A method and system for analyzing the airtightness data of electricity meters

Abstract

This application relates to the field of industrial automation testing technology, specifically to a method and system for analyzing the airtightness data of electricity meters. The method includes: simultaneously acquiring raw temperature and pressure data from the electricity meter and preprocessing them to obtain temperature and standardized pressure sequences; establishing a historical temperature segmentation model library based on historical samples, and calculating the current residual sequence according to the historical temperature segmentation model; calculating the importance of historical coefficient features based on the decomposition coefficients and corresponding labels of historical samples to obtain corresponding historical feature vectors, and training the model based on the historical feature vectors; solving for the current feature vector in the residual sequence, and automatically determining the airtightness of the current electricity meter using the pre-trained model. This application not only improves the accuracy, sensitivity, and efficiency of testing but also reduces testing costs and improves production safety and reliability.

Description

Technical Field

[0001] This application relates to the field of industrial automation testing technology, and in particular to a method and system for analyzing the airtightness data of an electricity meter. Background Technology

[0002] The sealing performance of the electricity meter casing is crucial to ensuring its operational reliability. During production, pressure changes within the sealed cavity are monitored using pressure sensing methods to determine the presence of leaks. However, dynamic fluctuations in ambient temperature significantly interfere with the accuracy of detection: the gas within the sealed cavity expands and contracts with temperature changes, causing the pressure sequence to exhibit a clear temperature-dependent trend. This trend is often much larger than the pressure changes of minor leaks, masking the true leak signal. This can easily lead to temperature fluctuations being misjudged as leaks, resulting in the incorrect rejection of qualified products, or the inability to identify minor leaks due to interference, allowing unqualified products to enter the market.

[0003] Existing solutions have significant limitations: constant temperature workshops require substantial investment and are difficult to adapt to dynamic production line conditions; simple linear temperature compensation can only handle basic scenarios and cannot address complex situations such as sudden temperature changes and nonlinear thermal effects; traditional judgment methods rely on thresholds or simple feature extraction, which fail to effectively separate temperature interference from leakage signals. Especially in the detection of minute leaks, due to the superposition of interference from sensor noise, mechanical vibration, etc., it is even more difficult to accurately identify key features such as turbulent fluctuations in gas escape. Summary of the Invention

[0004] To eliminate the trend interference of dynamic fluctuations in ambient temperature on pressure sequences and accurately extract minute leakage characteristics, this application provides a method and system for analyzing the airtightness data of electricity meters, realizing automated and high-precision determination of the airtightness of electricity meters on production lines, and reducing the false and false judgment rates.

[0005] In a first aspect, this application provides a method for analyzing the airtightness data of an electricity meter, employing the following technical solution:

[0006] A method for analyzing the airtightness data of an electricity meter, comprising the following steps:

[0007] Construct and train a network model for analyzing the airtightness data of electricity meters. The input is the current data of the electricity meter and the current basic information of the electricity meter. The output is the corresponding airtightness label.

[0008] The system simultaneously acquires raw temperature and pressure data from the electricity meter and preprocesses them to obtain temperature and standardized pressure sequences. It selects first historical samples that have been confirmed to be properly sealed, establishes a historical temperature segmentation model library based on these samples, and extracts residual sequences from the preprocessed data. It then selects second historical samples with different label categories, calculates the historical coefficient feature importance based on the decomposition coefficients and corresponding labels of the second historical samples, and obtains corresponding historical feature vectors. The model is then trained based on these historical feature vectors. Finally, it solves for the residual sequences to obtain the current feature vectors, and uses the pre-trained model to automatically determine the current airtightness of the electricity meter.

[0009] The process of establishing a historical temperature segmentation model library and extracting residual sequences based on preprocessed data includes: selecting a first historical sample labeled as well-sealed to determine temperature segments, establishing a first-order linear model for each temperature segment to extract model parameters, and then obtaining the segmentation model library; matching temperature segments according to the current temperature sequence, and calling the corresponding model parameters to obtain residual sequences;

[0010] The step of calculating the importance of historical coefficient features based on the decomposition coefficients of the second historical samples and the corresponding labels to obtain historical feature vectors includes: selecting second historical samples with different labels to decompose and extract second historical sample coefficient sequences; dividing each coefficient sequence into a label set based on the second historical sample labels and obtaining the label center sequence; calculating the importance of historical coefficient features based on the coefficient sequences and the corresponding label center sequences; and then determining the core coefficient sequences to obtain the corresponding historical feature vectors.

[0011] Furthermore, the preprocessing to obtain the temperature sequence and the standardized pressure sequence includes: removing outliers from the original temperature data and smoothing and correcting the temperature data to obtain the temperature sequence; collecting the original pressure data within the normal range and smoothing and correcting it, and then performing standardization processing to obtain the standardized pressure sequence; aligning the temperature sequence and the standardized pressure sequence with timestamps to form an ordered data pair.

[0012] Furthermore, the step of selecting a well-sealed first historical sample to determine temperature segments includes preprocessing the selected first historical sample, calculating a local covariance sequence based on the preprocessed data according to a fixed temperature range window, obtaining a covariance difference sequence between adjacent temperature windows, and obtaining N temperature segments based on the covariance difference sequence and a segmentation threshold; the segmentation threshold is taken as the 95th percentile value of the difference sequence.

[0013] Furthermore, each temperature segment in the segmented model library is bound and stored with its corresponding model parameters.

[0014] Furthermore, the step of calling the corresponding model parameters to obtain the residual sequence includes taking each data point in the current temperature sequence, inputting the first-order linear model parameters corresponding to its temperature segment, obtaining the temperature-driven pressure trend value, traversing the temperature sequence to obtain the pressure trend sequence, and calculating the corresponding data difference based on the current standardized pressure sequence and the pressure trend sequence to form the residual sequence.

[0015] Furthermore, the step of dividing each coefficient sequence into a label set and obtaining the label center sequence based on the second historical sample label includes dividing a single coefficient sequence into different label sets according to the second historical sample label, and calculating the average sequence of each set as its label center sequence.

[0016] Furthermore, the calculation of the importance of historical coefficient features based on the coefficient sequence and the corresponding label center sequence includes calculating the corresponding Euclidean distance set between each coefficient sequence and the corresponding label center sequence of each second historical sample, then calculating the mean of each Euclidean distance set, obtaining the total standard deviation of all distances, and using the maximum class distance difference and the total standard deviation to represent the importance of the corresponding coefficient features.

[0017] Furthermore, the step of determining the core coefficient sequence and obtaining the corresponding historical feature vector includes sorting the coefficient features corresponding to each coefficient sequence from high to low importance, selecting the top three sequences as core coefficient sequences, and concatenating the core coefficient sequences to form the corresponding historical feature vector.

[0018] Furthermore, the step of training the model based on historical feature vectors includes fitting the model using historical feature vectors, enabling the model to learn the mapping relationship between the core coefficient sequence and the airtightness label of the electricity meter.

[0019] Secondly, this application provides a data analysis system for the airtightness of electricity meters, which adopts the following technical solution:

[0020] An energy meter airtightness data analysis system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the energy meter airtightness data analysis method described above is implemented.

[0021] This application has the following technical effects:

[0022] This application uses a segmented model to remove the influence of temperature trends. By dividing the data into multiple time periods and establishing an independent model for each time period, the interference of temperature trends on the detection results is accurately identified and eliminated, which greatly reduces the false and false detection rates caused by temperature fluctuations, making the detection results more reliable and stable.

[0023] This application employs wavelet decomposition technology combined with feature screening to decompose complex signals into sub-signals of different frequencies, thereby better capturing the detailed features in the signal; through further feature screening, features related to minute leakage signals are selected from a large number of sub-signals, thereby efficiently capturing the features of minute leakage signals, significantly improving the detection sensitivity, and enabling the earlier detection of potential leakage problems.

[0024] This application tightly integrates the testing process with the production line, achieving automated data acquisition, processing, and analysis, significantly improving testing speed and efficiency. By analyzing historical data, this application establishes multiple models and selects the appropriate model for use in new testing processes based on actual conditions. This ensures both testing accuracy and efficiency, achieving a balance between the two.

[0025] In summary, this application not only improves the accuracy, sensitivity, and efficiency of detection, but also reduces detection costs and enhances the safety and reliability of production. Attached Figure Description

[0026] The above and other objects, features, and advantages of the present invention will become readily apparent from the following detailed description of exemplary embodiments, accompanied by the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein:

[0027] Figure 1 This is a flowchart of a method for analyzing the airtightness data of an electricity meter provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments 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 are within the scope of protection of the present invention.

[0029] This application discloses a method for analyzing the airtightness data of an electricity meter, referring to... Figure 1 ,include:

[0030] S101: Synchronously acquire raw temperature and pressure data from the electricity meter and preprocess them to obtain temperature and standardized pressure sequences.

[0031] Specifically, the preprocessing to obtain the temperature sequence and the standardized pressure sequence includes: removing outliers from the original temperature data and smoothing the temperature data to obtain the temperature sequence; collecting the original pressure data within the normal range and smoothing it, and then performing standardization processing to obtain the standardized pressure sequence; and aligning the temperature sequence and the standardized pressure sequence with timestamps to form an ordered data pair.

[0032] Specifically, in this embodiment, after the electricity meter casing is sealed, it is left to stand still for 3 seconds to allow the airflow to stabilize (based on fluid dynamics simulation and actual measurement verification, standing still for 3 seconds after sealing allows the gas turbulence inside the cavity to decay to a stable state). The pressure sensor (sampling interval 0.1s) and temperature sensor (sampling interval 0.1s) simultaneously collect data. In this embodiment, within a 30s pressure holding period, pressure and temperature sequences of 301 data points are acquired respectively, and the data sequences are bound to the electricity meter's unique identifier and test batch number and stored in a local database.

[0033] Specifically, the temperature and pressure sequences are preprocessed. For the temperature sequence, outliers exceeding the measurement range of 10–60°C are removed, and no-trend noise points with single temperature jumps exceeding 2°C are corrected. The remaining data are smoothed using a 3-point sliding window to obtain the final temperature sequence. For pressure sequences, if data points exceed the 0-500 Pa range (a normal range set based on historical qualified samples), it is considered a sensor malfunction and a re-acquisition is triggered. For isolated noise points with a single jump exceeding 30 Pa and no trend support, the mean of two adjacent points is used for correction; the remaining data are smoothed using a 3-point sliding window. The pressure unit is standardized to "Pa" and the temperature unit to "°C" to eliminate unit bias. The pressure sequence is standardized using Z-score processing.

[0034]

[0035] in, Indicates the first in real-time data Data points, ; Represents the original pressure sequence The first in One pressure data point; Map the pressure data to the mean and standard deviation of the sequence during the pressure holding period of this sample. The interval forms a standardized pressure sequence:

[0036]

[0037] Indexed by timestamps, the preprocessed standardized stress sequences are... With temperature sequence Alignment to form ordered data pairs This ensures consistency in timing for subsequent analyses.

[0038] S102: Establish a historical temperature segmentation model library based on the first historical sample, and obtain the current residual sequence according to the historical temperature segmentation model.

[0039] Specifically, based on a large number of historical qualified sealing samples, temperature segment boundaries are determined through a data-driven approach, pressure-temperature correlation models for each segment are trained, and a reusable static model library is formed.

[0040] Specifically, establishing a historical temperature segmentation model library and extracting residual sequences based on preprocessed data includes: selecting historical samples labeled as well-sealed to determine temperature segments and establishing a first-order linear model for each temperature segment to extract model parameters, thereby obtaining the segmentation model library; matching temperature segments according to the current temperature sequence and calling the corresponding model parameters to obtain the residual sequence.

[0041] Specifically, selecting well-sealed historical samples to determine temperature segments includes preprocessing the selected historical samples, calculating the local covariance sequence based on the preprocessed data according to a fixed temperature range window, obtaining the covariance difference sequence of adjacent temperature windows, and obtaining N temperature segments based on the covariance difference sequence and the segmentation threshold; the segmentation threshold is taken as the 95th percentile value of the difference sequence.

[0042] In this embodiment, sealed and qualified samples (sample size ≥ 1000) confirmed by manual annotation within the past 3 months were selected as the first historical samples. The temperature sequence of each sample (sampling interval 0.1s, 301 data points in total over a 30s period) was extracted. and standardized pressure sequence This ensures coverage of the entire temperature fluctuation range. For historical samples... and Calculate the local covariance within a fixed temperature range window. ( Indicates the first in historical data Data points, ), quantify the strength of the linear correlation between temperature and pressure within a specific temperature range; calculate the covariance difference between adjacent temperature windows. The 95th percentile of the differential sequence is used as the segmentation threshold. ,Will By merging continuous temperature intervals, N historical segments are obtained, and fixed temperature segment boundaries are acquired. For each historical segment, extract the subsequences from all first historical samples whose temperatures fall within that interval. and corresponding pressure subsequence Establish a first-order linear model ,in For temperature sensitivity coefficient, The offset term is used; the least squares method is employed to solve for the model parameters, minimizing the error between the predicted and historical pressure values. The segmented boundaries are defined. With corresponding model parameters Bind storage to form a historical temperature segmentation model library.

[0043] Specifically, real-time temperature values ​​are matched with historically preprocessed temperature segments, and the pressure-temperature correlation model parameters for the corresponding segments are directly called, avoiding real-time segmented calculations. This applies to real-time temperature sequences. Each data point in , Traverse the segment boundaries in the historical temperature segmentation model library and determine the interval to which it belongs: If ,but Belongs to the Each historical segment calls the model parameters for that segment. .

[0044] Specifically, calling the corresponding model parameters to obtain the residual sequence includes taking each data point in the current temperature sequence, inputting the first-order linear model parameters corresponding to its temperature segment, obtaining the temperature-driven pressure trend value, traversing the temperature sequence to obtain the pressure trend sequence, and calculating the corresponding data difference based on the current standardized pressure sequence and the pressure trend sequence to form the residual sequence.

[0045] This embodiment calculates the pressure trend value caused by real-time temperature changes based on the matched historical temperature segmentation model parameters, and quantifies the pure temperature disturbance component.

[0046] For each data point in the real-time sequence Substitute into its corresponding segment The model parameters are used to calculate the temperature-driven pressure trend value:

[0047]

[0048] in, This is the temperature sensitivity coefficient (slope) for this segment, which physically represents the average change in pressure for every 1°C change in temperature. This is a constant offset term, reflecting the offset of the reference pressure at a temperature of 0℃;

[0049] After traversing the entire sequence, the complete pressure trend term is obtained by concatenating the sequences:

[0050]

[0051] This sequence quantifies the dynamic impact of pure temperature changes on pressure within a period.

[0052] Specifically, by subtracting the temperature trend term from the standardized pressure sequence, a residual sequence signal reflecting the true airtightness is obtained:

[0053]

[0054] Output residual sequence:

[0055]

[0056] in, This involves removing the pure temperature interference component from the standardized pressure to obtain a clean pressure signal that only reflects the sealing condition. At this point, the influence of temperature fluctuations has been largely eliminated from the residual sequence. If the electricity meter is well-sealed, It exhibits irregular, minute noise fluctuations; if the electricity meter has a leak, It will show a significant and continuous downward trend, and will not be affected by changes in ambient temperature.

[0057] S103: Calculate the importance of historical coefficient features based on the decomposition coefficients of the second historical samples and the corresponding labels to obtain the corresponding historical feature vectors, and train the model based on the historical feature vectors; solve for the residual sequence to obtain the current feature vector, and combine the pre-trained model to automatically determine the airtightness of the current electricity meter.

[0058] Specifically, the process of calculating the importance of historical coefficient features based on the decomposition coefficients of the second historical samples and the corresponding labels to obtain historical feature vectors includes: selecting second historical samples with different labels to decompose and extract the coefficient sequences of the second historical samples; dividing each coefficient sequence into a label set based on the labels of the second historical samples and obtaining the label center sequence; calculating the importance of historical coefficient features based on the coefficient sequences and the corresponding label center sequences; and then determining the core coefficient sequences to obtain the corresponding historical feature vectors.

[0059] Specifically, for high-speed inspection scenarios on production lines, key information is screened based on the statistical importance of wavelet decomposition coefficients, and a support vector machine model is used to achieve automated and accurate determination of minute leaks, supporting real-time sorting decisions in the workshop. Low-frequency approximation coefficients and high-frequency detail coefficients are obtained through wavelet multi-scale decomposition. In production line inspection, residual signals are mixed with interference from workshop mechanical vibrations, sensor jitter, etc., while the dynamic characteristics of minute leak signals are implicit in low-frequency trends and high-frequency fluctuations. These are directly obtained through wavelet decomposition. In this embodiment, a 3-level db4 wavelet decomposition (balancing feature preservation and efficiency) is used to analyze the current residual sequence. After decomposition, we get:

[0060] Low-frequency approximation coefficients: (Finer scale, low frequency). (Medium-coarse scale, low frequency). (The coarsest scale, low frequency, with the lowest frequency, potentially containing the most stable downward trend).

[0061] High-frequency detail factor: (Highest frequency details, potentially including sensor jitter). (Mid-to-high frequency details, potentially including turbulent fluctuations) (Lower frequency details, potentially including mechanical vibrations).

[0062] Specifically, dividing each coefficient sequence into a label set based on the second historical sample label and obtaining the label center sequence includes dividing a single coefficient sequence into different label sets according to the second historical sample label, and calculating the average sequence of each set as its label center sequence.

[0063] Specifically, the importance of historical coefficient features is calculated based on the decomposition coefficients of the second historical samples and their corresponding labels. Historical manually labeled samples (well-sealed / suspected leak / leakage non-compliance) are used to quantify the importance of these features through statistical methods. The feature importance of a total of 6 coefficient sequences was assessed, and core coefficients that made significant contributions to classification were selected.

[0064] Specifically, the second historical samples were selected from the past three months and manually annotated (with ≥500 samples per category). Each sample contained six coefficient sequences derived from wavelet decomposition. And corresponding labels (0 = well sealed, 1 = suspected leak, 2 = leaking and unqualified). A typical sequence (center sequence) is constructed for each label, and the Euclidean distance between all sample coefficient sequences and the corresponding category center is calculated. The greater the difference in distance distribution between different labels (e.g., small distance for leaking samples, large distance for normal samples), the stronger the discriminative power of the sequence features for the label, and the higher its importance. For a single wavelet coefficient sequence (e.g., ... ), categorize all samples by label The sequence is divided into 3 sets: Well-sealed Sequence set, It is a sequence vector of length 301. Suspected leaks Sequence set; Leakage non-compliance A set of sequences.

[0065] Specifically, the average sequence of each set is calculated as the typical central sequence of that class (reflecting the sequence morphological characteristics of that class):

[0066] Central sequence (Well-sealed type):

[0067]

[0068] in, for The sample size, the first The value at each time point is All sequences in The mean of the points; The corresponding should be assigned to Well-sealed inside Sequence set;

[0069] Similarly, construct a central sequence for suspected leaks. and leakage non-compliance category center sequence .

[0070] Specifically, calculating the importance of historical coefficient features based on coefficient sequences and corresponding label center sequences includes calculating the corresponding Euclidean distance set between each coefficient sequence and its corresponding label center sequence for each second historical sample, then calculating the mean of each Euclidean distance set, obtaining the total standard deviation of all distances, and using the maximum class distance difference and the total standard deviation to represent the importance of the corresponding coefficient features. Determining the core coefficient sequences and obtaining corresponding historical feature vectors involves sorting the coefficient feature importance of each coefficient sequence from high to low, selecting the top three sequences as core coefficient sequences, and concatenating the core coefficient sequences to form the corresponding historical feature vectors. Training the model based on the historical feature vectors involves fitting the model using the historical feature vectors, enabling the model to learn the mapping relationship between the core coefficient sequences and the airtightness tags of the electricity meter.

[0071] For each sample For each sequence, calculate its Euclidean distance to the center sequence of its corresponding label (the smaller the distance, the more similar the sequence is to the typical morphology of that type), and obtain the distance set. Similarly, we can obtain ( Samples and (distance) and ( Samples and (distance). Calculate the mean of the distance sets for each label. Similarly, we can obtain , Calculate the total standard deviation of all distances (reflecting overall fluctuation). The ability to distinguish between classes is quantified by the ratio of the maximum class distance difference to the total standard deviation (the higher the ratio, the more significant the distribution difference).

[0072]

[0073] For the remaining 5 wavelet coefficient sequences Repeat the steps to calculate the importance of each historical coefficient feature; sort the historical coefficient features from high to low importance, and select the top 3 sequences as the core coefficient sequences (which contribute the most to label differentiation).

[0074] Specifically, the core coefficient sequence is obtained based on the importance of each historical coefficient feature, and then the historical feature vector is obtained. The vector machine model is then trained to obtain the trained vector machine model.

[0075] Specifically, an automated airtightness determination is achieved by training a vector machine model based on the selected core coefficient features. The specific process is as follows: For historical samples, the three core coefficient sequences selected in the above steps are extracted, and the three core coefficient sequences are directly concatenated to form a historical feature vector; a radial basis function (RBF kernel) is used, and the hyperparameters (penalty coefficient C, kernel function parameters) are optimized through grid search. The training set is used to fit the model, enabling it to learn the mapping relationship between core coefficient features and airtightness status. The goal is to minimize classification error and improve generalization ability. For the current data to be detected, the corresponding... The core coefficient sequence is selected based on the historical importance score. The feature vector formed by the three core sequences is input into the pre-trained vector machine model. The model outputs the classification result (0=good sealing, 1=suspected leakage, 2=leakage failure), realizing the automatic judgment of the airtightness of the electricity meter.

[0076] In fact, this embodiment can integrate the testing process with the production line to achieve automated data collection, processing and analysis, effectively improving the speed and efficiency of testing.

[0077] This application also discloses an energy meter airtightness data analysis system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, an energy meter airtightness data analysis method according to this application is implemented.

[0078] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0079] In this application, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as resistive random access memory (DRAM), dynamic random access memory (DRAM), static random access memory (SRAM), etc., or any other medium that can be used to store desired information and can be accessed by an application program, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device.

[0080] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for analyzing the airtightness data of an electricity meter, characterized in that, Including the following steps: Construct and train a network model for analyzing the airtightness data of electricity meters. The input is the current data of the electricity meter and the current basic information of the electricity meter. The output is the corresponding airtightness label. Simultaneously acquire raw temperature and pressure data from the electricity meter and preprocess them to obtain temperature and standardized pressure sequences; select the first historical sample that has been labeled and confirmed to be sealed, establish a historical temperature segmentation model library based on the first historical sample, and extract residual sequences based on the preprocessed data; select the second historical sample with different labeled categories, calculate the historical coefficient feature importance based on the decomposition coefficient of the second historical sample and the corresponding label to obtain the corresponding historical feature vector, and train the model based on the historical feature vector. The current feature vector is obtained by solving the residual sequence, and the airtightness of the current electricity meter is automatically determined by combining the pre-trained model. The process of establishing a historical temperature segmentation model library and extracting residual sequences based on preprocessed data includes: selecting a first historical sample labeled as well-sealed to determine temperature segments, establishing a first-order linear model for each temperature segment to extract model parameters, and then obtaining the segmentation model library; matching temperature segments according to the current temperature sequence, and calling the corresponding model parameters to obtain residual sequences; The step of calculating the importance of historical coefficient features based on the decomposition coefficients of the second historical samples and the corresponding labels to obtain historical feature vectors includes: selecting second historical samples with different labels to decompose and extract second historical sample coefficient sequences; dividing each coefficient sequence into a label set based on the second historical sample labels and obtaining the label center sequence; calculating the importance of historical coefficient features based on the coefficient sequences and the corresponding label center sequences; and then determining the core coefficient sequences to obtain the corresponding historical feature vectors.

2. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The preprocessing to obtain the temperature sequence and standardized pressure sequence includes: removing outliers from the original temperature data and smoothing the temperature data to obtain the temperature sequence; collecting the original pressure data within the normal range and smoothing it, and then performing standardization processing to obtain the standardized pressure sequence; aligning the temperature sequence and the standardized pressure sequence with timestamps to form an ordered data pair.

3. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The process of selecting a well-sealed first historical sample to determine temperature segments includes preprocessing the selected first historical sample, calculating a local covariance sequence based on the preprocessed data according to a fixed temperature range window, obtaining a covariance difference sequence between adjacent temperature windows, and obtaining N temperature segments based on the covariance difference sequence and a segmentation threshold; the segmentation threshold is taken as the 95th percentile value of the difference sequence.

4. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, In the segmented model library, each temperature segment is bound to its corresponding model parameter and stored.

5. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The step of calling the corresponding model parameters to obtain the residual sequence includes taking each data point in the current temperature sequence, inputting the first-order linear model parameters corresponding to its temperature segment, obtaining the temperature-driven pressure trend value, traversing the temperature sequence to obtain the pressure trend sequence, and calculating the corresponding data difference based on the current standardized pressure sequence and the pressure trend sequence to form the residual sequence.

6. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The step of dividing each coefficient sequence into a label set and obtaining the label center sequence based on the second historical sample label includes dividing a single coefficient sequence into different label sets according to the second historical sample label, and calculating the average sequence of each set as its label center sequence.

7. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The calculation of historical coefficient feature importance based on coefficient sequence and corresponding label center sequence includes calculating the corresponding Euclidean distance set between each coefficient sequence and its corresponding label center sequence for each second historical sample, then calculating the mean of each Euclidean distance set, obtaining the total standard deviation of all distances, and using the maximum class distance difference and the total standard deviation to represent the corresponding coefficient feature importance.

8. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The process of determining the core coefficient sequence and obtaining the corresponding historical feature vector includes sorting the coefficient features corresponding to each coefficient sequence from high to low importance, selecting the top three sequences as the core coefficient sequences, and concatenating the core coefficient sequences to form the corresponding historical feature vector.

9. The method for analyzing the airtightness data of an electricity meter according to claim 1, characterized in that, The process of training the model based on historical feature vectors includes fitting the model using historical feature vectors, enabling the model to learn the mapping relationship between the core coefficient sequence and the airtightness label of the electricity meter.

10. A data analysis system for the airtightness of an electricity meter, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the method for analyzing the airtightness data of an electricity meter according to any one of claims 1-9.

Images