Then, considering the system structural dependencies and execution constraints between the target modules, the priority sequence is adjusted for path constraints. For target modules with sequential dependencies, their dependent modules are upgraded before the modules they depend on. For target modules without direct dependencies, their priority-based sorting order is maintained. Based on the adjusted module sequence, the upgrade execution order of each target module is determined one by one, and the execution orders are combined to form an upgrade path.

[0097] In this embodiment of the invention, by normalizing the posterior failure probability of each target module and constructing a relative deviation coefficient by combining the distribution mean and dispersion, the upgrade priority weight is calculated based on a comprehensive reflection of the absolute risk level and relative risk prominence of the module. Furthermore, the upgrade order is optimized by combining the structural dependencies and execution constraints between modules. This enables refined sorting and reasonable scheduling of the upgrade strategies for each module of the smart meter, thereby avoiding upgrade failures caused by delayed processing of critical high-risk modules or dependency conflicts. It improves the orderliness and execution efficiency of the upgrade process, reduces the risk of system operation interruption, and enhances the security and reliability of the overall upgrade scheme.

[0098] S8: Execute the upgrade path to complete the upgrade of the smart meter.

[0099] Specifically, upgrade instructions for the corresponding target modules are generated based on the upgrade path. The upgrade package, configuration parameters, or firmware file matching the target module is retrieved. The upgrade content is distributed sequentially according to the execution order determined by the upgrade path, and version verification, integrity verification, and operational status self-check are performed after each target module upgrade is completed. If the verification passes, the system switches to the upgraded version; if the verification fails, a rollback or retry is performed. After all target modules have been upgraded, an upgrade result record is generated, and the module version status information of the smart meter is updated.

[0100] Reference manual attached Figure 2 The diagram shows a schematic representation of a smart meter upgrade system provided in an embodiment of the present invention.

[0101] This invention provides a smart meter upgrade system 20, including: a processor 201 and a memory 202; The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the smart meter upgrade method described above and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0102] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0103] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for upgrading a smart meter, characterized in that, include: S1: Obtain historical and real-time operating data of the smart meter; S2: Based on the historical operating data and the real-time operating data, calculate the operating capability impact coefficient of the smart meter at the current moment using the kernel principal component analysis algorithm; S3: Based on the operational capability impact coefficient, predict the operational capability degradation trajectory of the smart meter in the future period; S4: Based on the operational capability degradation trajectory, calculate the operational capability degradation coefficient of the smart meter in the future time period, and determine whether the operational capability degradation coefficient is greater than the preset operational capability degradation coefficient; If so, determine that the smart meter needs to be upgraded, and proceed to S5; Otherwise, it is determined that the smart meter does not require an upgrade; S5: Using the smart meter upgrade identification model, identify multiple target modules in the smart meter that need to be upgraded; S6: Perform reverse reasoning on each of the target modules to obtain the posterior failure probability of each of the target modules; S7: Calculate the upgrade priority weight of each target module based on the posterior failure probability, and determine the upgrade path of each target module according to the upgrade priority weight; S8: Execute the upgrade path to complete the upgrade of the smart meter.

2. The smart meter upgrade method according to claim 1, characterized in that, After S1 and before S2, it also includes: S1A: Preprocess the historical operation data, including data cleaning, data alignment, data noise reduction, data labeling, and data normalization. S1B: Based on the preprocessed historical operating data, determine the core operating status features used to describe the operating status of the smart meter.

3. The smart meter upgrade method according to claim 2, characterized in that, The S1B specifically includes: S1B1: Extract the basic operational features of the preprocessed historical operational data, generate derived operational features based on the basic operational features, combine the basic operational features and the derived operational features to construct a candidate core operational status feature pool; S1B2: Redundancy filtering and saliency filtering are performed sequentially on the candidate core operating state feature pool to obtain the filtered candidate core operating state feature pool. S1B3: Calculate the entropy weight of each running feature in the filtered candidate core running state feature pool using the entropy weight method. S1B4: Using each of the aforementioned operational features as input features and the smart meter operational status label as output, train a random forest model and calculate the feature importance score of each of the aforementioned operational features using the Gini index decrease. S1B5: Normalize the entropy weight method weights and the feature importance scores, and then perform weighted fusion on the normalized entropy weight method weights and feature importance scores to obtain the final importance score of each of the running features. S1B6: Based on the final importance score, sort each operating feature in the filtered candidate core operating status feature pool in descending order, and determine the preset number of operating features at the top of the sort as the core operating status features used to describe the operating status of the smart meter.

4. The smart meter upgrade method according to claim 2, characterized in that, S2 specifically includes: S201: Perform correlation analysis on the preprocessed historical operating data and the core operating status characteristics, filter out the target operating data that affects the core operating status characteristics from the preprocessed historical operating data, and construct an operating data input matrix; S202: Standardize the running data input matrix to obtain a standardized running data input matrix, and map each data sample in the standardized running data input matrix to a high-dimensional feature space; S203: Calculate the covariance matrix of the standardized running data input matrix mapped to the high-dimensional feature space, and perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; S204: Represent the feature vector as a linear combination of each of the data samples in the high-dimensional feature space to obtain the sample coefficient representation corresponding to the feature vector, so as to transform the feature decomposition problem in the high-dimensional feature space into a sample coefficient solving problem; S205: Based on the sample coefficient representation and the inner product relationship of each data sample in the high-dimensional feature space, construct a kernel matrix, and transform the feature decomposition process corresponding to the covariance matrix into a feature solving process on the kernel matrix; S206: Based on the eigenvalues ​​of the kernel matrix, extract multiple uncorrelated kernel principal components and calculate the variance contribution rate of each kernel principal component; S207: Calculate the cumulative variance contribution rate based on each of the variance contribution rates, and select the target kernel principal component from the kernel principal components based on the cumulative variance contribution rate; S208: Take the data sample corresponding to the real-time running data as the new data sample, calculate the projection value of the new data sample on the direction of each target kernel principal component, and obtain the kernel principal element corresponding to the new data sample; S209: Using the variance contribution rate of each of the target kernel principal components as a weighting coefficient, the kernel principal components corresponding to the newly added data samples are weighted and fused to obtain the operational capability impact coefficient of the smart meter at the current moment.

5. The smart meter upgrade method according to claim 1, characterized in that, S3 specifically includes: S301: Construct a time series of core operating status features based on the core operating status features corresponding to the historical operating data and the real-time operating data; S302: The core operating state feature time series is stabilized by conventional difference and seasonal difference to obtain a stable core operating state feature time series; S303: Based on the time series of characteristics of the stable core operating state and the operating capacity influence coefficient, construct a smart meter operating capacity degradation prediction model with the operating capacity influence coefficient as an exogenous variable; S304: Construct an adaptive function for the smart meter operation capability degradation prediction model, and determine the optimal order combination of the smart meter operation capability degradation prediction model by optimizing the algorithm with the goal of minimizing the function value of the adaptive function. S305: Based on the optimal order combination, initialize the smart meter operation capability degradation prediction model, and update the model parameters of the smart meter operation capability degradation prediction model online using the least squares method; S306: The average value of the historical operating capacity influence coefficient is used as the operating capacity influence coefficient for the next future moment and input into the updated smart meter operating capacity degradation prediction model. By performing backward multi-step rolling prediction, a sequence of operating capacity prediction values ​​for multiple future moments is obtained, and the sequence of operating capacity prediction values ​​is determined as the operating capacity degradation trajectory of the smart meter in the future period.

6. The smart meter upgrade method according to claim 1, characterized in that, The calculation method for the operational capability degradation coefficient specifically includes: S401: Based on the operational capability degradation trajectory, extract the predicted operational capability value and the first arrival time of the error limit for each core operational status feature at a single future moment; S402: Calculate the individual degradation coefficient of each of the core operating state characteristics based on the predicted operating capacity values; S403: Calculate the weighting coefficient of each of the core operating state features based on the historical observation data of each of the core operating state features; S404: The individual degradation coefficients and the weight coefficients are weighted and fused to obtain the initial operating degradation coefficients for the future single moment; S405: Calculate the time degradation correction coefficient based on the first arrival time of the error limit for each of the core operating state characteristics; S406: Using the time degradation correction coefficient, the initial operating degradation coefficient is corrected to obtain the operating capability degradation coefficient for the future single moment; S407: Repeat S401 to S406 to calculate the operational capability degradation coefficient for each future moment within the future time period, and select the maximum operational capability degradation coefficient as the operational capability degradation coefficient of the smart meter within the future time period.

7. The smart meter upgrade method according to claim 1, characterized in that, The smart meter upgrade identification model is specifically a smart meter upgrade identification model based on a long short-term memory neural network. The long short-term memory neural network includes a fusion memory gate and an output gate. The fusion memory gate is formed by fusing the forget gate and the input gate of the original long short-term memory neural network.

8. The smart meter upgrade method according to claim 1, characterized in that, S6 specifically includes: S601: Construct a dynamic Bayesian network model based on the functional relationships between the target modules and the influence of each target module on the overall operating status of the smart meter. S602: Discretize the operating status of each target module, and determine the prior probability of each target module being in an abnormal degradation state or a fault state based on the historical operating data and the real-time operating data. S603: Calculate the influence weight of each target module on the overall operating status of the smart meter, and construct a conditional probability relationship model based on the influence weight; S604: Based on the comparison result between the operational capability degradation coefficient and the preset operational capability degradation coefficient, the overall operational status of the smart meter is in an abnormal degradation state or a fault state as an observation condition and input into the dynamic Bayesian network model, and output the initial posterior fault probability of each target module. S605: Based on the coupling relationship between each of the target modules, the initial posterior fault probability is weighted and corrected to obtain the posterior fault probability of each of the target modules.

9. The smart meter upgrade method according to claim 1, characterized in that, The calculation method for the upgrade priority weight specifically includes: S701: Normalize the posterior failure probabilities of each of the above-mentioned faults to obtain the basic risk proportion of each of the target modules; S702: Calculate the mean and dispersion of the posterior fault probability distribution, and calculate the relative deviation coefficient of each of the target modules based on the mean and dispersion. S703: Calculate the priority score for each of the target modules based on the basic risk ratio and the relative deviation coefficient; S704: Normalize each of the priority scores to obtain the upgrade priority weight of each of the target modules; S705: Determine the upgrade path for each target module according to the upgrade priority weight.

10. A smart meter upgrade system, characterized in that, include: Processor and memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the smart meter upgrade method as described in any one of claims 1 to 9.

Images