An electric energy meter operation state inspection and evaluation method and system

Abstract

The application discloses a kind of electric energy meter operating state inspection evaluation method and system, it is related to electric power metering safety technical field, including, the electrical working data of electric energy meter is collected and preprocessed, the reference current and working current of electric energy meter are obtained and error feature is extracted, current error signal is generated;Electric energy meter is carried out voiceprint collection, captures the voiceprint signal of ammeter;Through coding algorithm, voiceprint signal is converted into discrete pulse sequence, through dimension conversion and optimization compression, obtain standardized pulse sequence;Pulse neural network is applied to the feature extraction of standardized pulse sequence, obtains the fault probability distribution after pulse sequence feature;The application generates feedback control signal and electric energy meter evaluation report based on fault probability distribution, solves the problem of poor real-time of fault mode identification.

Description

Technical Field

[0001] This invention relates to an apparatus and method for measuring electrical and magnetic variables, specifically belonging to the field of power metering safety, and particularly to a method and system for inspecting and evaluating the operating status of an electricity meter. Background Technology

[0002] As the core equipment for electricity metering in a power system, the accurate assessment of the operating status of electricity meters directly affects the level of intelligence in electricity measurement. Existing technologies can verify and evaluate the operating status of electricity meters. For example, clustering algorithms based on random matrix theory identify anomalies through the statistical characteristics of electrical parameters such as voltage and current, improving the sensitivity of detection for common fault modes; transfer learning models, combined with cross-scenario data transfer capabilities, reduce the difficulty of model generalization between different electricity meter models.

[0003] Existing technologies have made significant progress in the inspection and evaluation of the operating status of electricity meters through multi-dimensional data fusion and algorithm optimization, but there are still shortcomings. First, the clustering algorithm based on random matrix theory in existing technologies determines the error based on the reference current, which has the problem of insufficient accuracy in the reference current detection. In addition, the transfer learning model requires a large amount of historical data as a basis, and has the pain points of error accumulation due to environmental interference and poor real-time performance of electricity meter fault mode recognition. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for inspecting and evaluating the operating status of electricity meters to solve the problems of insufficient accuracy of reference current detection and poor real-time performance of fault mode identification.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for inspecting and evaluating the operating status of an electricity meter, comprising,

[0008] Collect and preprocess the electrical operating data of the electricity meter, obtain the reference current and operating current of the electricity meter and extract error features to generate a current error signal;

[0009] The soundprint of the electricity meter is collected, and the sampling frequency and duration of the soundprint collection device are adjusted based on the current error signal to capture the soundprint signal of the electricity meter.

[0010] The voiceprint signal is converted into a discrete pulse sequence through an encoding algorithm. Based on the discrete pulse sequence, a standardized pulse sequence is obtained through dimensional transformation and optimized compression.

[0011] A spiking neural network is used to extract features from a standardized pulse sequence. After obtaining the pulse sequence features, the fault probability distribution is obtained by calculating synaptic weights and updating neuron states.

[0012] Feedback control signals and energy meter evaluation reports are generated based on the fault probability distribution.

[0013] As a preferred embodiment of the method for testing and evaluating the operating status of an electricity meter according to the present invention, the electrical operating data includes the quantum Hall voltage and the original operating current of the electricity meter.

[0014] The preprocessing includes voltage conversion of the quantum Hall voltage to obtain the reference current of the energy meter, noise reduction filtering of the original operating current to obtain the operating current of the energy meter;

[0015] The extraction of error features refers to calculating the amplitude deviation, detecting the phase deviation, and analyzing the harmonic distortion of the reference current and the operating current. After obtaining the amplitude deviation, phase deviation, and harmonic distortion, the data is integrated to generate a current error signal.

[0016] As a preferred embodiment of the energy meter operation status inspection and evaluation method of the present invention, the specific steps for adjusting the sampling frequency and duration of the acoustic fingerprint acquisition device based on the current error signal to capture the acoustic fingerprint signal of the ammeter are as follows.

[0017] The sampling frequency of the voiceprint acquisition device is set according to the amplitude deviation, and the duration of the voiceprint acquisition device is set according to the harmonic distortion.

[0018] Start the voiceprint acquisition device to obtain the initial voiceprint information of the electricity meter;

[0019] The initial acoustic signature information is adjusted based on the phase deviation to generate the acoustic signature signal of the ammeter.

[0020] As a preferred embodiment of the energy meter operation status inspection and evaluation method of the present invention, the step of converting the voiceprint signal into a discrete pulse sequence through an encoding algorithm, and obtaining a standardized pulse sequence based on the discrete pulse sequence through dimensional transformation and optimized compression, is as follows:

[0021] By using an encoding algorithm, the waveform of the voiceprint signal is scanned and dynamic threshold mutation is determined to obtain a discrete pulse sequence;

[0022] Transform the time axis dimension of the discrete pulse sequence to establish a pulse timing matrix;

[0023] The discrete pulse sequence is optimized and compressed based on the pulse timing matrix, merging adjacent pulses and deleting isolated pulses to generate a standardized pulse sequence.

[0024] As a preferred embodiment of the energy meter operation status inspection and evaluation method of the present invention, the specific steps of applying a spiking neural network to extract features from the standardized pulse sequence are as follows.

[0025] The pulse density of the standardized pulse sequence is identified by a spiking neural network to obtain the density region distribution of the standardized pulse sequence.

[0026] Extract and label the pulse timestamp differences of the standardized pulse sequence to obtain the interval feature label of the standardized pulse sequence;

[0027] Gradient calculation is performed based on standardized pulse sequences, mutation thresholds are set, mutation events are identified and marked, and a list of mutation events of the standardized pulse sequences is obtained.

[0028] The pulse sequence characteristics of the normalized pulse sequence are obtained by integrating the density region distribution, interval feature markers, and mutation event list of the normalized pulse sequence.

[0029] As a preferred embodiment of the energy meter operation status inspection and evaluation method of the present invention, the following steps are taken: after obtaining the pulse sequence features, the fault probability distribution is obtained through synaptic weight calculation and neuron state update.

[0030] Based on pulse sequence characteristics, fault weights of energy meters are calculated and synaptic current pulse sequences are generated in a pulse neural network.

[0031] A sequence of synaptic current pulses is injected into the neurons of a spiking neural network, and after triggering a membrane potential response, the neuron state is updated and a fault pulse event is emitted.

[0032] The number of fault pulse events emitted by neurons is counted and normalized to obtain the fault probability distribution of the electricity meter.

[0033] As a preferred embodiment of the energy meter operation status inspection and evaluation method of the present invention, the specific steps for generating feedback control signals and energy meter evaluation reports based on fault probability distribution are as follows:

[0034] By analyzing the numerical distribution in the failure probability distribution, a multi-level response strategy is issued based on different intervals of the numerical distribution;

[0035] Match control strategies based on the fault types in the fault probability distribution;

[0036] Encapsulate multi-level response and control strategies, generate feedback control signals, and record them as the operation log of the energy meter;

[0037] The current error signal, acoustic signal, fault probability distribution, and operation log of the encapsulated ammeter are used to obtain an energy meter evaluation report.

[0038] Secondly, the present invention provides a system for testing and evaluating the operating status of an electricity meter, comprising,

[0039] The error extraction module collects and preprocesses the electrical operating data of the energy meter, obtains the reference current and operating current of the energy meter, extracts error features, and generates a current error signal.

[0040] The voiceprint acquisition module collects the voiceprint of the electricity meter and adjusts the sampling frequency and duration of the voiceprint acquisition device based on the current error signal to capture the voiceprint signal of the electricity meter.

[0041] The sequence conversion module converts the voiceprint signal into a discrete pulse sequence through an encoding algorithm. Based on the discrete pulse sequence, it obtains a standardized pulse sequence through dimensional transformation and optimized compression.

[0042] The fault identification module uses a spiking neural network to extract features from a standardized pulse sequence. After obtaining the pulse sequence features, the fault probability distribution is obtained through synaptic weight calculation and neuron state update.

[0043] The feedback evaluation module generates feedback control signals and energy meter evaluation reports based on the fault probability distribution.

[0044] The beneficial effects of this invention are as follows: It obtains a more accurate reference current through the quantum Hall effect, replacing the traditional detection method that relies on an external reference source, thus eliminating reference drift error at the physical level; it adjusts the acoustic signature sampling parameters in real time based on the current error signal and dynamically corrects the acoustic signature signal using phase deviation, effectively suppressing environmental noise interference; it applies a pulse neural network to process the standardized pulse sequence of the energy meter, obtains the fault probability distribution, and completes rapid fault type location and identification in real time; and its multi-level response strategy based on the fault probability distribution accurately matches maintenance priorities, promptly selects and issues control strategies, solving the problem of poor real-time performance in fault mode recognition. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the method for inspecting and evaluating the operating status of electricity meters.

[0047] Figure 2 This is a schematic diagram of an electricity meter operation status inspection and evaluation system.

[0048] Figure 3 This is a flowchart of voiceprint signal conversion and processing.

[0049] Figure 4 This is a flowchart for processing spiking neural networks. Detailed Implementation

[0050] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0051] This invention relates to apparatus and methods for measuring electrical and magnetic variables, specifically belonging to the field of power metering safety, with reference to... Figures 1-4 As one embodiment of the present invention, this embodiment provides a method for testing and evaluating the operating status of an electricity meter, comprising the following steps:

[0052] S1. Collect and preprocess the electrical operating data of the electricity meter, obtain the reference current and operating current of the electricity meter, extract error features, and generate a current error signal.

[0053] S1.1 Collect the electrical operating data of the electricity meter, and preprocess the electrical operating data to obtain the reference current and operating current of the electricity meter.

[0054] Specifically, a quantum Hall current sensor is used to collect the quantum Hall voltage and raw operating current of the energy meter to obtain the electrical operating data of the energy meter; the quantum Hall voltage is converted into a digital voltage value using an analog-to-digital converter (ADC), and the reference current of the energy meter is obtained based on the principle of the quantum Hall effect; a mid-range filtering algorithm is applied to the raw operating current to eliminate instantaneous noise, and the filtering window length is set to a fixed value, for example, a continuous filtering window length of 5, and the median value of a fixed number of sampling points is taken as the specific value of the current sampling point; the noise-reduced current is output to obtain the operating current of the energy meter.

[0055] S1.2 Extract the error characteristics of the reference current and the operating current to generate a current error signal.

[0056] Specifically, the reference current and the operating current are compared point by point to calculate the absolute deviation of their amplitudes. The amplitude deviations are then obtained and arranged along the time axis to generate an amplitude deviation sequence. A Hilbert transform is performed on the waveforms of the reference current and the operating current to extract their instantaneous phases and calculate their absolute deviations. The phase deviations are then obtained and arranged along the time axis to generate a phase deviation sequence. The fundamental frequency and harmonic components of the reference current and the operating current are separated. For example, when the fundamental frequency is 50Hz, the amplitudes of the second to fifth harmonic components are extracted, corresponding to a frequency range of 100Hz to 250Hz. The fundamental amplitude and the amplitudes of each harmonic component are compared and analyzed to calculate the proportion of non-fundamental components in the current waveform (the ratio of the non-fundamental portion of the reference current and the operating current in their respective current signals), thus obtaining the harmonic distortion. The amplitude deviation sequence, phase deviation sequence, and harmonic distortion are aligned by timestamps to generate a current error signal.

[0057] S2. Collect the acoustic signature of the electricity meter, and adjust the sampling frequency and duration of the acoustic signature collection device based on the current error signal to capture the acoustic signature signal of the electricity meter.

[0058] S2.1 Set the sampling frequency of the voiceprint acquisition device according to the amplitude deviation, and set the duration of the voiceprint acquisition device according to the harmonic distortion.

[0059] Specifically, a high-sensitivity microphone array is used as the voiceprint acquisition device to traverse the amplitude deviation sequence, extract the maximum instantaneous value, locate the time point at which the maximum instantaneous value is obtained, and obtain the specific value of the operating current at the time point at which the maximum instantaneous value is obtained; based on the basic operating parameters of the electricity meter, the highest operating frequency of the electricity meter is obtained (for example, for a common electricity meter, the highest operating frequency is 500Hz under the working condition of 80% of the rated working load), and the sampling frequency is set according to the highest operating frequency using the Nyquist sampling theorem.

[0060] It should be noted that the Nyquist sampling theorem requires the sampling frequency to be greater than or equal to twice the highest operating frequency. For example, if the highest operating frequency is 500Hz, then the sampling frequency should be set to 1000Hz.

[0061] Furthermore, a distortion threshold is set. For example, in power systems, a harmonic distortion of less than or equal to 5% is widely accepted as the normal range for power grid signal quality. The duration of the voiceprint acquisition device is set according to the harmonic distortion. During a single measurement, when the harmonic distortion is less than the distortion threshold, the default voiceprint scanning time is used. When the harmonic distortion is greater than or equal to the distortion threshold, the voiceprint scanning time is extended. For example, the default voiceprint scanning time of the Shure SM81 condenser microphone array is 5 seconds. When the harmonic distortion is greater than or equal to the distortion threshold, the voiceprint scanning time is extended to 10 seconds.

[0062] S2.2 Start the voiceprint acquisition device to obtain the initial voiceprint information of the electricity meter;

[0063] Specifically, a high-sensitivity microphone array is placed on the surface of the electricity meter, the sampling frequency and duration are set, the device is started and the voiceprint information is recorded, the original voiceprint waveform data is recorded, and the initial voiceprint information of the electricity meter is obtained.

[0064] S2.3 Adjust the initial acoustic signature information based on the phase deviation to generate the acoustic signature signal of the ammeter.

[0065] Specifically, the original acoustic signature signal is subjected to time-domain filtering. A fixed-length Hanning window is used to window the original acoustic signature signal (e.g., a fixed length of one short frame, frame length of N points, N=2048) to remove noise from the original acoustic signature signal. A fast Fourier transform is used to convert the denoised original acoustic signature signal into a frequency domain signal to obtain the frequency domain original acoustic signature signal. The harmonic components and fundamental components of the frequency domain original acoustic signature signal are separated by a low-pass filter (LPF) to obtain the separated fundamental component. The phase data of each harmonic of the frequency domain original acoustic signature signal is extracted. Based on the phase deviation sequence obtained by S1, the negative direction of the values ​​in the phase deviation sequence is used as the compensation direction. The values ​​in the phase deviation sequence are linearly superimposed onto the phase data of each harmonic of the frequency domain original acoustic signature signal after aligning the time axis to obtain the harmonic components after superimposing the phase data. The harmonic components after superimposing the phase data and the separated fundamental component are resynthesized into a time domain signal to obtain the acoustic signature signal of the ammeter.

[0066] S3. Convert the voiceprint signal into a discrete pulse sequence using an encoding algorithm. Based on the discrete pulse sequence, obtain a standardized pulse sequence through dimensional transformation and optimized compression.

[0067] S3.1. Through the encoding algorithm, the waveform of the voiceprint signal is scanned and dynamic threshold change is determined to obtain the discrete pulse sequence;

[0068] Specifically, the time-domain waveform of the voiceprint signal is scanned point by point to extract the amplitude value of each sampling point, and the phase deviation between adjacent points is read (the phase deviation comes from the phase deviation sequence, specifically the value of the phase deviation sequence at this sampling point); based on the local amplitude and phase deviation, the average amplitude within the sampling point interval is taken to obtain the local average amplitude, and the dynamic threshold is adjusted using the sliding window method, expressed as:

[0069] ;

[0070] in, It is a dynamic threshold. It is the local average amplitude. This is a weighting factor; for example, in the testing and evaluation of the DD862 single-phase mechanical energy meter, a typical value is 1.5. It's a phase deviation.

[0071] Furthermore, when the local average amplitude is greater than or equal to the dynamic threshold of the sampling point, the sampling point is determined to be a pulse event, and the timestamp and amplitude value are recorded to form a single-point pulse signal. When the local average amplitude is less than or equal to the dynamic threshold of the sampling point, no recording or output is performed. All determined pulse events are sorted by timestamp, and the single-point pulse signals are merged on the time axis to form a discrete pulse sequence.

[0072] S3.2. Transform the time axis dimension of the discrete pulse sequence and establish the pulse timing matrix;

[0073] Specifically, the timestamps of the discrete pulse sequence are mapped to a normalized time axis, and the time axis is normalized to complete the transformation of the time axis dimension.

[0074] Specifically, all timestamps are extracted from the discrete pulse sequence, which record the occurrence time of each pulse event; the total duration of the signal is calculated, which is the difference between the earliest and latest timestamps in the discrete pulse sequence; the original timestamps are mapped to a unified time axis range; and each timestamp is normalized to the [0,1] interval by calculating the ratio of the difference between the current timestamp and the minimum timestamp to the total duration of the signal, thus obtaining the discrete pulse sequence on the normalized time axis, and aligning the time axis of all pulse events.

[0075] Furthermore, the discrete pulse sequence on the normalized time axis is traversed, and the time interval between adjacent pulse events is read. For example, if the time of one pulse event is 0.01, and the time of the next pulse event is 0.03 according to the time progression, then the time interval between these two adjacent pulse events is 0.02. At the same time, the amplitude difference between adjacent pulse events is read, and all time intervals and amplitude differences are combined according to the time axis sequence to establish a pulse time sequence matrix.

[0076] It should be noted that each row of the pulse timing matrix contains a pair of intervals and differences. For example, if the discrete pulse sequence contains 200 events, the pulse timing matrix contains 199 rows of data. The pulse timing matrix is ​​used to quantify the timing characteristics of pulse events.

[0077] S3.3. Based on the pulse timing matrix, the discrete pulse sequence is optimized and compressed by merging adjacent pulses and deleting isolated pulses to generate a standardized pulse sequence.

[0078] Specifically, statistical calculations are performed on the time interval data and amplitude difference data in the pulse timing matrix to obtain the average value and standard deviation of the time intervals, and the maximum value of the amplitude difference is extracted. The discrete pulse sequence is optimized and compressed by setting a time interval merging threshold and an amplitude difference elimination threshold. The time interval merging threshold is set as the sum of the average value and standard deviation of the time intervals, and the amplitude difference elimination threshold is a fixed proportion of the maximum value of the amplitude difference.

[0079] It should be noted that in power systems, data with time intervals lower than the sum of the average and standard deviation of the time intervals are generally considered to be abnormal time intervals that significantly deviate from the normal range. Abnormal time intervals may be caused by clock drift, communication delays, or external interference. Therefore, a time interval merging threshold is set as the sum of the average and standard deviation of the time intervals. Pulses with time intervals lower than the time interval merging threshold need to be merged to avoid data distortion. The maximum amplitude difference usually corresponds to extreme fluctuations in the power system. In order to retain the main signal characteristics while filtering noise and abnormal fluctuations, an amplitude difference rejection threshold is set as a fixed percentage of the maximum amplitude difference. Pulse events with amplitude differences lower than the amplitude difference rejection threshold are rejected. For example, the amplitude difference rejection threshold is set to 20% of the maximum amplitude difference.

[0080] Furthermore, two adjacent pulse events with a time interval lower than the time interval merging threshold are merged, and pulse events with an amplitude difference lower than the amplitude difference removal threshold are marked as isolated pulses and removed. Two adjacent pulses are merged to obtain a new pulse. The timestamp of the new pulse is the average of the timestamps of the two adjacent pulses, and the amplitude is the average of the amplitudes of the two adjacent pulses. Removal directly deletes isolated pulses from the discrete pulse sequence and shifts the time axis to fill the missing timestamps.

[0081] Furthermore, after optimization and compression, the length of the discrete pulse sequence will change. The merged pulse events and the pulse events that were not deleted will be reordered according to their timestamps and normalized again to form a standardized pulse sequence.

[0082] S4. Apply a spiking neural network to extract features from the standardized pulse sequence. After obtaining the pulse sequence features, calculate the synaptic weights and update the neuron state to obtain the fault probability distribution.

[0083] S4.1. Use a spiking neural network to identify the pulse density of the standardized pulse sequence and obtain the density region distribution of the standardized pulse sequence;

[0084] Specifically, the process of constructing a spiking neural network includes: constructing the input layer of the spiking neural network using an input neuron layer, setting the number of neurons in the input layer to be consistent with the dimension of the historical binary matrix, and setting the initial membrane potential and resting potential of the input layer to zero; adding a time-stamped convolutional layer, setting the kernel size of the convolutional layer, for example, setting the kernel size to 3, the stride to 1, and setting the weights of the convolutional layer to small random numbers; adding a max pooling layer after the convolutional layer, setting the pooling window size, for example, the pooling window size to 2, the stride to 2; using a LIF neuron model, constructing the intermediate and output layers by adding fully connected layers, and setting the number of neurons, for example, setting the number of neurons in the intermediate layer to 100 and the number of neurons in the output layer to 10, thus completing the construction of the spiking neural network;

[0085] Historical voiceprint signals from electricity meters under normal operating conditions were collected using voiceprint acquisition devices under various working environments and conditions. These signals were then converted into historical discrete pulse sequences using an encoding algorithm. The historical discrete pulse sequences were further transformed and compressed to obtain standardized historical pulse sequences. By intervening in the electricity meter's operating state, various typical faults (such as battery undervoltage, power failure, metering inaccuracy, and electricity theft) were simulated. Voiceprint information generated by the electricity meter under different fault conditions was collected using voiceprint acquisition devices to obtain abnormal voiceprint signals, which included the correspondence between different typical faults and voiceprint information. These abnormal voiceprint signals were converted into abnormal discrete pulse sequences using an encoding algorithm. The abnormal discrete pulse sequences were then transformed and compressed to obtain standardized abnormal pulse sequences. Finally, the historical standardized pulse sequences and the abnormal standardized pulse sequences were compiled into a training dataset.

[0086] The training dataset is divided into a training set and a validation set. During each training iteration, a batch of pulse sequences from the training set is randomly selected and input into the spiking neural network to obtain the membrane potential of the training output layer. By analyzing the corresponding characteristics of the membrane potential of the training output layer, the operating status of the electricity meter is determined (normal operation, typical fault types, such as power failure, which is manifested by the membrane potential of the output layer neurons remaining in a low potential state). The accuracy of the judgment is checked by comparing it with the data in the validation set. If accurate, the next round of training continues. If inaccurate, a smoothing function is used to approximate the gradient of the input layer neurons. The adjustment direction is determined according to the positive or negative characteristics of the gradient. For example, when the gradient is positive, the value of the convolutional layer weights is reduced, and the convolutional layer weights of the spiking neural network are updated by the gradient substitution method. During this period, the accuracy of the validation set is calculated after each round, and the change in the accuracy of the validation set after each iteration is recorded. The accuracy of the validation set is the ratio of the number of correctly predicted samples to the total number of samples in the validation set. The entire training process continues for several rounds until the accuracy on the validation set no longer changes or the preset maximum number of iterations (e.g., 100 times) is reached, indicating that the spiking neural network training is complete, and the trained spiking neural network is obtained.

[0087] The trained spiking neural network is invoked, and time-stamped convolutional layers are applied to identify the pulse density of the standardized pulse sequence. A fixed convolutional kernel size is used, and a sliding window is used to acquire local pulse patterns at three consecutive time points. The local pulse patterns are stacked in chronological order to generate a pulse event feature set. The pulse event feature set is transmitted to the input layer. The input layer neurons (e.g., LIF neurons) read the pulse information in the pulse event feature set and determine whether a pulse event exists to decide whether to fire a membrane potential. If a pulse event exists, a membrane potential signal is fired; otherwise, the neuron remains silent. After the sliding window acquires all the pulse event information in the standardized pulse sequence, a pulse event feature map is generated and the initial state of the membrane potential is obtained. The initial state of the membrane potential is synchronized to the input layer of the spiking neural network.

[0088] Furthermore, a max pooling layer is used in the spiking neural network to reduce the dimensionality of the spiking event feature map. Pooling reduces the data dimensionality while retaining key features, thus obtaining a pooled spiking event feature map. The pooled spiking event feature map is divided into fixed-length sliding windows (e.g., window length is 0.1 seconds) according to the timestamp. The number of spiking events in each window is counted, the average number of spiking events per second is calculated, and the spiking event density value in each sliding window is obtained. The spiking event density values ​​of all windows are sorted by timestamp, and the spiking event density values ​​are mapped to the [0,1] interval to obtain the density region distribution of the standardized spiking sequence.

[0089] S4.2 Extract and mark the pulse timestamp difference of the standardized pulse sequence to obtain the interval feature mark of the standardized pulse sequence;

[0090] Specifically, the standardized pulse sequence is arranged in timestamp order and transformed into a two-dimensional pulse matrix. The rows of the matrix represent timestamps, and the columns represent the existence status of pulse events (e.g., 1 indicates the existence of a pulse event, and 0 indicates the absence of a pulse event). This generates the interval feature markers of the standardized pulse sequence.

[0091] S4.3. Perform gradient calculation based on the standardized pulse sequence, set the mutation threshold, identify and mark mutation events, and obtain a list of mutation events for the standardized pulse sequence;

[0092] Specifically, timestamps of all pulse events are extracted from the standardized pulse sequence and arranged in chronological order to form a sequence; the absolute difference between adjacent timestamps is calculated to generate a gradient sequence. For example, if the timestamp sequence is [0.1 0.12 0.14], then the gradient sequence is [0.02 0.02]; a mutation threshold is set based on the preset current fluctuation threshold and the operating frequency of the energy meter for current mutation detection. For example, for the DD862 single-phase mechanical energy meter, the preset current fluctuation threshold of the current mutation detection module is 0.25A. Under the working condition of 80% rated working load, the maximum operating frequency is 500Hz. The mutation threshold is set as the ratio of the current fluctuation threshold to the operating frequency, specifically 0.025.

[0093] Furthermore, to distinguish between normal fluctuations and abnormal mutations, the values ​​in the gradient sequence are compared with the mutation threshold one by one. If the current gradient value is greater than or equal to the mutation threshold, it is determined to be a mutation event and marked as 1. If the current gradient value is less than the mutation threshold, it is determined to be a non-mutation event and marked as 0, thus generating a binary mutation sequence.

[0094] Align the binarized marker sequence with the normalized pulse sequence timestamps to form a list of mutation events. Each marker in the list corresponds to the position of the normalized pulse sequence timestamp, clearly recording the time point when the mutation event occurred.

[0095] S4.4. Integrate the density region distribution, interval feature markers, and mutation event list of the standardized pulse sequence to obtain the pulse sequence characteristics of the standardized pulse sequence;

[0096] Specifically, the density region distribution is normalized to obtain a normalized density region distribution. The binary data format of the interval feature markers and the mutation event list is preserved. The normalized density region distribution, interval feature markers, and mutation event list are unified in terms of time dimension. The data information corresponding to the timestamps is reflected on a discrete common time axis. Zero values ​​are inserted to fill in the parts of the density region distribution, interval feature marker sequence, and mutation event list that are not defined on the common time axis. The results are integrated into a three-dimensional feature matrix to obtain the pulse sequence features of the standardized pulse sequence.

[0097] S4.5 Based on pulse sequence characteristics, calculate the fault weight of the energy meter and generate a synaptic current pulse sequence in the pulse neural network;

[0098] Specifically, based on the contribution ratio of the pulse sequence characteristics of the standardized pulse sequence in fault diagnosis, corresponding fault weights are assigned. For example, based on the typical contribution ratio of various fault characteristics in the field of power metering, the density region distribution accounts for 50%, the interval feature label accounts for 30%, and the abrupt event list accounts for 20%. The fault weights and pulse sequence characteristics are weighted and calculated to generate a synaptic current pulse sequence, which serves as the input signal for the pulse neural network.

[0099] S4.6. Inject the synaptic current pulse sequence into the neurons of the spiking neural network, and after triggering the membrane potential response, update the neuron state and issue a fault pulse event;

[0100] Specifically, the number of input layer neurons in the spiking neural network is set to be consistent with the dimension of the synaptic current pulse sequence. Current injection: The synaptic current pulse sequence is injected into the input layer neurons, and the membrane potential of the input layer neurons is updated using LIF. The calculation expression is as follows:

[0101] ;

[0102] in, It is the updated membrane potential voltage. This is the membrane potential voltage before the update, during the first calculation. The value is consistent with the initial state of the membrane potential. It is a membrane resistor (e.g., in a pulsed version of the ResNet spiking neural network). =0.5Ω), It is the current synaptic current value of the neuron, which is determined by the values ​​in the synaptic current pulse sequence.

[0103] The output of the input layer neurons is transmitted to the intermediate layer neurons through synaptic connections, and the output of the intermediate layer neurons is transmitted to the output layer neurons through synaptic connections. Finally, the membrane potential of the output layer neurons determines the fault type classification result of the electricity meter.

[0104] It should be noted that the membrane potential of the output layer neurons can be used to directly determine the fault type of the electricity meter. For example, in determining a battery undervoltage fault, the intermediate layer neurons activate the output layer through abrupt event markers (such as pulse interruption at the moment of power failure), causing a brief increase in the intermediate layer membrane potential. The output layer neuron membrane potential is not triggered, and in this case, the electricity meter fault type is determined to be a battery undervoltage fault. In determining a power supply fault, the output layer neuron membrane potential remains at a low potential state (because the power supply fault causes no current injection into the input layer neurons), but the intermediate layer neurons activate the output layer through abrupt event markers (such as voltage interruption), causing a brief increase in the output layer membrane potential. If the meter fluctuates and then remains at a low potential, the fault type is determined to be a power supply fault. For metering inaccuracy faults, if the output layer neuron membrane potential remains at a high potential (e.g., dense creeping pulses), and a mutation event directly triggers the output layer neuron membrane potential while the intermediate layer neuron points change normally, the fault type is determined to be a metering inaccuracy fault. For electricity theft, if the output layer neuron membrane potential is directly activated by a mutation event, briefly reaches a high potential, and then remains at a low potential, and the intermediate layer neuron membrane potential also briefly reaches a high potential and then remains at a low potential under the influence of the input layer neuron membrane potential, this is determined to be electricity theft.

[0105] S4.7 Count the number of fault pulse events emitted by the neuron and normalize them to obtain the fault probability distribution of the electricity meter;

[0106] Specifically, the number of membrane potential outputs of output layer neurons after determining the fault type is summed with the total number of pulse firings of all neurons in the input, intermediate, and output layers. The fault probability is obtained by calculating the sum of the number of membrane potential outputs of output layer neurons after determining the fault type and the total number of pulse firings of all neurons in the input, intermediate, and output layers. This probability is then correlated with the fault type of the electricity meter. After collecting the fault probability and fault type of several electricity meters of the same type (for example, collecting 100 data points to ensure the reference value of the sample), the fault type and fault probability data are normalized to obtain the fault probability distribution of the electricity meter, which includes the corresponding occurrence probability of each fault type.

[0107] S5. Generate feedback control signals and energy meter evaluation reports based on fault probability distribution.

[0108] S5.1. By analyzing the numerical distribution in the fault probability distribution, a multi-level response strategy is issued based on different intervals of the numerical distribution.

[0109] Specifically, the probability values ​​of all fault types are extracted from the fault probability distribution, and the corresponding occurrence probabilities of each fault type are sorted according to their numerical values. The sorting is done in descending order to ensure that the highest fault probability value is placed first and processed first.

[0110] Based on the fault diagnosis history of electricity meters, probability distribution thresholds are set, dividing the probability distribution into three intervals: high probability interval (probability value ≥ threshold 1), medium probability interval (threshold 1 > probability value ≥ threshold 2), and low probability interval (probability value < threshold 2). For example, in the fault diagnosis history of common electricity meters, when the fault probability exceeds 40%, the electricity meter cannot be repaired solely under remote control, which is mostly due to hardware-level faults of the electricity meter or human intervention factors such as electricity theft. Threshold 1 is set at 45%. When the fault probability is less than 10%, the fault diagnosis results of the electricity meter are mostly false alarms caused by environmental factors. Threshold 2 is set at 10%.

[0111] Based on different intervals of the numerical distribution, a multi-level response strategy is generated. For fault types in the high-probability interval, a first-level response strategy is triggered, including immediately generating a high-priority feedback control signal, marking the fault type as "emergency handling," and generating feedback information containing the fault type, fault probability, and recommended handling measures. For fault types in the medium-probability interval, a second-level response strategy is triggered, including generating a second-priority feedback control signal, marking the fault type as "to be observed," and generating feedback information containing the fault type, fault probability, and recommended monitoring measures. For fault types in the low-probability interval, a third-level response strategy is triggered, including marking the fault type as "routine recording," and generating feedback information containing the fault type, fault probability, and recommended periodic inspection measures.

[0112] S5.2 Match the control strategy according to the fault type in the fault probability distribution;

[0113] Specifically, based on the fault type in the fault probability distribution, control strategies are matched and distributed to maintenance stations according to the priority order in the multi-level response strategy, and a control strategy table is generated. For example, for a battery undervoltage fault, the matching strategy is to replace the meter battery and check for poor battery contact; for a power supply fault, the matching strategy is to replace the meter power module and check for power supply line connectivity issues; for a metering inaccuracy fault, the matching strategy is to check the metering circuit of the meter and recalibrate the metering parameters; for electricity theft, the matching strategy is to check the meter seal and replenish the electricity.

[0114] S5.3 Encapsulate multi-level response strategies and control strategies, generate feedback control signals, and record them as the operation log of the energy meter;

[0115] Specifically, the system reads multi-level response strategies (such as "emergency handling," "waiting to be observed," and "routine recording") and their corresponding response levels (such as high, medium, and low), and reads the control strategy table, which contains the mapping relationship between fault types and corresponding handling measures. Based on the multi-level response strategies and the control strategy table, the system defines the data structure for feedback control signals, including fault type identifiers (unique codes referencing fault types, such as "Err-04" indicating battery undervoltage), response level identifiers (letters representing response levels, such as "L1" indicating high priority and "L2" indicating medium priority), control strategy codes (referencing handling measure codes in the control strategy table, such as "ACT-01" indicating battery replacement), and additional parameters (including fault probability values ​​and execution time). The system then obtains the feedback control signals. Finally, the generated feedback control signals are written to the energy meter's operation log in a fixed format, including a timestamp (recording the specific time the signal was generated), fault type, response level, and control strategy.

[0116] S5.4. Obtain the current error signal, acoustic signal, fault probability distribution, and operation log of the encapsulated ammeter to obtain the energy meter evaluation report;

[0117] Specifically, data sources, including current error signals, acoustic signals, fault probability distributions, and operation logs from electricity meters, are integrated, processed in a structured format, classified by fault type, and compiled into storage devices to form independent files for electricity meter fault diagnosis, thereby obtaining electricity meter assessment reports.

[0118] This embodiment also provides a system for testing and evaluating the operating status of electricity meters, including:

[0119] The error extraction module collects and preprocesses the electrical operating data of the energy meter, obtains the reference current and operating current of the energy meter, extracts error features, and generates a current error signal.

[0120] The voiceprint acquisition module collects the voiceprint of the electricity meter and adjusts the sampling frequency and duration of the voiceprint acquisition based on the current error signal to capture the voiceprint signal of the electricity meter.

[0121] The sequence conversion module converts the voiceprint signal into a discrete pulse sequence through an encoding algorithm. Based on the discrete pulse sequence, it obtains a standardized pulse sequence through dimensional transformation and optimized compression.

[0122] The fault identification module uses a spiking neural network to extract features from a standardized pulse sequence. After obtaining the pulse sequence features, the fault probability distribution is obtained through synaptic weight calculation and neuron state update.

[0123] The feedback evaluation module generates feedback control signals and energy meter evaluation reports based on the fault probability distribution.

[0124] In summary, this invention achieves the following: It obtains a more accurate reference current through the quantum Hall effect, replacing the traditional detection method that relies on an external reference source, thus eliminating reference drift error at the physical level; it adjusts the acoustic signature sampling parameters in real time based on the current error signal and dynamically corrects the acoustic signature signal using phase deviation, effectively suppressing environmental noise interference; it applies a pulse neural network to process the standardized pulse sequence of the energy meter to obtain the fault probability distribution, enabling rapid fault type location and identification in real time; and it employs a multi-level response strategy based on the fault probability distribution to accurately match maintenance priorities and promptly select and issue control strategies, thus solving the problem of poor real-time performance in fault mode recognition.

[0125] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for inspecting and evaluating the operating status of an electricity meter, characterized in that: include, Collect and preprocess the electrical operating data of the electricity meter, obtain the reference current and operating current of the electricity meter and extract error features to generate a current error signal; The soundprint of the electricity meter is collected, and the sampling frequency and duration of the soundprint collection device are adjusted based on the current error signal to capture the soundprint signal of the electricity meter. The voiceprint signal is converted into a discrete pulse sequence through an encoding algorithm. Based on the discrete pulse sequence, a standardized pulse sequence is obtained through dimensional transformation and optimized compression. A spiking neural network is used to extract features from a standardized pulse sequence. After obtaining the pulse sequence features, the fault probability distribution is obtained by calculating synaptic weights and updating neuron states. Generate feedback control signals and energy meter evaluation reports based on fault probability distribution; The electrical operating data includes the quantum Hall voltage and raw operating current of the energy meter; The preprocessing includes voltage conversion of the quantum Hall voltage to obtain the reference current of the energy meter, noise reduction filtering of the original operating current to obtain the operating current of the energy meter; The extraction of error features refers to calculating the amplitude deviation, detecting the phase deviation, and analyzing the harmonic distortion of the reference current and the operating current. After obtaining the amplitude deviation, phase deviation, and harmonic distortion, the data is integrated to generate a current error signal. The specific steps for adjusting the sampling frequency and duration of the acoustic signature acquisition device based on the current error signal to capture the acoustic signature signal of the ammeter are as follows. The sampling frequency of the voiceprint acquisition device is set according to the amplitude deviation, and the duration of the voiceprint acquisition device is set according to the harmonic distortion. Start the voiceprint acquisition device to obtain the initial voiceprint information of the electricity meter; The initial acoustic signature information is adjusted based on the phase deviation to generate the acoustic signature signal of the ammeter.

2. The method for inspecting and evaluating the operating status of an electricity meter as described in claim 1, characterized in that: The process involves converting the voiceprint signal into a discrete pulse sequence using an encoding algorithm, and then obtaining a standardized pulse sequence through dimensionality transformation and optimized compression based on the discrete pulse sequence. The specific steps are as follows. By using an encoding algorithm, the waveform of the voiceprint signal is scanned and dynamic threshold mutation is determined to obtain a discrete pulse sequence; Transform the time axis dimension of the discrete pulse sequence to establish a pulse timing matrix; The discrete pulse sequence is optimized and compressed based on the pulse timing matrix, merging adjacent pulses and deleting isolated pulses to generate a standardized pulse sequence.

3. The method for inspecting and evaluating the operating status of an electricity meter as described in claim 1, characterized in that: The application of a spiking neural network to extract features from the standardized pulse sequence involves the following specific steps. The pulse density of the standardized pulse sequence is identified by a spiking neural network to obtain the density region distribution of the standardized pulse sequence. Extract and label the pulse timestamp differences of the standardized pulse sequence to obtain the interval feature label of the standardized pulse sequence; Gradient calculation is performed based on standardized pulse sequences, mutation thresholds are set, mutation events are identified and marked, and a list of mutation events of the standardized pulse sequences is obtained. The pulse sequence characteristics of the normalized pulse sequence are obtained by integrating the density region distribution, interval feature markers, and mutation event list of the normalized pulse sequence.

4. The method for inspecting and evaluating the operating status of an electricity meter as described in claim 1, characterized in that: After acquiring the pulse sequence features, the fault probability distribution is obtained through synaptic weight calculation and neuron state update. The specific steps are as follows. Based on pulse sequence characteristics, fault weights of energy meters are calculated and synaptic current pulse sequences are generated in a pulse neural network. A sequence of synaptic current pulses is injected into the neurons of a spiking neural network, and after triggering a membrane potential response, the neuron state is updated and a fault pulse event is emitted. The number of fault pulse events emitted by neurons is counted and normalized to obtain the fault probability distribution of the electricity meter.

5. The method for inspecting and evaluating the operating status of an electricity meter as described in claim 1, characterized in that: The specific steps for generating feedback control signals and energy meter evaluation reports based on fault probability distribution are as follows: By analyzing the numerical distribution in the failure probability distribution, a multi-level response strategy is issued based on different intervals of the numerical distribution; Match control strategies based on the fault types in the fault probability distribution; Encapsulate multi-level response and control strategies, generate feedback control signals, and record them as the operation log of the energy meter; The current error signal, acoustic signal, fault probability distribution, and operation log of the encapsulated ammeter are used to obtain an energy meter evaluation report.

6. A system for inspecting and evaluating the operating status of an electricity meter, based on the method for inspecting and evaluating the operating status of an electricity meter according to any one of claims 1 to 5, characterized in that: include, The error extraction module collects and preprocesses the electrical operating data of the energy meter, obtains the reference current and operating current of the energy meter, extracts error features, and generates a current error signal. The voiceprint acquisition module collects the voiceprint of the electricity meter and adjusts the sampling frequency and duration of the voiceprint acquisition device based on the current error signal to capture the voiceprint signal of the electricity meter. The sequence conversion module converts the voiceprint signal into a discrete pulse sequence through an encoding algorithm. Based on the discrete pulse sequence, it obtains a standardized pulse sequence through dimensional transformation and optimized compression. The fault identification module uses a spiking neural network to extract features from a standardized pulse sequence. After obtaining the pulse sequence features, the fault probability distribution is obtained through synaptic weight calculation and neuron state update. The feedback evaluation module generates feedback control signals and energy meter evaluation reports based on the fault probability distribution.

Images