Intelligent electric energy meter degradation prediction method and system based on pre-training

Abstract

The invention discloses an intelligent electric energy meter degradation prediction method and system based on pre-training. The method comprises the steps of obtaining operation state data X of an intelligent electric energy meter in a target typical environment; the operation state data X comprises a part or all of an environment stress parameter, data acquisition time, a load power level of the intelligent electric energy meter, a load power factor and a performance reduction index; dividing the running state data X into a plurality of slices and constructing a slice sequence Xp; and obtaining a degradation prediction result of the intelligent electric energy meter by using a pre-trained degradation trend prediction model DeGra-PTM on the slice sequence Xp, wherein the degradation trend prediction model DeGra-PTM is pre-trained to establish a mapping relation between the input slice sequence Xp and the output basic error. The method aims to improve the generalization ability and training efficiency of the prediction model, and achieves the accurate prediction of the degradation trend of the intelligent electric energy meter in the high dry heat environment at a lower training cost.

Description

Technical Field

[0001] This invention relates to the field of smart meter degradation trend prediction technology, specifically to a pre-trained smart meter degradation prediction method and system. Background Technology

[0002] In recent years, predicting the aging trends of industrial products has become increasingly important for ensuring operational reliability and reducing lifecycle costs, especially for highly reliable, long-life devices such as smart meters. As a fundamental component of modern power systems, smart meters are responsible for accurate energy metering, real-time monitoring, and supporting smart grid functions. By the end of 2023, the number of grid-connected smart meters in China exceeded 670 million, each with a designed lifespan of at least 16 years and a mandatory recalibration interval generally not exceeding 8 years. These meters degrade under environmental stress and operating conditions, and without proactive management, this could lead to serious technical and economic consequences.

[0003] Performance degradation in smart meters typically manifests as inaccurate metering, communication failures, power supply anomalies, or complete device malfunction. For example, metering errors can lead to systemic billing errors, undermining the fairness of electricity trading and causing economic losses to power companies or consumers. Communication interruptions can hinder remote data acquisition and control, affecting the efficiency of grid monitoring. Clock battery failures can result in incorrect electricity usage time markings, interfering with time-of-use pricing applications. In extreme cases, undetected performance degradation can trigger catastrophic failures, such as overheating or combustion, posing a direct risk to grid safety and stability.

[0004] Beyond the technological implications, the traditional approach of replacing smart meters at fixed time intervals incurs significant costs. These costs include the purchase of new meters, installation labor, and the environmental burden of handling e-waste. It is estimated that shifting from a traditional "time-based replacement" strategy to a condition-based "fail-to-replace" policy could save the nation trillions of dollars annually, while also significantly reducing plastic and metal waste. Therefore, accurate assessment of smart meter degradation trends is not merely a technical task, but a crucial tool for achieving sustainable, cost-effective, and reliable grid operation.

[0005] Currently, there are many methods for assessing the degradation trends and analyzing the reliability of smart meters. The main research directions can be divided into modeling based on accelerated degradation tests and modeling based on online condition monitoring databases. Modeling based on accelerated degradation tests uses elevated stress conditions to expose product defects in a short period. However, this method requires carefully designed test plans based on specific conditions (such as environmental, electrical, or mechanical loads), making it both complex and expensive. More importantly, modeling based on accelerated degradation tests struggles to capture the dynamic degradation trends of smart meters under actual operating conditions, because stress curves in real-world operating environments are typically variable and multidimensional.

[0006] In contrast, data-driven methods utilizing online condition monitoring are better suited for predictive analysis of degradation trends in field applications. By continuously collecting degradation data (such as temperature, vibration, and electrical parameters), these methods can reflect the actual operating status of smart meters. Artificial intelligence technology further enhances the ability to simulate complex degradation processes under real-world operating conditions. While many researchers have employed different deep learning models to predict and analyze the degradation trends of smart meters, their models more or less neglect device-specific attributes and cross-device variability, thus weakening their inductive power. Current models are typically trained on limited datasets from a single batch or manufacturer, ignoring key device-level features such as manufacturing tolerances, historical load curves, and material degradation characteristics. Furthermore, the lack of standardized stress modeling protocols between different experimental designs further exacerbates interoperability issues. For example, while CNNs have been applied to specific scenarios (such as high-dry-heat environments), their performance degrades when applied to new batches of devices or operating environments, thus requiring repeated retraining. Given that the field data collection cycle can take 12-30 months, this "one device, one model" approach is computationally too demanding and violates the principle of broad predictive applicability.

[0007] Furthermore, pre-training models have become a fundamental approach, driving significant progress in fields such as natural language processing and computer vision, and providing a promising method for addressing the critical challenge of predicting the degradation trends of industrial smart meters. The success of this model depends on two key factors: first, it fully learns the complex temporal variation patterns of each input feature, ensuring the accuracy of degradation trend prediction; second, massive, high-quality training data is the foundation for learning generalizable representations, and using this data during model pre-training allows for the full learning of common information among different smart meters. How to implement this model for predicting the degradation of smart meters has become a critical technical problem that urgently needs to be solved. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to provide a pre-trained method and system for predicting the degradation of smart energy meters, which aims to improve the generalization ability and training efficiency of the prediction model and achieve accurate prediction of the degradation trend of smart energy meters under high dry and hot environments with lower training costs.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A pre-trained method for predicting the degradation of smart energy meters includes the following steps: S101, Obtain the operating status data of the smart energy meter under the target typical environment. X The running status data X This includes environmental stress parameters, data acquisition time, the load power level of the smart meter, the load power factor, and some or all of the performance degradation indicators. S102, will transfer the running status data X Divide into several slices and construct a slice sequence X p ; S103, slice the sequence X p The degradation prediction results of smart energy meters are obtained by using a pre-trained degradation trend prediction model DeGra-PTM, which is pre-trained to establish the input slice sequence. X p The mapping relationship between the basic error of the output.

[0010] Optionally, in step S101, the environmental stress parameters include some or all of the following: temperature, humidity, air pressure, wind speed, light intensity, precipitation, altitude, and ultraviolet intensity; and the performance degradation indicators include some or all of the following: basic error, power-on error, power-off error, and daily timing error.

[0011] Optionally, in step S103, the degradation trend prediction model DeGra-PTM includes a word embedding layer, a position embedding layer, a multi-head self-attention mechanism layer, a first residual module, an expert layer, a second residual module, a feature extraction module, and a linear layer connected in sequence. The expert layer is constructed based on a feedforward network. The first residual module is used to concatenate the input and output features of the multi-head self-attention mechanism layer and output the result. The second residual module is used to concatenate the input and output features of the expert network and output the result. The input slice sequence... X p The basic error of the degradation trend prediction model DeGra-PTM is obtained from the word embedding layer and the output is obtained through the linear layer.

[0012] Optionally, the functional expression of the multi-head self-attention mechanism layer is: ; in, The output features of the multi-head self-attention mechanism layer, The input features are for the multi-head self-attention mechanism layer. For feature splicing operations, ~ These are the output features of the attention heads from the 1st to the hth. Let be the output projection matrix of the multi-head self-attention mechanism, and let the functional expression of the output features of any attention head be: ; in, For the output features of the i-th attention head, For attention operations, , and Let be the query matrix, key matrix, and value matrix of the i-th attention head, respectively. for transpose, for The dimensions are defined by the query matrix, key matrix, and value matrix, which are the input features. It is obtained by mapping different linear layers.

[0013] Optionally, the expert layer based on the feedforward network consists of a linear layer, a sigmoid function, and another linear layer, and the function expression of the expert layer based on the feedforward network is: , in, The output features of the expert layer based on the feedforward network, For the input features of the expert layer, and The output projection matrix of the linear layer. This is the Sigmoid function.

[0014] Optionally, the degradation trend prediction model DeGra-PTM includes, during pre-training, the current time window... slice sequence X p The degradation trend prediction model DeGra-PTM is used to obtain the predicted degradation results, and the predicted degradation results are then mapped through a linear layer as the next time window. The label values ​​of the basic error used when training the DeGra-PTM degradation trend prediction model, where For the current time, This represents the size of the time window.

[0015] Optionally, the loss function used in the pre-training of the degradation trend prediction model DeGra-PTM has the following expression: ; in, For loss function, For the next time window The basic error is obtained through the DeGra-PTM degradation trend prediction model. For the next time window The label value of the basic error.

[0016] The present invention also provides a pre-trained smart meter degradation prediction system, comprising a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the pre-trained smart meter degradation prediction method.

[0017] The present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the pre-trained smart meter degradation prediction method by a processor.

[0018] The present invention also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the pre-trained smart meter degradation prediction method by a processor.

[0019] Compared with existing technologies, the present invention mainly achieves the following beneficial effects: The pre-trained smart meter degradation prediction method of the present invention includes acquiring the operating status data of smart meters under typical target environments. X The running status data X This includes environmental stress parameters, data acquisition time, the load power level of the smart meter, load power factor, and some or all of the performance degradation indicators; and operational status data. X Divide into several slices and construct a slice sequence X p ; slice sequence X p The degradation prediction results of smart energy meters are obtained by using a pre-trained degradation trend prediction model DeGra-PTM, which is pre-trained to establish the input slice sequence. X p The mapping relationship between the basic error of the output and the prediction model. This invention aims to improve the generalization ability and training efficiency of the prediction model, and to achieve accurate prediction of the degradation trend of smart energy meters under high dry and hot environments with lower training costs. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the network structure of the DeGra-PTM degradation trend prediction model in an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the pre-training process of the DeGra-PTM degradation trend prediction model in an embodiment of the present invention.

[0023] Figure 4 This is a schematic diagram of the dataset partitioning during the experiment in this embodiment of the invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0025] like Figure 1 As shown, the pre-trained smart meter degradation prediction method in this embodiment includes the following steps: S101, Obtain the operating status data of the smart energy meter under the target typical environment. X The running status data X This includes environmental stress parameters, data acquisition time, the load power level of the smart meter, the load power factor, and some or all of the performance degradation indicators. S102, will transfer the running status data X Divide into several slices and construct a slice sequence X p ; S103, slice the sequence X p The degradation prediction results of smart energy meters are obtained by using a pre-trained degradation trend prediction model DeGra-PTM, which is pre-trained to establish the input slice sequence. X p The mapping relationship between the basic error of the output.

[0026] In step S101, the environmental stress parameters include some or all of temperature, humidity, air pressure, wind speed, light intensity, precipitation, altitude, and ultraviolet intensity. The performance degradation indicators include some or all of basic error, power-on error, power-off error, and daily timing error. As an optional implementation, in this embodiment, the smart energy meter under test operates in real-time under high-dry-heat natural conditions, consuming electricity using a standard source at the State Grid Xinjiang High-Dry-Heat Test Base. The standard meter measures the measurement error of the smart energy meter under test, and sensors such as temperature, humidity, air pressure, wind speed, and ultraviolet radiation are used to monitor and collect environmental data in real time. The test platform in this embodiment consists of a highly representative smart energy meter system, containing 1369 energy meters from 149 manufacturers. Environmental stress parameters and smart energy meter operating data are collected every 4 hours at the test base. Environmental stress parameters include temperature (…). Te. ),humidity( Hu. ), air pressure ( AP ), wind speed ( WS ), light intensity ( Il. ), precipitation ( Pr. ), altitude ( Al. ) and ultraviolet ( UR It also implements seven load power levels: 0.05Ps, 0.1Ps, 0.2Ps, 0.5Ps, Ps, 0.5Pmax, and Pmax, where Ps represents the load power and Pmax = 12Ps represents the maximum power. Five load power factor settings are also configured: 1, 0.25L, 0.05L, 0.5C, and 0.8C, where 1 represents a purely resistive load, L represents an inductive load, and C represents a capacitive load. Furthermore, data on various performance degradation indicators of the smart meter, such as basic error (…), were collected. BE The data includes power-on / power-off errors and daily timing errors. Finally, this experimental information is integrated into the base station to form the operational status data for the DeGra-PTM degradation trend prediction model. X : ; In the formula, Te. Indicates temperature; Hu. Indicates humidity; AP Indicates air pressure; WS Indicates wind speed; Il. Indicates light intensity; Pr. Indicates precipitation; Al. Indicates altitude; UR Indicates ultraviolet radiation; PL Indicates power rating; PF Indicates the power factor; Ph. Indicates the number of phases in a smart energy meter; Man. Indicates the manufacturer; ID Indicates the electricity meter code; t Indicates the data collection time, including year, month, day, hour, minute, and second; BE This indicates the basic error.

[0027] Since the input to the DeGra-PTM degradation trend prediction model is typical time series data, to improve prediction accuracy, the model in this embodiment needs to learn the correlation between data within each time step. However, most existing studies use point data as input, which often leads to insufficient extraction of valuable information. Therefore, in step S102 of this embodiment, the running status data of the DeGra-PTM degradation trend prediction model is used as input. X Converting data into fragments makes it possible to capture local information from the input data, thereby improving the comprehensiveness of the extracted information. Specifically, the original runtime state data is first... X Divide into several slices, and then combine them into a new slice sequence. X p Furthermore, this embodiment also includes slice sequences. X p Normalization was performed to mitigate the distribution bias between the training and test data. In particular, normalization was performed separately for each feature curve, thus avoiding the problem of dealing with inconsistencies in the shape of the input data.

[0028] Considering that this invention aims to achieve a unified model to learn general knowledge to handle the degradation trends of different energy meters, a large amount of degradation data from different energy meters is required for sufficient training. This heterogeneous smart energy meter data increases the training difficulty, thus requiring a more powerful model. This embodiment combines the idea of ​​pre-training and utilizes the Transformer architecture to propose a degradation trend prediction model, DeGra-PTM. This model not only solves the problem that traditional time-series prediction models can only focus on a single energy meter or meters from a single manufacturer and have low generalization ability, but also significantly improves the model's analytical capabilities without increasing the corresponding computational load. Figure 2 As shown, the degradation trend prediction model DeGra-PTM in step S103 of this embodiment includes a word embedding layer, a position embedding layer, a multi-head self-attention mechanism layer, a first residual module, an expert layer, a second residual module, a feature extraction module, and a linear layer connected in sequence. The expert layer is built based on a feedforward network. The first residual module is used to concatenate the input and output features of the multi-head self-attention mechanism layer and output the result. The second residual module is used to concatenate the input and output features of the expert network and output the result. The input slice sequence... X p The basic error of the degradation trend prediction model DeGra-PTM is obtained from the word embedding layer and the output is obtained through the linear layer.

[0029] Although the input slice sequence X p It is no longer point-like, but in this embodiment, the input slice sequence will still be processed. X p Encoding is performed to improve its representational power. This will help the unified model better unleash its powerful computational capabilities, thereby extracting more comprehensive information from the input features. In this embodiment, the function expression for word embedding in the word embedding layer is: , In the above formula, X p Represents a slice sequence; H This represents the embedding vector after word embedding; Linear Indicates a linear transformation layer; T Indicates the slice length; d Indicates the dimension of the input features.

[0030] To preserve the location information in the input data, the DeGra-PTM degradation trend prediction model performs location embedding based on the different times of the feature values. Specifically, the function expression for location embedding in the location embedding layer is as follows: ; ; In the above formula, P Indicates a positional embedding, with the subscript corresponding to the specific embedding position; pos This indicates the location of a specific record within the entire dataset's time frame. This represents the dimension of the position embedding vector.

[0031] Since both the input and output of the degradation trend prediction model DeGra-PTM have temporal characteristics, capturing the temporal patterns of each feature is crucial for accurately predicting the degradation trend of smart meters. There are many methods for processing time-series data; in this embodiment, the DeGra-PTM degradation trend prediction model employs a multi-head self-attention mechanism to capture the long-term dependencies of each feature over time. The advantage of this method is that it is not affected by the length of the time window, and the information collected by multiple attention heads is richer and more comprehensive than that collected by a single attention head. The function expression of the multi-head self-attention mechanism layer is: ; in, The output features of the multi-head self-attention mechanism layer, The input features are for the multi-head self-attention mechanism layer. For feature splicing operations, ~ These are the output features of the attention heads from the 1st to the hth. Let be the output projection matrix of the multi-head self-attention mechanism, and let the functional expression of the output features of any attention head be: ; in, For the output features of the i-th attention head, For attention operations, , and Let be the query matrix, key matrix, and value matrix of the i-th attention head, respectively. for transpose, for The dimensions are defined by the query matrix, key matrix, and value matrix, which are the input features. It is obtained through different linear layer mappings. And it has... , h The number of attention heads; This represents the dimension of the position embedding vector.

[0032] Smart meters typically consist of eight components: a main power module, a microcontroller, a metering module, a communication module, a security module, a relay module, a clock module, and a display module. While the degradation trends of different smart meters vary, their shared basic structure means that smart meters from different manufacturers share certain commonalities in performance degradation. This makes it possible to extract commonalities from existing smart meter degradation trends and then use this general information to predict new degradation trends. Therefore, in this embodiment, the degradation trend prediction model DeGra-PTM constructs an expert layer based on a feedforward network to focus on learning the unified laws between features and different meter degradation trends, such as... Figure 2 As shown, the expert layer based on the feedforward network consists of a linear layer, a sigmoid function, and another linear layer. Its function expression is: , in, The output features of the expert layer based on the feedforward network, For the input features of the expert layer, and The output projection matrix of the linear layer. This is the Sigmoid function.

[0033] like Figure 3 As shown, in this embodiment, the degradation trend prediction model DeGra-PTM includes, during pre-training, the current time window... slice sequence X pThe degradation trend prediction model DeGra-PTM is used to obtain the predicted degradation results, and the predicted degradation results are then mapped through a linear layer as the next time window. The label values ​​of the basic error used when training the DeGra-PTM degradation trend prediction model, where For the current time, This represents the size of the time window. In this embodiment, the loss function used by the DeGra-PTM degradation trend prediction model during pre-training has the following expression: ; in, For loss function, For the next time window The basic error is obtained through the DeGra-PTM degradation trend prediction model. For the next time window The label value of the basic error. By dynamically adjusting the parameters in modules such as the expert layer through the loss between the predicted and the true values, the expert layer and other modules can learn the general expression directly between the input features and the degradation trends of different energy meters. Due to its unique pre-training task, it can learn the general mapping relationship between external stress and the degradation trend of smart energy meters, thereby enhancing the generalization ability of the model.

[0034] Finally, the DeGra-PTM degradation trend prediction model, which is pre-trained and constructed using a multi-head self-attention mechanism and an expert layer, is used to predict the degradation trend of electricity meters. During the training phase, this model learns both the temporal relationships of different input features and the unified pattern between input features and different electricity meter degradation trends, enabling it to be applied to new electricity meter degradation trend prediction and analysis tasks.

[0035] To verify the effectiveness of the degradation trend prediction model DeGra-PTM in this embodiment, a pre-trained DeGra-PTM model is used to predict the degradation trend of new energy meters. The prediction results are compared with those of classic time-series prediction models such as Kernel Support Vector Return (KSVR), Weighted Fusion Bayesian (WFB), Backpropagation Neural Network (BPNN), Multi-View Convolutional Neural Network (MVCNN), and Long Short-Term Memory Network with Macro-Micro Attention (LSTM) on different datasets (dataset SMA to dataset SMC), demonstrating the superiority and generalization ability of the DeGra-PTM model in this embodiment. Three evaluation metrics are also introduced: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Decision Coefficient R², whose calculation expressions are as follows: ; ; ; Where N is the sample size. , and These represent the label, predicted value, and average value of the basic error for the i-th sample, respectively. For fair comparison, the hyperparameters of the above comparative models were optimized using grid search, and the ratio of training set, validation set, and test set is consistent with the model in this embodiment. The dataset partitioning method is as follows: Figure 4 As shown, the smart meters collected from 130 manufacturers between 2016 and 2025 are divided as follows: 1233 smart meters from 100 manufacturers are divided into a training set, as shown below. Figure 4 As shown in Figure ①, 50 electricity meters from 5 manufacturers are divided into a verification set, as follows: Figure 4 As shown in Figure ②, 136 electricity meters from 25 manufacturers were divided into a test set, as follows: Figure 4 As shown in ③. To quantitatively analyze the predictive performance of different models, three evaluation indicators, RMSE, MAE, and R2, were used to evaluate the different models, and the results are shown in Table 1.

[0036] Table 1 Comparison of the predictive effects of smart energy meter degradation trends

[0037] As shown in Table 1, compared with existing kernel support vector regression (KSVR), weighted fusion Bayesian (WFB), backpropagation neural network (BPNN), multi-view convolutional neural network (MVCNN), and long short-term memory network with macro and micro attention (LSTM), the degradation trend prediction model DeGra-PTM of this embodiment achieves better degradation trend prediction performance for smart meters on three datasets and on the three evaluation indicators of root mean square error (RMSE), mean absolute error (MAE), and decision coefficient (R²).

[0038] In summary, the method of this embodiment includes acquiring a dataset of smart meter operating status under high dry and hot conditions; encoding features such as external stress and basic errors of smart meters using the degradation trend prediction model DeGra-PTM, and simultaneously performing positional encoding based on the different time periods of the feature values; employing a multi-head self-attention mechanism to capture the complex temporal variation patterns of different input features; utilizing an expert layer to capture the unified expression relationship between input features and degradation trends of smart meters from different manufacturers and models; and using the trained degradation trend prediction model DeGra-PTM to predict the degradation trend of new smart meters. Through these methods, the generalization ability and training efficiency of the prediction model can be improved, achieving accurate prediction of smart meter degradation trends under high dry and hot conditions with relatively low training costs.

[0039] This embodiment also provides a pre-trained smart meter degradation prediction system, including a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to execute the pre-trained smart meter degradation prediction method.

[0040] This embodiment also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the pre-trained smart meter degradation prediction method by a processor.

[0041] This embodiment also provides a computer program product, including a computer program or instructions that are programmed or configured to execute the pre-trained smart meter degradation prediction method via a processor.

[0042] Those skilled in the art will understand that the technical solutions provided by this invention may take the form of a method, system, or computer program product. Therefore, this invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention may take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce an implementation of the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0043] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A pre-trained method for predicting the degradation of smart energy meters, characterized in that, Includes the following steps: S101, Obtain the operating status data of the smart energy meter under the target typical environment. X The running status data X This includes environmental stress parameters, data acquisition time, the load power level of the smart meter, the load power factor, and some or all of the performance degradation indicators. S102, running status data X Divide into several slices and construct a slice sequence X p ; S103, slice sequence X p The degradation prediction model DeGra-PTM, which is pre-trained, is used to obtain degradation prediction results for smart meters. The DeGra-PTM model is pre-trained to establish the input slice sequence. X p The mapping relationship between the basic error of the output.

2. The pre-trained smart energy meter degradation prediction method according to claim 1, characterized in that, In step S101, the environmental stress parameters include some or all of the following: temperature, humidity, air pressure, wind speed, light intensity, precipitation, altitude, and ultraviolet intensity. The performance degradation indicators include some or all of the following: basic error, power-on error, power-off error, and daily timing error.

3. The pre-trained smart energy meter degradation prediction method according to claim 1, characterized in that, In step S103, the degradation trend prediction model DeGra-PTM includes a word embedding layer, a position embedding layer, a multi-head self-attention mechanism layer, a first residual module, an expert layer, a second residual module, a feature extraction module, and a linear layer connected in sequence. The expert layer is built based on a feedforward network. The first residual module is used to concatenate the input and output features of the multi-head self-attention mechanism layer and output the result. The second residual module is used to concatenate the input and output features of the expert network and output the result. The input slice sequence... X p The basic error of the degradation trend prediction model DeGra-PTM is obtained from the word embedding layer and the output is obtained through the linear layer.

4. The pre-trained smart energy meter degradation prediction method according to claim 3, characterized in that, The expert layer based on the feedforward network consists of a linear layer, a sigmoid function, and another linear layer. The function expression of the expert layer based on the feedforward network is as follows: ， in, The output features of the expert layer based on the feedforward network, For the input features of the expert layer, and The output projection matrix of the linear layer. This is the Sigmoid function.

5. The pre-trained smart energy meter degradation prediction method according to claim 3, characterized in that, The function expression for the multi-head self-attention mechanism layer is: ； in, The output features of the multi-head self-attention mechanism layer, The input features are for the multi-head self-attention mechanism layer. For feature splicing operations, ~ These are the output features of the attention heads from the 1st to the hth. Let be the output projection matrix of the multi-head self-attention mechanism, and let the functional expression of the output features of any attention head be: ； in, For the output features of the i-th attention head, For attention operations, , and Let be the query matrix, key matrix, and value matrix of the i-th attention head, respectively. for transpose, for The dimensions are defined by the query matrix, key matrix, and value matrix, which are the input features. It is obtained by mapping different linear layers.

6. The pre-trained smart energy meter degradation prediction method according to claim 2, characterized in that, The degradation trend prediction model DeGra-PTM, during pre-training, includes the current time window... slice sequence X p The degradation trend prediction model DeGra-PTM is used to obtain the predicted degradation results, and the predicted degradation results are then mapped through a linear layer as the next time window. The label values ​​of the basic error used when training the DeGra-PTM degradation trend prediction model, where For the current time, This represents the size of the time window.

7. The pre-trained smart energy meter degradation prediction method according to claim 6, characterized in that, The loss function used in the pre-training of the degradation trend prediction model DeGra-PTM has the following expression: ； in, For loss function, For the next time window The basic error is obtained through the DeGra-PTM degradation trend prediction model. For the next time window The label value of the basic error.

8. A pre-trained smart energy meter degradation prediction system, comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to execute the pre-trained smart meter degradation prediction method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the pre-trained smart meter degradation prediction method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute, via a processor, the pre-trained smart meter degradation prediction method according to any one of claims 1 to 7.

Images