A method and system for detecting faults in an electricity metering chip

Abstract

The present application relates to the field of power metering fault detection, and more particularly to a kind of fault detection method and system of electric meter metering chip, comprising: obtaining the three-phase voltage and current analog signal of electric meter metering chip, extract features to construct window feature vector and label operating state, form history sample set.Set feature interval number range, generate candidate partition mode, carry out feature discretization and interval block construction, mapping sample and classification, calculate balance and select executable partition mode;Pre-set wavelet base is carried out wavelet decomposition to special interval block, assess degree of discrimination to determine optimal partition mode;Wave feature vector is extracted to special interval block, and exclusive logic regression model is trained.Real-time data acquisition, calculate window feature vector, match interval block to determine fault, realize electric meter metering chip fault detection.The present application improves the accuracy, reliability and real-time performance of electric meter metering chip fault detection through feature discretization, wavelet decomposition and exclusive model training.

Description

Technical Field

[0001] This invention relates to the field of power metering fault detection. In particular, it relates to a method and system for detecting faults in power metering chips. Background Technology

[0002] The metering chip is a core component of an electricity meter, primarily used for the accurate measurement of electricity consumption. It collects signals such as voltage and current, and calculates the amount of electricity used based on these signals, thus achieving precise metering. The performance of the metering chip directly determines the accuracy and reliability of electricity metering, which is crucial for the stable operation of the power system and the fairness of electricity trading. With the development of smart grids, metering chips not only need to possess high-precision metering functions but also need to support multiple communication protocols to enable remote meter reading and data transmission, further improving the intelligent management level of the power system.

[0003] A malfunction in the metering chip can lead to deviations in electricity metering data, affecting billing for electricity users and the operational efficiency of the power system. Timely and accurate fault detection allows for rapid chip repair or replacement, reducing metering errors, ensuring the fairness and reliability of electricity metering, and protecting the legitimate rights and interests of both electricity users and power supply companies.

[0004] When detecting faults in electricity meter chips, fixed wavelet bases or filtering algorithms are typically used, which are difficult to handle complex harmonic interference in industrial scenarios. This leads to confusion between interference signals and normal signals, resulting in low feature extraction accuracy and an inability to accurately reflect the true state of electricity. In addition, the lack of stratified processing of samples according to feature distribution and the direct use of a unified model for judgment when normal and faulty samples are mixed can easily lead to misjudgment of "difficult-to-distinguish samples" generated under weak or complex interference conditions, affecting the accuracy and reliability of fault detection. Summary of the Invention

[0005] To address the problems in the fault detection of electricity meter chips, where traditional methods are unable to cope with complex harmonic interference and lack sample stratification, resulting in low feature extraction accuracy, easy misjudgment, and difficulty in distinguishing samples, thus affecting the accuracy and reliability of detection, this invention provides solutions in the following aspects.

[0006] In a first aspect, a method for fault detection of an electricity meter chip includes: acquiring three-phase voltage analog signals and three-phase current analog signals of the electricity meter chip within a preset window length, extracting several features to form a window feature vector, labeling the operating status of the electricity meter chip within the preset window length, and constructing a historical sample set; setting the range of interval values ​​for each feature, generating several sets of candidate partitioning methods, discretizing feature intervals and constructing interval blocks for each candidate partitioning method, mapping samples in the historical sample set to corresponding interval blocks, classifying interval blocks according to rules, calculating the balance of each candidate partitioning method, filtering the set of executable partitioning methods based on the balance, and obtaining the corresponding intervals. Interval block information, where the interval blocks are classified into: normal interval blocks, abnormal interval blocks, and special interval blocks; wavelet decomposition is performed on the special interval blocks under each set of executable partitioning methods to calculate the total harmonic content and evaluate the discriminative power of the wavelet basis. The executable partitioning method with the highest total discriminative power is selected as the optimal partitioning method, and the corresponding interval block data is determined; wavelet feature vectors are extracted for each special interval block using wavelet decomposition to construct a dedicated model sample set and train a dedicated logistic regression model to obtain a significant logistic regression model; real-time data is acquired, real-time window feature vectors are calculated, and the window feature vectors are matched with the interval blocks under the optimal partitioning method to determine whether a fault exists, thus completing the fault detection of the electricity metering chip.

[0007] Preferably, the interval block construction step includes:

[0008] The maximum and minimum values ​​of each feature in the historical sample set are counted to determine the range of values ​​for each feature. Based on the range of values, the range of values ​​for each feature is divided into several continuous and non-overlapping discretized intervals at equal intervals. The discretized intervals of each feature are combined by Cartesian product to form interval blocks in multidimensional space. Each feature interval combination corresponds to a unique interval block.

[0009] Preferably, the classification of interval blocks according to rules includes:

[0010] The samples in the historical sample set are mapped to the corresponding interval blocks. Interval blocks with 0 samples are removed. For each interval block, the label distribution of all samples contained therein is calculated, where the labels include: 1 for normal and 0 for fault.

[0011] A block is marked as normal if all samples within it are labeled 1, and as abnormal if all samples within it are labeled 0. A block is marked as special if it contains both samples labeled 1 and samples labeled 0.

[0012] Preferably, the method for calculating the balance of the candidate partitioning methods includes:

[0013] Calculate the entropy of the number of samples in all special interval blocks under each candidate partitioning method, and sum them as the balance of the candidate partitioning method to quantify the uniformity of sample distribution among special interval blocks.

[0014] Preferably, the set of executable partitioning methods based on balance selection includes:

[0015] Candidate partitioning methods with balance less than or equal to the balance threshold are discarded; conversely, candidate partitioning methods with balance greater than the balance threshold are formed into a set of executable partitioning methods.

[0016] Preferably, selecting the executable partitioning method with the highest total distinguishability as the optimal partitioning method further includes:

[0017] Several wavelet bases are preset, and special sample sets corresponding to all special interval blocks in any executable partitioning method are obtained. Wavelet decomposition is performed on each sample in the special interval block according to the preset number of layers. The effective values ​​of the fundamental wave and each harmonic are calculated, and the total harmonic content of voltage and current of all samples in the special interval block is obtained.

[0018] The special sample set is divided into normal samples and abnormal samples. The sum of the absolute values ​​of the differences in the total harmonic content between normal samples and abnormal samples is calculated to obtain the discrimination of each special interval block. The maximum value of each discrimination is selected as the optimal discrimination of the special interval block. The optimal discrimination of all special interval blocks under any executable partitioning method is summed to obtain the total discrimination of any executable partitioning method. The partitioning method with the largest total discrimination is selected as the optimal partitioning method.

[0019] Preferably, the training steps of the significant logistic regression model include:

[0020] Wavelet decomposition is performed on the original waveforms of all samples within a special interval block to extract the effective values ​​of the fundamental voltage wave, the effective values ​​of the fundamental current wave, the total harmonic content of the voltage wave, and the total harmonic content of the current wave as features. Maximum and minimum value normalization is then performed to construct wave feature vectors.

[0021] The wave feature vector and corresponding label of each sample are used to form a special model sample set for a specific interval block. Based on the special model sample set, a special logistic regression model is trained. The wave feature vector is used as input and the output is the probability of the sample label. The model parameters are optimized by using the cross-entropy loss function and gradient descent method to complete the model training and obtain a significant logistic regression model.

[0022] Preferably, the step of determining whether a fault exists includes:

[0023] If the real-time window feature vector matches a normal interval block, it is determined that there is no fault; if it matches an abnormal interval block, it is determined that there is a fault; if it matches a special interval block, wavelet decomposition is performed using the optimal wavelet basis, the real-time wavelet feature vector is extracted and input into the significant logistic regression model, and the normal probability is output. If the normal probability is greater than the preset probability, it is determined that there is no fault; otherwise, if the normal probability is less than or equal to the preset probability, it is determined that there is a fault.

[0024] Preferably, the method for obtaining the window feature vector includes:

[0025] Based on the three-phase voltage simulation signal and the three-phase current simulation signal, the difference between the maximum and minimum values ​​of the voltage waveform of each phase is calculated, and the average value of all phase differences is calculated to obtain the three-phase average voltage amplitude.

[0026] The three-phase voltage instantaneous value sequence is decomposed using the symmetrical component method to obtain a positive-sequence voltage component and a negative-sequence voltage component. The ratio of the positive-sequence voltage component to the negative-sequence voltage component is calculated and converted into a percentage to obtain the three-phase voltage imbalance.

[0027] Calculate the difference between the maximum and minimum values ​​of the current waveform for each phase, and calculate the average of all phase differences to obtain the three-phase average current amplitude.

[0028] The three-phase current instantaneous value sequence is decomposed by the symmetrical component method to obtain a positive sequence current component and a negative sequence current component. The ratio of the positive sequence current component to the negative sequence current component is calculated and converted into a percentage to obtain the three-phase current imbalance.

[0029] For the voltage and current waveforms of a single phase, record the time when the voltage crosses zero and the time when the current crosses zero. Calculate the difference between the two times and divide it by the preset window length to obtain the phase difference of a single phase. Take the average of the three phase differences to obtain the average phase difference of the voltage-current three phases.

[0030] Using the three-phase average amplitude of voltage, the three-phase unbalance of voltage, the three-phase average amplitude of current, the three-phase unbalance of current, and the three-phase average phase difference between voltage and current as feature dimensions, a window feature vector is constructed. Each window corresponds to one sample and includes the window feature vector and a label.

[0031] Secondly, an electricity meter chip fault detection 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 above-mentioned electricity meter chip fault detection method is implemented.

[0032] The present invention has the following effects:

[0033] 1. This invention, by introducing feature interval discretization and interval block classification, combined with wavelet decomposition and total harmonic content calculation, can effectively cope with complex harmonic interference and improve the accuracy of feature extraction. Simultaneously, by performing stratified processing of samples, it avoids directly using a unified model for judgment when normal and faulty samples are mixed, reducing misjudgments of difficult-to-distinguish samples and significantly improving the accuracy and reliability of fault detection.

[0034] 2. This invention employs a balanced selection method for executable partitioning to ensure uniform distribution of samples across specific interval blocks, thereby improving the model's adaptability to different situations. Furthermore, by dynamically selecting the optimal wavelet basis and training a dedicated logistic regression model, it can better adapt to different feature distributions and fault modes, enhancing the model's generalization ability.

[0035] 3. This invention achieves accurate classification of each special interval block by constructing a dedicated model sample set and training a dedicated logistic regression model, reducing false positives and false negatives, quickly calculating real-time window feature vectors and matching interval blocks under the optimal partitioning method, and can quickly determine whether a fault exists, thus improving the efficiency and real-time performance of fault detection. Attached Figure Description

[0036] Figure 1 This is a flowchart of steps S1-S5 in a method for detecting faults in an electricity meter chip according to an embodiment of the present invention.

[0037] Figure 2 This is a structural block diagram of an electricity metering chip fault detection system according to an embodiment of the present invention. Detailed Implementation

[0038] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0039] Reference Figure 1 A method for detecting faults in an electricity meter's metering chip includes steps S1-S5, as detailed below:

[0040] S1: Obtain the three-phase voltage analog signal and three-phase current analog signal of the metering chip within the preset window length, extract several features to form a window feature vector, set labels for the operating status of the metering chip within the preset window length, and construct a historical sample set.

[0041] Based on three-phase voltage and current analog signals, including: acquiring the A-phase voltage waveform, B-phase voltage waveform, C-phase voltage waveform, A-phase current waveform, B-phase current waveform, and C-phase current waveform, the following features are extracted to construct a window feature vector:

[0042] Calculate the difference between the maximum and minimum values ​​of the voltage waveforms for phases A, B, and C respectively, and then calculate the average of these three differences to obtain the three-phase average voltage amplitude. This reflects the average amplitude level of the three-phase voltage and helps to assess the stability and fluctuation of the voltage.

[0043] The instantaneous three-phase voltage sequence is decomposed using the symmetrical component method to obtain positive-sequence and negative-sequence voltage components. The ratio of the positive-sequence to the negative-sequence voltage components is calculated and converted into a percentage to obtain the three-phase voltage unbalance. This reflects the degree of balance of the three-phase voltage; a higher unbalance indicates poorer symmetry of the three-phase voltage, which may affect the normal operation of the equipment.

[0044] Calculate the difference between the maximum and minimum values ​​of the current waveforms for phases A, B, and C respectively, and then calculate the average of these three differences to obtain the three-phase average current amplitude. This reflects the average amplitude level of the three-phase current and helps to assess the stability and fluctuation of the current.

[0045] The instantaneous value sequence of three-phase current is decomposed using the symmetrical component method to obtain the positive-sequence and negative-sequence current components. The ratio of the positive-sequence to the negative-sequence current components is calculated and converted into a percentage to obtain the three-phase current imbalance. This reflects the degree of balance of the three-phase current; a higher imbalance indicates poorer symmetry of the three-phase current, which may affect the normal operation of the equipment.

[0046] For the voltage and current waveforms of a single phase, record the times when the voltage and current cross zero. Calculate the difference between the two times and divide it by a preset window length to obtain the phase difference of a single phase. Take the average of the three-phase phase differences to obtain the average phase difference between voltage and current. This reflects the phase relationship between voltage and current and helps to evaluate the power factor and the load characteristics of the circuit.

[0047] The three-phase average amplitude of voltage, the three-phase unbalance of voltage, the three-phase average amplitude of current, the three-phase unbalance of current, and the three-phase average phase difference between voltage and current are combined into a window feature vector. Each window corresponds to a sample and includes the window feature vector and a label.

[0048] S2: Set the range of interval numbers for each feature, generate several sets of candidate partitioning methods, discretize the feature intervals and construct interval blocks for each candidate partitioning method, map the samples in the historical sample set to the corresponding interval blocks, classify the interval blocks according to the rules, calculate the balance of each candidate partitioning method, filter the set of executable partitioning methods based on the balance, and obtain the corresponding interval block information.

[0049] The steps for constructing a range block include:

[0050] The maximum and minimum values ​​of each feature in the historical sample set are counted to determine the range of values ​​for each feature. Based on the range of values, the range of values ​​for each feature is divided into several continuous and non-overlapping discretized intervals at equal intervals. The discretized intervals of each feature are combined by Cartesian product to form interval blocks in multidimensional space. Each feature interval combination corresponds to a unique interval block.

[0051] Specifically, suppose we have a feature The maximum value in the historical sample set is 100, and the minimum value is 0. We set the number of intervals to 5, then the width of each interval is... Therefore, features The range of values ​​is divided into the following 5 intervals: Interval 1: Interval 2: Interval 3: Interval 4: Interval 5: .

[0052] If features The value is 35. Based on the above division, the feature... It falls in interval 2: Within this, continuous feature values ​​can be converted into discrete interval labels.

[0053] The interval blocks are classified according to the rules, including:

[0054] The samples in the historical sample set are mapped to the corresponding interval blocks. Interval blocks with 0 samples are removed. For each interval block, the label distribution of all samples contained therein is calculated, where the labels include: 1 for normal and 0 for fault.

[0055] A block is marked as normal if all samples within it are labeled 1, and as abnormal if all samples within it are labeled 0. A block is marked as special if it contains both samples labeled 1 and samples labeled 0.

[0056] The methods for calculating balance include:

[0057] Calculate the entropy of the number of samples in all special interval blocks under each candidate partitioning method, and sum them as the balance of the candidate partitioning method to quantify the uniformity of sample distribution among special interval blocks.

[0058] Specifically, the balance satisfies the following relationship:

[0059] ;

[0060] in, It represents the balance of the current candidate partitioning method, that is, the entropy of the number of samples corresponding to multiple special interval blocks; This represents the total number of special interval blocks in the interval block combination corresponding to the current candidate partitioning method. This represents the total number of samples in all special interval blocks within the interval block combinations corresponding to the current candidate partitioning method. This indicates the current candidate partitioning method corresponding to the first... The number of samples in a specific interval block. This represents an exponential function with a base of 2.

[0061] Entropy is used to calculate the balance of the current candidate partitioning. A higher entropy value indicates a more uniform distribution of samples across different blocks; a lower entropy value indicates that samples are concentrated in some blocks and sparsely distributed in others.

[0062] Candidate partitioning methods with balance less than or equal to the balance threshold are discarded; conversely, candidate partitioning methods with balance greater than the balance threshold are formed into a set of executable partitioning methods.

[0063] For example, the balance threshold is 0.8, which can be adjusted according to specific circumstances.

[0064] S3: Perform wavelet decomposition on the special interval blocks under each set of executable partitioning methods, calculate the total harmonic content, evaluate the discriminative power of the wavelet basis, select the executable partitioning method with the highest total discriminative power as the optimal partitioning method, and determine the corresponding interval block data.

[0065] Example 1: Taking any partitioning method in the set of executable partitioning methods as an example, that is, one partitioning method corresponds to one special sample set, and one special sample set contains all special interval blocks. Choose any wavelet basis, and perform wavelet decomposition on each sample in the special interval block according to a preset number of layers. In this example, it is set to 5 layers, which can be adjusted according to specific circumstances. Calculate the effective values ​​of the fundamental wave and each harmonic to obtain the total harmonic content of voltage and current of all samples in the special interval block.

[0066] The special interval blocks are divided into normal samples and abnormal samples. The sum of the absolute values ​​of the differences in the total harmonic content between normal samples and abnormal samples is calculated to obtain the discrimination degree of the special interval blocks. The sum of the discrimination degrees of all special interval blocks under each executable partitioning method is calculated to obtain the total discrimination degree of each executable partitioning method. The executable partitioning method with the largest total discrimination degree is selected as the optimal partitioning method.

[0067] Example 2: Several wavelet bases are preset. For example, the wavelet bases can be selected as: db4 (Daubechies4, fourth-order Daubechies wavelet base), db8 (Daubechies8, eighth-order Daubechies wavelet base), sym4 (Symlet4, fourth-order Symlet wavelet base), coif3 (Coiflet3, third-order Coiflet wavelet base), etc. Similarly, taking any partitioning method in the set of executable partitioning methods as an example, one partitioning method corresponds to one special sample set. One special sample set contains all special interval blocks. Wavelet analysis is performed on the special interval blocks using different wavelet bases. Specifically, the number of wavelet decomposition layers is preset. In this example, it is set to 5 layers, which can be adjusted according to the specific situation. Each sample in the special sample set is decomposed into 5 layers of wavelet decomposition, and the effective values ​​of the fundamental wave and each harmonic are calculated to obtain the total harmonic content of voltage and current.

[0068] The special sample set is divided into normal samples and abnormal samples. The sum of the absolute values ​​of the differences in the total harmonic content between normal samples and abnormal samples is calculated to obtain the discrimination of each special interval block. The maximum value of each discrimination is selected as the optimal discrimination of the special interval block. The optimal discrimination of all special interval blocks under any executable partitioning method is summed to obtain the total discrimination of any executable partitioning method.

[0069] The partitioning method with the highest overall discrimination is selected as the optimal partitioning method, and the corresponding interval block data is recorded, including special interval block combinations, the optimal wavelet basis for a single special interval block, and special sample sets.

[0070] S4: Use wavelet decomposition to extract wavelet feature vectors for each special interval block, construct a dedicated model sample set, and train a dedicated logistic regression model to obtain a significant logistic regression model.

[0071] For each special interval block, the optimal wavelet basis is used, and wavelet decomposition is performed on the six original waveforms of all samples in the special interval block. The six waveforms refer to the A-phase voltage waveform, B-phase voltage waveform, C-phase voltage waveform, A-phase current waveform, B-phase current waveform, and C-phase current waveform. The effective value of the fundamental voltage waveform, the effective value of the fundamental current waveform, the total harmonic content of the voltage waveform, and the total harmonic content of the current waveform are extracted as features, and the maximum and minimum values ​​are normalized to construct a 4-dimensional wave feature vector.

[0072] The wave feature vector and corresponding label of each sample are used to form a special model sample set for a specific interval. Based on the special model sample set, a special logistic regression model is trained. The 4-dimensional wave feature vector is used as input and the output is the probability of the sample label. The model parameters are optimized by using the cross-entropy loss function and gradient descent method to complete the model training and obtain a significant logistic regression model.

[0073] Based on a significant logistic regression model, the input wave feature vector outputs a normal probability, with the normal probability taking values ​​within a certain range. This allows for precise classification of each specific interval block.

[0074] S5: Acquire real-time data, calculate the real-time window feature vector, and match the window feature vector with the interval blocks under the optimal partitioning method to determine whether a fault exists, thus completing the fault detection of the electricity metering chip.

[0075] If the real-time window feature vector matches a normal interval block, it is determined that there is no fault; if it matches an abnormal interval block, it is determined that there is a fault; if it matches a special interval block, wavelet decomposition is performed using the optimal wavelet basis, the real-time wavelet feature vector is extracted and input into the significant logistic regression model, and the normal probability is output. If the normal probability is greater than the preset probability, it is determined that there is no fault; otherwise, if the normal probability is less than or equal to the preset probability, it is determined that there is a fault.

[0076] For example, the preset probability is 0.5, which can be adjusted according to specific circumstances.

[0077] This invention also provides a fault detection system for electricity meter measuring chips. For example... Figure 2 As shown, the system includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a method for detecting faults in an electricity meter chip according to the first aspect of the present invention. The system also includes other components well-known to those skilled in the art, such as a communication bus and a communication interface. Their configurations and functions are known in the art and will not be described further here.

[0078] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting faults in an electricity metering chip, characterized by, The method comprises the following steps: acquiring three-phase voltage analog signals and three-phase current analog signals of a power metering chip within a preset window length, extracting a plurality of features to form a window feature vector, setting a label for the running state of the power metering chip within the preset window length, and constructing a historical sample set; setting the interval number value range of each feature, generating a plurality of groups of candidate partition modes, for each candidate partition mode, performing feature interval discretization and interval block construction, mapping the samples in the historical sample set to the corresponding interval blocks, classifying the interval blocks according to a rule, calculating the balance of each candidate partition mode, screening an executable partition mode set based on the balance, and acquiring corresponding interval block information, wherein the classification of the interval blocks includes normal interval blocks, abnormal interval blocks and special interval blocks; performing wavelet decomposition on each special interval block under each executable partition mode set, calculating the total harmonic content, and evaluating the discrimination degree of the wavelet basis, selecting the executable partition mode with the maximum total discrimination degree as the optimal partition mode, and determining the corresponding interval block data; extracting a wave feature vector using wavelet decomposition for each special interval block, constructing an exclusive model sample set, and training an exclusive logistic regression model to obtain a significant logistic regression model; acquiring real-time data, calculating a real-time window feature vector, matching the window feature vector with the interval blocks under the optimal partition mode to determine whether a fault exists, and completing power metering chip fault detection; wherein the calculation method of the balance of the candidate partition mode comprises: calculating the sample number entropy of all special interval blocks under each candidate partition mode, and accumulating the sample number entropy as the balance of the candidate partition mode to quantify the uniformity of the sample distribution between the special interval blocks; the training steps of the significant logistic regression model comprise: performing wavelet decomposition on the original waveforms of all samples in the special interval block, extracting the voltage fundamental effective value, the current fundamental effective value, the voltage total harmonic content and the current total harmonic content as features, and performing maximum and minimum value normalization processing to construct a wave feature vector; constructing an exclusive model sample set of the special interval block based on the wave feature vector and the corresponding label of each sample, training an exclusive logistic regression model based on the exclusive model sample set, taking the wave feature vector as the input, and outputting the probability of the sample label, optimizing the model parameters using the cross-entropy loss function and the gradient descent method, completing the model training, and obtaining the significant logistic regression model; the step of determining whether a fault exists comprises: in response to the real-time window feature vector matching to a normal interval block, it is determined that there is no fault; matching to an abnormal interval block, it is determined that there is a fault; matching to a special interval block, performing wavelet decomposition using the optimal wavelet basis, extracting a real-time wave feature vector and inputting the significant logistic regression model, outputting a normal probability, and in response to the normal probability being greater than a preset probability, it is determined that there is no fault, otherwise, the normal probability is less than or equal to the preset probability, and it is determined that there is a fault.

2. The method of claim 1, wherein the method further comprises: the interval block construction step comprises: The maximum value and the minimum value of each feature in the historical sample set are counted, interval number value ranges of the features are determined, the value ranges of the features are uniformly divided into a plurality of continuous and non-overlapping discretization intervals according to the interval numbers, and the discretization intervals of the features are combined by Cartesian product to form interval blocks in a multi-dimensional space, and each feature interval combination corresponds to a unique interval block.

3. The method of claim 1, wherein the method further comprises: The interval blocks are classified according to rules, including: The samples in the historical sample set are mapped to corresponding interval blocks, the interval blocks with a sample number of 0 in the interval blocks are removed, and for each interval block, the label distribution of all samples contained in the interval block is counted, wherein the label includes: 1 for normal and 0 for fault; The interval blocks in which all the labels of the samples in the interval blocks are 1 are marked as normal interval blocks, the interval blocks in which all the labels of the samples in the interval blocks are 0 are marked as abnormal interval blocks, and the interval blocks in which the samples with labels of 1 and the samples with labels of 0 are contained are marked as special interval blocks.

4. The method of claim 1, wherein the method further comprises: The executable division mode set is screened based on the balance, including: In response to the balance being less than or equal to the balance threshold, the candidate division mode is discarded, otherwise, the balance is greater than the balance threshold, and the candidate division mode forms the executable division mode set.

5. The method of claim 1, wherein the method further comprises: The executable division mode with the maximum total interval degree is selected as the optimal division mode, and further includes: A plurality of wavelet bases are preset, a special sample set corresponding to the special interval blocks in any executable division mode is obtained, each sample in the special interval blocks is decomposed by wavelet according to a preset number of layers, the effective values of the fundamental wave and each harmonic wave are calculated, and the total harmonic content of the voltage and the current of all the samples in the special interval blocks is obtained; The special sample set is divided into normal samples and abnormal samples, the sum of the absolute values of the differences between the total harmonic contents of the normal samples and the abnormal samples is calculated, the interval degrees of the special interval blocks are obtained, the maximum value of the interval degrees is selected as the optimal interval degree of the special interval blocks, the optimal interval degrees of all the special interval blocks in any executable division mode are summed, the total interval degree of any executable division mode is obtained, and the division mode with the maximum total interval degree is selected as the optimal division mode.

6. The method of claim 1, wherein the method further comprises: The window feature vector is obtained in the following manner: Based on the three-phase voltage analog signal and the three-phase current analog signal, the difference between the maximum value and the minimum value of each phase voltage waveform is calculated, the mean value of all phase difference values is calculated, and the voltage three-phase average amplitude is obtained; The three-phase voltage instantaneous value sequence is decomposed by the symmetrical component method to obtain a positive sequence voltage component and a negative sequence voltage component, the ratio of the positive sequence voltage component to the negative sequence voltage component is calculated and converted into a percentage, and the voltage three-phase unbalance degree is obtained; The difference between the maximum value and the minimum value of each phase current waveform is calculated, the mean value of all phase difference values is calculated, and the current three-phase average amplitude is obtained; The three-phase current instantaneous value sequence is decomposed by the symmetrical component method to obtain a positive sequence current component and a negative sequence current component, the ratio of the positive sequence current component to the negative sequence current component is calculated and converted into a percentage, and the current three-phase unbalance degree is obtained; For the voltage waveform and the current waveform of a single phase, the voltage zero-crossing time and the current zero-crossing time are recorded, the difference between the two times is calculated and divided by a preset window length to obtain the phase difference of the single phase, the average of the phase differences of the three phases is taken to obtain the voltage-current three-phase average phase difference; The voltage three-phase average amplitude, the voltage three-phase unbalance degree, the current three-phase average amplitude, the current three-phase unbalance degree and the voltage-current three-phase average phase difference are taken as characteristic dimensions to construct a window feature vector, one window corresponding to one sample, containing the window feature vector and the label.

7. An electric metering chip failure detection system characterized by, Comprise: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, the metering chip fault detection method according to any one of claims 1-6 is realized.

Images