A synchronous electric energy meter error verification and correction method, medium and electronic device

Abstract

The present invention provides a synchronous electric energy meter error verification and correction method, medium and electronic device, belonging to the technical field of electric energy meters, comprising: collecting electric energy metering data of the electric energy meter and current, voltage and frequency signals of connected equipment; using a multivariate mixed decomposition method to decompose the collected current and voltage signals into fundamental and harmonic components; calculating the active load; detecting load waveform characteristics, such as the active resonance density of peak and valley points; establishing an electric energy meter error model considering voltage, current and frequency factors according to the load waveform characteristics; substituting the collected electrical parameter data into the error model to calculate the basic error of the electric energy meter; comparing the basic error with the actual error and correcting the error model parameters; iteratively obtaining a corrected error model; using the corrected error model to correct the measurement error of the electric energy meter; solving the technical problem that the prior art is difficult to correct the error of the electric energy meter for changes in power quality and load uncertainty.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric energy meters, and in particular, relates to a synchronous electric energy meter error verification and correction method, a medium and an electronic device. Background Art

[0002] Electronic watt-hour meters are currently widely used in the power industry. Compared with traditional mechanical watt-hour meters, electronic watt-hour meters have the advantages of small size, low error, and easy remote meter reading. They are widely used in residential life, industrial and commercial electricity, power grid dispatching and other fields. However, with the continuous development of the power system, electronic watt-hour meters are also facing some new metering problems in practical applications.

[0003] First, with the widespread application of power electronic equipment, the voltage distortion and current harmonics problems in the power grid are becoming increasingly serious. These power quality problems will have an adverse effect on the measurement accuracy of the electric energy meter. As a measurement device based on sampling and digital signal processing, the measurement error of the electronic electric energy meter is easily affected by factors such as voltage distortion and current harmonics, thereby affecting the measurement accuracy. Traditional electric energy meter error calibration methods, such as regular calibration using standard electric energy meters and correction using error curves, can no longer effectively solve the impact of power quality changes on measurement accuracy.

[0004] Secondly, the uncertainty of power load also brings new challenges to the measurement of electric energy meters. The load patterns of industrial and commercial electricity and residential electricity are becoming increasingly complex and diverse, and different types of loads have different degrees of influence on the measurement accuracy of electric energy meters. For example, switching power supply loads will produce serious current harmonics, which will cause great interference to the measurement accuracy; and motor loads will cause changes in the phase difference between voltage and current, which will also affect the measurement accuracy of active power. The traditional single electric energy meter error calibration method is difficult to fully consider the impact of different loads on the measurement accuracy.

[0005] Thirdly, the intelligent trend of power system requires electric energy meter to have higher measurement accuracy and dynamic tracking ability. With the promotion and application of new power systems such as smart grid and microgrid, electric energy meter needs to realize high-frequency sampling and real-time measurement of parameters such as instantaneous voltage and current to meet the needs of power grid dispatching, load forecasting and other applications. The traditional electric energy meter error calibration method is difficult to realize real-time tracking and dynamic correction of electric energy meter measurement error.

[0006] Therefore, in view of the new challenges brought by changes in power quality and load uncertainty to the measurement accuracy of electricity meters, it is urgent to study a new type of electricity meter error verification and correction technology to improve the measurement accuracy of electricity meters in complex power environments and meet the needs of future power system development. Summary of the invention

[0007] In view of this, the present invention provides a synchronous electric energy meter error detection and correction method, medium and electronic device, which can solve the technical problem that the prior art is difficult to correct the error of the electric energy meter according to the change of power quality and load uncertainty.

[0008] The present invention is achieved in that:

[0009] A first aspect of the present invention provides a method for error detection and correction of a synchronizable electric energy meter, which comprises the following steps:

[0010] S10, obtaining the electric metering data of the electric energy meter and the current, voltage and frequency signals collected by the standard electric energy detection equipment connected to the electric energy meter;

[0011] S20, using a multivariate hybrid decomposition method to decompose the collected current and voltage signals into different characteristic components, and extracting fundamental components and harmonic components from the decomposed characteristic components;

[0012] S30, calculating active load, including fundamental wave active power and harmonic active power;

[0013] S40, detecting the active resonance density at the peak and valley points of the load as a load waveform feature;

[0014] S50, establishing an error model of the electric energy meter according to the load waveform characteristics, including voltage, current, and frequency as influencing factors;

[0015] S60, substituting the detected current, voltage and frequency data into the error model to calculate the basic error of the electric energy meter;

[0016] S70, using the collected voltage, current, and frequency signals to calculate actual electric energy data, comparing it with the electric energy meter reading, and calculating the difference between the basic error and the actual error;

[0017] S80, using the difference to correct error model parameters;

[0018] S90, repeating steps S60 to S80 until the error model converges, and obtaining a corrected error model for error correction of the electric energy meter.

[0019] Wherein, the step S10 specifically includes: collecting the electric metering data of the electric energy meter from the communication interface of the electric energy meter, including active power, reactive power, voltage, current, power factor and other information; at the same time, collecting the current, voltage, frequency and other signals collected by the standard equipment through the measurement interface of the standard electric energy detection equipment; the electric metering data of the electric energy meter and the current, voltage and frequency signals collected by the standard electric energy detection equipment provide necessary input for subsequent error analysis.

[0020] Among them, the step S20 specifically uses the wavelet analysis method to decompose the collected current and voltage signals into fundamental components and harmonic components; by performing wavelet decomposition on the current and voltage signals, the characteristic values ​​of the fundamental and harmonic components can be extracted to provide a basis for subsequent error analysis.

[0021] Among them, the step S30 first calculates the fundamental active power and the harmonic active power according to the decomposed current and voltage signals, and then takes the sum of the fundamental active power and the harmonic active power as the active load, reflecting the total active power measured by the electric energy meter.

[0022] Among them, the step S40, detecting the active resonance density in the load waveform characteristics, specifically includes: dividing the load waveform characteristic curve into multiple time windows, counting the active harmonic components in each time window, calculating the harmonic density index, and then averaging the harmonic density index of each time window to obtain the active resonance density of the entire load waveform. This index can reflect the degree of distortion of the load waveform and provide a basis for subsequent error modeling.

[0023] Among them, the step S50 adopts an error model structure composed of a long-term branch, a medium-term branch, a short-term branch and a fusion branch, wherein the long-term branch adopts a long short-term memory (LSTM) network, the input is a long-term load waveform feature sequence, and the output is a long-term error component prediction; the medium-term branch adopts a convolutional neural network (CNN), the input is a medium-term load waveform feature matrix, and the output is a medium-term error component prediction; the short-term branch adopts a feedforward neural network, the input is a short-term load waveform feature vector, and the output is a short-term error component prediction; the fusion branch performs weighted summation on the prediction results of the three branches to obtain the final error prediction value.

[0024] Among them, the step S60 specifically includes: inputting the detected current, voltage and frequency data into the error model established in step S50, obtaining three error component predictions through short-term branches, medium-term branches and long-term branches respectively, and then performing weighted summation through fusion branches to obtain the final basic error prediction value of the electric energy meter.

[0025] Among them, the step S70 specifically includes: calculating the actual active power and reactive power based on the collected voltage, current and frequency signals, and then obtaining the actual electric energy data; comparing the actual electric energy data with the electric energy meter reading to calculate the actual error; comparing the basic error obtained in step S60 with the actual error, and calculating the difference between the two.

[0026] Among them, the step S80 specifically includes: analyzing the difference between the basic error and the actual error to identify the systematic deviation in the error model; adjusting the weight parameters of each branch in the error model according to the size and direction of the difference to reduce the prediction error; if necessary, the hyperparameters of each branch neural network can also be fine-tuned to further optimize the model performance.

[0027] Among them, the step S90 specifically includes: applying the corrected error model again to a new round of current, voltage, and frequency data to calculate the basic error; comparing it with the actual error to obtain a new difference; further adjusting the model parameters based on the new difference; repeating the above process until the model prediction error converges to a satisfactory accuracy level.

[0028] Among them, the error model includes a long-term parameter branch, a medium-term parameter branch, a short-term parameter branch and a fusion branch. The long-term parameter branch is used to predict the metering error of the electric energy meter based on the load waveform characteristics obtained by long-term continuous collection and calculation. The medium-term parameter branch is used to predict the metering error of the electric energy meter based on the load waveform characteristics obtained by medium-term continuous collection and calculation. The short-term parameter branch is used to predict the metering error of the electric energy meter based on the load waveform characteristics obtained by short-term continuous collection and calculation. The fusion branch is used to integrate the errors predicted in the three time scales of long, medium and short to obtain the final error prediction value.

[0029] Furthermore, the long-term parameter branch adopts a long short-term memory network, the input is a long-term load feature sequence, and the output is a long-term error component prediction.

[0030] Furthermore, the mid-term parameter branch adopts a convolutional network, the input is the mid-term load feature matrix, and the output is the mid-term error component prediction.

[0031] Furthermore, the short-term parameter branch adopts a feedforward network, the input is the short-term load characteristic vector, and the output is the short-term error component prediction.

[0032] Furthermore, the fusion branch is used to perform weighted summation on the predictions of the three branches to obtain a final error prediction value.

[0033] Furthermore, the short term is several hours or days; the medium term is several weeks or months; and the long term is several months or years.

[0034] Among them, the error correction formula of the electric energy meter is described as: corrected electric energy meter reading = original electric energy meter reading / (1+error rate predicted by the error model); or corrected electric energy meter reading = original electric energy meter reading + predicted error median value.

[0035] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for error verification and correction of a synchronizable electric energy meter.

[0036] A third aspect of the present invention provides an electronic device, which comprises a processor and a memory, wherein the memory is used to store the step program of the above-mentioned method, and the processor reads the memory and executes the steps.

[0037] Compared with the prior art, the beneficial effects of the synchronous electric energy meter error calibration and correction method, medium and electronic device provided by the present invention are: using the electric energy meter's own electricity measurement data and the voltage and current signals collected by the standard detection equipment, and using advanced technologies such as wavelet analysis and neural networks, a long, medium and short time scale fusion electric energy meter error prediction model is constructed. This model can effectively capture the characteristics of electric energy meter errors changing with time and load, and by continuously optimizing the model parameters, it can ultimately achieve accurate prediction and correction of electric energy meter errors. Compared with the traditional single electric energy meter error calibration method, the scheme of the present invention has the following technical advantages:

[0038] 1. In response to the problem of deteriorating power quality in the power grid, the solution of the present invention uses wavelet analysis technology to extract fundamental and harmonic characteristics from the collected voltage and current signals, laying the foundation for error modeling of the power meter. This error modeling method that takes power quality factors into consideration can better reflect the metering characteristics of the power meter in a complex power environment and has higher metering accuracy than traditional methods.

[0039] 2. Aiming at the metering error problem caused by load uncertainty, the solution of the present invention constructs an electric energy meter error prediction model that includes long-term, medium-term and short-term factors. The model can capture the impact of load characteristics on metering errors at different time scales, obtain the final error prediction through adaptive fusion, and better adapt to the metering needs of electric energy meters under complex load environments.

[0040] 3. Aiming at the requirements for real-time metering accuracy of electric energy meters under the trend of intelligent power system, the scheme of the present invention realizes real-time tracking and dynamic correction of metering errors of electric energy meters. By repeatedly optimizing the error model parameters, the scheme can continuously improve the metering accuracy of electric energy meters and meet the needs of future power systems for high-precision metering.

[0041] In summary, the synchronous electric energy meter error detection and correction method proposed in the present invention makes full use of advanced signal processing and machine learning technologies to construct an electric energy meter error prediction model with adaptive learning capabilities. The model can accurately capture the dynamic characteristics of the electric energy meter error as the power quality of the power grid and the load characteristics change, realize the real-time tracking and effective correction of the electric energy meter measurement error, greatly improve the measurement accuracy of the electric energy meter in a complex power environment, and solve the technical problem that the existing technology is difficult to correct the error of the electric energy meter for changes in power quality and load uncertainty. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0043] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0045] like Figure 1 FIG. 1 is a flow chart of a synchronous electric energy meter error detection and correction method provided by the present invention. The method comprises the following steps:

[0046] S10, obtaining the electric metering data of the electric energy meter and the current, voltage and frequency signals collected by the standard electric energy detection equipment connected to the electric energy meter;

[0047] S20, using a multivariate hybrid decomposition method to decompose the collected current and voltage signals into different characteristic components, and extracting fundamental components and harmonic components from the decomposed characteristic components;

[0048] S30, calculating active load, including fundamental wave active power and harmonic active power;

[0049] S40, detecting the active resonance density at the peak and valley points of the load as a load waveform feature;

[0050] S50, establishing an error model of the electric energy meter according to the load waveform characteristics, including voltage, current, and frequency as influencing factors;

[0051] S60, substituting the detected current, voltage and frequency data into the error model to calculate the basic error of the electric energy meter;

[0052] S70, using the collected voltage, current, and frequency signals to calculate actual electric energy data, comparing it with the electric energy meter reading, and calculating the difference between the basic error and the actual error;

[0053] S80, using the difference to correct the error model parameters;

[0054] S90, repeating steps S60 to S80 until the error model converges, and obtaining a corrected error model for error correction of the electric energy meter.

[0055] The specific implementation methods of the above steps are described in detail below:

[0056] The specific implementation method of step S10 is: first, the electric metering data of the electric energy meter is obtained through the communication interface of the electric energy meter, including active power, reactive power, voltage, current, power factor and other information. At the same time, the current, voltage, frequency and other signals collected by the standard equipment are collected through the measurement interface of the standard electric energy detection equipment. These data provide necessary input for subsequent error analysis.

[0057] The specific implementation of step S20 is: using a multivariate hybrid decomposition method to decompose the collected current and voltage signals into different characteristic components. Specifically, the signal can be decomposed into fundamental and harmonic components using wavelet analysis. Wavelet analysis is a time-frequency analysis method that can effectively separate the fundamental and harmonic components of a signal. By performing wavelet decomposition on the current and voltage signals, the characteristic values ​​of the fundamental and harmonic waves can be extracted to provide a basis for subsequent error analysis.

[0058] The specific implementation of step S30 is: first, calculate the fundamental wave active power according to the decomposed current and voltage signals. The formula is as follows:

[0059]

[0060] Among them, u(t) and i(t) represent the fundamental voltage and fundamental current respectively, and T represents one cycle.

[0061] Secondly, calculate the harmonic active power. The formula is as follows:

[0062]

[0063] Among them, u h (t) and i h (t) represent harmonic voltage and harmonic current respectively.

[0064] Finally, the active load is equal to the sum of the fundamental wave active power and the harmonic active power. By calculating the active load, the total active power measured by the electric energy meter can be reflected.

[0065] The specific implementation method of step S40 is: detecting the active resonance density in the load waveform characteristics. The so-called active resonance density refers to the degree to which the active harmonic components are densely concentrated on a certain load waveform characteristic curve. The specific calculation method is as follows:

[0066] 1) Divide a load waveform characteristic curve into multiple time windows;

[0067] 2) Count the active harmonic components in each time window and calculate the harmonic density index;

[0068] 3) Take the average of the harmonic density index of each time window to obtain the active resonance density of the entire load waveform.

[0069] This indicator can reflect the degree of distortion of the load waveform and provide a basis for subsequent error modeling. Generally speaking, the higher the active resonance density, the more serious the load waveform distortion, and the greater the measurement error of the energy meter.

[0070] The specific implementation of step S50 is: according to the load waveform characteristics obtained above, establish an error model of the electric energy meter. The input variables of the model include voltage, current, frequency, etc., and the output variable is the measurement error of the electric energy meter. The specific model structure is as follows:

[0071] Error model = long-term branch + medium-term branch + short-term branch + fusion branch

[0072] in:

[0073] The long-term branch is a long short-term memory (LSTM) network, which takes as input a long-term (e.g., months or years) load waveform feature sequence and outputs a long-term error component prediction. The LSTM network can effectively capture long-term dependencies in time series data and is suitable for modeling the error characteristics of electric energy meters that change slowly over time.

[0074] The medium-term branch is a convolutional neural network (CNN), which takes as input the medium-term (e.g., weeks or months) load waveform feature matrix and outputs the medium-term error component prediction. CNN is good at extracting local correlations in the feature matrix and is suitable for modeling the medium-term error characteristics of the energy meter that change periodically with the load type.

[0075] The short-term branch is a feedforward neural network, which takes as input the short-term (such as hours or days) load waveform feature vector and outputs the short-term error component prediction. The feedforward network has a simple structure and is suitable for modeling the sensitivity of the energy meter to short-term fluctuations in power quality.

[0076] The fusion branch performs weighted summation of the prediction results of the three branches to obtain the final error prediction value. The weight of the fusion branch can be adaptively learned or set to a fixed value based on experience.

[0077] Through the separate modeling and adaptive fusion of long, medium and short time scales, the multi-time scale characteristics of the electric energy meter error can be better captured and the prediction accuracy can be improved. During training, the parameters of the three branches and the fusion branch can be jointly optimized at the same time.

[0078] The specific implementation of step S60 is: substituting the detected current, voltage and frequency data into the aforementioned error model to calculate the basic error of the electric energy meter. According to the structure of the error model, this step includes the following sub-steps:

[0079] 1) Input the short-term load waveform characteristics into the short-term branch to obtain the short-term error component prediction;

[0080] 2) Input the mid-term load waveform characteristics into the mid-term branch to obtain the mid-term error component prediction;

[0081] 3) Input the long-term load waveform characteristics into the long-term branch to obtain the long-term error component prediction;

[0082] 4) The prediction results of the three branches are weighted summed through the fusion branch to obtain the final basic error prediction of the electricity meter.

[0083] The key to this step is how to effectively connect the real-time collected current, voltage, and frequency data with the trained error model to obtain the basic error prediction value of the electricity meter.

[0084] The specific implementation of step S70 is: using the collected voltage, current, and frequency signals to calculate the actual power data, comparing it with the power meter reading, and calculating the difference between the basic error and the actual error. Specifically, this step includes:

[0085] 1) Calculate the actual active power and reactive power based on the collected voltage, current and frequency signals, and then obtain the actual electric energy data;

[0086] 2) Compare the actual power data with the power meter readings and calculate the actual error;

[0087] 3) Compare the basic error obtained in step S60 with the actual error and calculate the difference between the two.

[0088] This difference reflects the prediction accuracy of the error model and provides a basis for subsequent model optimization.

[0089] The specific implementation of step S80 is: using the difference to correct the error model parameters. Specifically, this step includes:

[0090] 1) Analyze the difference between the basic error and the actual error, and identify the systematic deviations in the error model;

[0091] 2) According to the size and direction of the difference, adjust the weight parameters of each branch in the error model to reduce the prediction error;

[0092] 3) If necessary, the hyperparameters of each branch neural network can be fine-tuned to further optimize the model performance.

[0093] By continuously correcting the model parameters, the basic error of the prediction is gradually approached to the actual error, and finally an accurate and reliable electric energy meter error prediction model is established.

[0094] The specific implementation of step S90 is: repeating steps S60 to S80 until the error model converges to obtain a corrected error model. Specifically, this step includes:

[0095] 1) Apply the corrected error model to a new round of current, voltage, and frequency data to calculate the basic error;

[0096] 2) Compare with the actual error to obtain a new difference;

[0097] 3) Further adjust the model parameters according to the new difference;

[0098] 4) Repeat the above process until the model prediction error converges to a satisfactory accuracy level; the satisfactory level here can be considered that the difference between the two output results does not exceed 5%;

[0099] 5) Once the model converges, the following error correction formula is used to correct the meter reading: Corrected meter reading = original meter reading / (1+prediction error rate); or Corrected meter reading = original meter reading + predicted error median value;

[0100] 6) Repeat steps 1-5 periodically to adapt to changes in meter performance over time and ensure continued accuracy of the corrections.

[0101] Through continuous iterative optimization, an accurate and reliable electric energy meter error prediction model is finally obtained, which provides a basis for the error verification and correction of the electric energy meter.

[0102] In summary, this synchronous electric energy meter error verification and correction method makes full use of the electric energy meter's own electricity measurement data and the current, voltage, and frequency signals collected by standard detection equipment, and uses advanced technologies such as multivariate hybrid decomposition and neural network modeling to construct an electric energy meter error prediction model that integrates long, medium, and short time scales. This model can effectively capture the characteristics of electric energy meter errors that change with time and load, and by continuously optimizing model parameters, it can ultimately achieve accurate prediction and correction of electric energy meter errors.

[0103] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for error verification and correction of a synchronizable electric energy meter.

[0104] A third aspect of the present invention provides an electronic device, which comprises a processor and a memory, wherein the memory is used to store the step program of the above-mentioned method, and the processor reads the memory and executes the steps.

[0105] In order to better implement the present invention, the following provides a specific embodiment 1 of the present invention in a computer readable storage medium or an electronic device such as a computer, for a computer program. In the embodiment 1, in step S10, firstly, the electric metering data of the electric energy meter is obtained from the communication interface of the electric energy meter, including the active power P a , reactive power Q a , voltage U, current I, power factor cosφ and other information. At the same time, the current I collected by the standard equipment is collected through the measurement interface of the standard power detection equipment. s , voltage U s , frequency f s These data provide the necessary input for subsequent error analysis.

[0106] In step S20, the collected current and voltage signals are decomposed into different characteristic components by using a multivariate hybrid decomposition method. Specifically, the signal is decomposed into fundamental components and harmonic components by using wavelet analysis. Wavelet analysis is a time-frequency analysis method that can effectively separate the fundamental and harmonic components of a signal.

[0107] The basic idea of ​​wavelet analysis is to transform the continuous time signal x(t) to be analyzed through a wavelet transform composed of basis functions ψ(t) to obtain the dual representation of the signal in the time domain and frequency domain, namely the wavelet coefficient W(a, b). The wavelet transform formula is as follows:

[0108]

[0109] Among them, a represents the scale factor, b represents the translation factor, ψ * (t) is the complex conjugate function.

[0110] By performing wavelet decomposition on the current and voltage signals, the characteristic values ​​of the fundamental and harmonic waves can be extracted, providing a basis for subsequent error analysis. Assume that after wavelet decomposition, the current signal i(t) can be expressed as:

[0111]

[0112] Among them, i f (t) represents the fundamental current, in (t) represents the nth harmonic current, N is the highest harmonic order considered. Similarly, the voltage signal u(t) can also be decomposed into:

[0113]

[0114] With the characteristic values ​​of the fundamental wave and the harmonic wave, we can proceed to step S30 to calculate the active load.

[0115] First, the fundamental wave active power P is calculated based on the decomposed current and voltage signals. f :

[0116]

[0117] Where T represents a cycle.

[0118] Secondly, calculate the harmonic active power P h :

[0119]

[0120] Active load P is equal to fundamental active power P f and harmonic active power P h Sum:

[0121] P=P f +P h

[0122] By calculating the active load P, the total active power measured by the electric energy meter can be reflected.

[0123] In step S40, it is necessary to detect the active resonance density in the load waveform characteristics. The so-called active resonance density refers to the degree to which the active harmonic components are densely concentrated on a certain load waveform characteristic curve. The specific calculation method is as follows:

[0124] 1) Divide a load waveform characteristic curve into M time windows;

[0125] 2) For each active harmonic component P in the time window m h,m Perform statistics and calculate the harmonic density index D m :

[0126]

[0127] Among them, P h,m,n Represents the active power of the nth harmonic in the mth time window.

[0128] 3) Harmonic density index D for each time window m Taking the average, we can get the active resonance density D of the entire load waveform:

[0129]

[0130] This index D can reflect the degree of distortion of the load waveform and provide a basis for subsequent error modeling. Generally speaking, the higher the active resonance density D, the more serious the load waveform distortion, and the greater the measurement error of the energy meter.

[0131] In step S50, an error model of the electric energy meter is established based on the load waveform characteristics obtained above. The input variables of the model include voltage U, current I, frequency f, etc., and the output variable is the metering error ε of the electric energy meter. The specific model structure is as follows:

[0132] ε=ε long +ε mid +ε short +ε fusion

[0133] in:

[0134] Long-term branch ε long The long short-term memory (LSTM) network is used, and the input is a long-term (such as several months or years) load waveform feature sequence. The output is the long-term error component prediction:

[0135]

[0136] Mid-term branch ε mid A convolutional neural network (CNN) is used, with the input being a medium-term (e.g., weeks or months) load waveform feature matrix. The output is the medium-term error component forecast:

[0137] ε mid =f CNN (D m )

[0138] Short-term branch ε short A feedforward neural network is used, with the input being a short-term (such as hours or days) load waveform feature vector The output is the short-term error component forecast:

[0139] ε short =f FNN (d s )

[0140] Fusion branch ε fusion The prediction results of the three branches are weighted and summed to obtain the final error prediction value:

[0141] ε fusion =w long ε long +Wmid ε mid +W short ε short

[0142] Among them, W long ,W mid ,W short is the fusion weight for adaptive learning.

[0143] Through the separate modeling and adaptive fusion of long, medium and short time scales, the multi-time scale characteristics of the electric energy meter error can be better captured and the prediction accuracy can be improved. During training, the parameters of the three branches and the fusion branch can be jointly optimized at the same time.

[0144] In step S60, the detected current I, voltage U, and frequency f data are substituted into the aforementioned error model to calculate the basic error ε of the electric energy meter. basic Specifically, this step includes:

[0145] 1) The short-term load waveform characteristics d s Enter the short branch f FNN , get the short-term error component prediction ε short ;

[0146] 2) The mid-term load waveform characteristic D m Enter mid-term branch f CNN , and obtain the medium-term error component prediction ε mid ;

[0147] 3) Long-term load waveform characteristics Enter the long-term branch f LSTM , get the long-term error component prediction ε long ;

[0148] 4) The prediction results of the three branches are weighted and summed through the fusion branch to obtain the final basic error prediction ε of the electric energy meter basic :

[0149] ε basic =w long ε long +w mid ε mid +w short ε short

[0150] The key to this step is how to effectively connect the real-time collected current, voltage, and frequency data with the trained error model to obtain the basic error prediction value of the electricity meter.

[0151] In step S70, the actual electric energy data E is calculated using the collected voltage U, current I, and frequency f signals. actual, and the energy meter reading E meter Compare and calculate the basic error ε basic The actual error ε actual Specifically, this step includes:

[0152] 1) Calculate the actual active power P based on the collected voltage U, current I, and frequency f signals actual , and then get the actual power data E actual :

[0153] P actual =UI cosφ

[0154]

[0155] 2) The actual electric energy data E actual With the energy meter reading E meter Compare and calculate the actual error ε actual :

[0156]

[0157] 3) The basic error ε obtained in step S60 is basic The actual error ε actual Compare and calculate the difference Δε between the two:

[0158] Δε=ε actual -ε basic

[0159] This difference Δε reflects the prediction accuracy of the error model and provides a basis for subsequent model optimization.

[0160] In step S80, the error model parameters are corrected using the difference Δε. Specifically, this step includes:

[0161] 1) Analysis of basic error ε basic The actual error ε actual The difference Δε between them identifies the systematic deviation in the error model;

[0162] 2) According to the size and direction of the difference Δε, adjust the weight parameter w of each branch in the error model long ,w mid ,w short , to reduce the prediction error;

[0163] 3) If necessary, the hyperparameters of each branch neural network can be fine-tuned to further optimize the model performance.

[0164] By continuously modifying the model parameters, the basic error of the prediction is basicGradually approaching the actual error ε actual ,Finally, an accurate and reliable electric energy meter error prediction model is established.

[0165] Finally, in step S90, steps S60 to S80 are repeated until the error model converges to obtain a corrected error model. Specifically, this step includes:

[0166] 1) Apply the corrected error model to the new round of current I, voltage U, and frequency f data to calculate the basic error ε basic ;

[0167] 2) and the actual error ε actual Compare and obtain a new difference Δε;

[0168] 3) Further adjust the model parameters according to the new difference Δε;

[0169] 4) Repeat the above process until the model prediction error ε basic Converging to a satisfactory accuracy level, it is used to correct the error of the electricity measurement of the energy meter.

[0170] Through continuous iterative optimization, an accurate and reliable electric energy meter error prediction model is finally obtained, which provides a basis for the error verification and correction of the electric energy meter.

[0171] In addition, for step S20, multivariate mixed decomposition can also be used for calculation, wherein multivariate mixed decomposition (MEMD) is an adaptive time-frequency analysis method suitable for multivariate signal decomposition, which is a generalization of empirical mode decomposition (EMD). Unlike wavelet analysis, MEMD does not require any basis function to be set in advance, and can adaptively decompose different modal components according to the characteristics of the signal itself.

[0172] Specifically, for the current signal i(t) and the voltage signal u(t), they can be combined into a multivariate signal x(t) = [i(t), u(t)] T , and then perform MEMD decomposition. The decomposition steps are as follows:

[0173] 1. Construct an appropriate number of projection vectors Used to orient x(t);

[0174] 2. For each projection vector Project x(t) to get the projection signal

[0175] 3. For all projection signals Apply standard EMD to get the local average and local ring

[0176] 4. Average all local averages to get the global average

[0177] 5. Order If d(t) is a multivariate intrinsic mode function, then d(t) is the first modal component. Otherwise, take d(t) as the new input signal and repeat steps 2-5 until d(t) meets the stopping condition.

[0178] Through the above steps, the original signal x(t) can be decomposed into a series of intrinsic modal components d j (t) and residual r(t):

[0179]

[0180] Among them, d j (t) represents the jth modal component, which contains certain frequency components; r(t) is the residual, which represents the part that cannot be further decomposed.

[0181] For current and voltage signals, the modal components with lower frequencies can be regarded as fundamental components, and the modal components with higher frequencies can be regarded as harmonic components. Specifically, if the first M modal components d 1 (t),d 2 (t),…,d M (t) is regarded as the fundamental component, and the remaining d M+1 (t),…,d n (t) is regarded as a harmonic component, then the fundamental current i f (t) and fundamental voltage u f (t) can be expressed as:

[0182]

[0183] Harmonic current i h (t) Harmonic voltage u h (t) can be expressed as:

[0184]

[0185]

[0186] The superscripts i and u represent the modal components corresponding to current and voltage, respectively.

[0187] In this way, through MEMD decomposition, the original current and voltage signals can be decomposed into fundamental and harmonic parts, laying the foundation for subsequent active power calculation and error analysis.

[0188] Specifically, the principle of the present invention is:

[0189] 1. Obtain the electricity metering data and the voltage, current and frequency signals collected by the standard electricity detection equipment. The electricity metering data includes active power, reactive power, voltage, current, power factor, etc. These data can be directly collected through the communication interface of the electricity meter. At the same time, the standard electricity detection equipment is also used to collect the real-time voltage, current and frequency signals in the power grid. These data provide the necessary input for subsequent error analysis.

[0190] 2. Use wavelet analysis to decompose the collected voltage and current signals into fundamental and harmonic components. Wavelet analysis is a time-frequency analysis method that can effectively separate the fundamental and harmonic components of the signal. By performing wavelet decomposition on the voltage and current signals, the characteristic values ​​of the fundamental and harmonic components can be obtained, providing a basis for subsequent error modeling.

[0191] 3. Calculate the active load, including fundamental wave active power and harmonic wave active power. The size and waveform characteristics of the active load are important factors affecting the measurement accuracy of the electric energy meter. After calculating the active load, the relationship between the load characteristics and the electric energy meter error can be further analyzed.

[0192] 4. Detect the active resonance density in the load waveform characteristics. The so-called active resonance density refers to the degree to which the active harmonic components are densely concentrated on the load waveform characteristic curve. This indicator can reflect the severity of load distortion and provide a basis for subsequent error modeling. Generally speaking, the higher the active resonance density, the more serious the load distortion, and the greater the measurement error of the electric energy meter.

[0193] 5. According to the load waveform characteristics obtained above, an error prediction model for the electric energy meter is established. The model adopts the structure of long-term branch, medium-term branch, short-term branch and fusion branch, which can capture the impact of load characteristics on metering errors at different time scales. The long-term branch focuses on slowly changing factors, such as aging of electric energy meters and environmental impacts; the medium-term branch focuses on periodic changes, such as seasonal changes in load types; and the short-term branch focuses on short-term fluctuations in power quality. The prediction results of the three branches are adaptively weighted through the fusion branch to obtain the final error prediction value.

[0194] 6. Input the detected voltage, current and frequency data into the error prediction model to calculate the basic error of the electric energy meter. At the same time, use the collected voltage, current and frequency signals to calculate the actual electric energy data, compare it with the electric energy meter reading, and get the actual error. Compare the basic error with the actual error to get the difference between the two, which provides a basis for subsequent model optimization.

[0195] 7. According to the difference between the basic error and the actual error, adjust the parameters of the error prediction model, including the weights of each branch and the hyperparameters of the neural network. By continuously optimizing the model parameters, the predicted basic error gradually approaches the actual error, and finally an accurate and reliable electric energy meter error prediction model is established.

[0196] 8. Repeat the above steps and continue to optimize the error prediction model until the model prediction accuracy converges to a satisfactory level.

[0197] The reason why the solution of the present invention can effectively solve the new challenges faced by the accuracy of electric energy meter measurement is mainly due to the following key points:

[0198] Firstly, the wavelet analysis technology is fully utilized to extract fundamental and harmonic characteristics from voltage and current signals, laying a foundation for error modeling. This error modeling method that considers power quality factors can more accurately reflect the metering characteristics of electric energy meters in complex power environments.

[0199] Secondly, the neural network model structure that integrates long, medium and short time scales can capture the impact of load characteristics on metering errors at different time scales. Through adaptive learning, the model can continuously improve the prediction accuracy of metering errors of electric energy meters and meet the needs of future power systems for high-precision metering.

[0200] Thirdly, the scheme realizes the real-time tracking and dynamic correction of the metering error of the energy meter. By continuously optimizing the error prediction model, the metering accuracy of the energy meter can be continuously improved, which meets the needs of the intelligent development of the power system.

[0201] In summary, the synchronous electric energy meter error detection and correction method proposed in this invention fully integrates advanced signal processing and machine learning technologies, and constructs an electric energy meter error prediction model with adaptive learning capabilities. This model can accurately capture the dynamic characteristics of the electric energy meter measurement error, realize real-time tracking and effective correction of the error, greatly improve the measurement accuracy of the electric energy meter in a complex power environment, and provide strong technical support for the construction of new power systems such as smart grids.

[0202] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A synchronous electric energy meter error verification and correction method, characterized in that: The following steps are involved: S10, obtaining the electric metering data of the electric energy meter and the current, voltage and frequency signals collected by the standard electric energy detection equipment connected to the electric energy meter; S20, using a multivariate hybrid decomposition method to decompose the collected current and voltage signals into different characteristic components, and extracting fundamental components and harmonic components from the decomposed characteristic components; S30, calculating the active load, including fundamental wave active power and harmonic active power; S40, detecting the active resonance density at the peak and valley points of the load as a load waveform feature; S50, establishing an error model of the electric energy meter according to the load waveform characteristics, including voltage, current, and frequency as influencing factors; S60, substituting the detected current, voltage and frequency data into the error model to calculate the basic error of the electric energy meter; S70, using the collected voltage, current, and frequency signals to calculate the actual power data, compare it with the power meter reading, and calculate the difference between the basic error and the actual error; S80, using the difference to correct the error model parameters; S90, repeating steps S60 to S80 until the error model converges, and obtaining a corrected error model for error correction of the electric energy meter; The error model includes a long-term parameter branch, a medium-term parameter branch, a short-term parameter branch and a fusion branch. The long-term parameter branch is used to predict the metering error of the electric energy meter according to the load waveform characteristics obtained by long-term continuous collection and calculation. The medium-term parameter branch is used to predict the metering error of the electric energy meter according to the load waveform characteristics obtained by medium-term continuous collection and calculation. The short-term parameter branch is used to predict the metering error of the electric energy meter according to the load waveform characteristics obtained by short-term continuous collection and calculation. The fusion branch is used to integrate the errors predicted by the three time scales of long, medium and short to obtain the final error prediction value. The long-term parameter branch adopts a long short-term memory network, the input is a long-term load characteristic sequence, and the output is a long-term error component prediction; The mid-term parameter branch adopts a convolutional network, the input is the mid-term load characteristic matrix, and the output is the mid-term error component prediction; The short-term parameter branch adopts a feedforward network, the input is the short-term load characteristic vector, and the output is the short-term error component prediction; The fusion branch is used to perform weighted summation on the predictions of the three branches to obtain a final error prediction value.

2. A synchronous electric energy meter error verification and correction method according to claim 1, characterized in that: The short term is hours or days; the medium term is weeks or months; and the long term is months or years.

3. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for error verification and correction of a synchronizable electric energy meter as described in any one of claims 1-2.

4. An electronic device, characterized in that: A processor and a memory are provided, wherein the memory is used to store the step program of the method described in any one of claims 1 to 2, and the processor reads the memory to execute the steps.

Images