A dynamic perception-based intelligent electric energy meter state self-diagnosis method and electric energy meter
By constructing a dynamic health baseline model and a Gaussian kernel membership fusion algorithm, combined with time series analysis and adaptive compensation, the environmental adaptability and robustness issues of smart meter status diagnosis are solved, achieving highly accurate and self-healing smart meter status monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU INST OF METROLOGY
- Filing Date
- 2026-05-12
- Publication Date
- 2026-06-09
AI Technical Summary
Existing methods for diagnosing the condition of smart meters are inadequate in terms of environmental adaptability, multi-feature fusion, and robustness, resulting in high false alarm rates, unstable diagnostic results, and difficulty in accurately reflecting the health status of the meters.
By collecting state characteristics and environmental factor data through the sensing module, a dynamic health baseline model is constructed. The Gaussian kernel membership fusion algorithm and time series analysis are used, combined with weighted deviation and drift trend prediction, to achieve accurate quantification and adaptive compensation of multi-dimensional features.
It improves the diagnostic accuracy and robustness of smart energy meters in complex environments, reduces false alarm rates, achieves self-repairing closed loop, and extends equipment lifespan.
Smart Images
Figure CN122172108A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of fault diagnosis technology, specifically to a method for self-diagnosing the status of a smart energy meter based on dynamic sensing, and the energy meter itself. Background Technology
[0002] As the core terminal for data acquisition in smart grids, the reliability of smart meters directly affects the fairness of electricity metering and the accuracy of grid load management. Traditional periodic manual inspection and maintenance methods are no longer sufficient to meet the maintenance needs of massive numbers of metering devices; therefore, intelligent and automated self-diagnostic technologies have become an inevitable trend in the industry.
[0003] Currently, methods for diagnosing the condition of smart meters are typically based on static threshold judgment or simple trend analysis. However, existing technologies have certain drawbacks: First, they suffer from poor environmental adaptability and high false alarm rates. The parameter characteristics of the internal components of smart meters are highly sensitive to environmental factors such as temperature and humidity. Existing technologies often use fixed health thresholds, which cannot distinguish between normal physical characteristic deviations of components in extreme environments and actual performance degradation faults. For example, in high-temperature environments, the crystal oscillator frequency will experience normal physical drift; if a static threshold is used, it is very easy to misjudge this as a fault, leading to false alarms. On the other hand, relaxing the threshold may miss early minor hardware faults. Second, they lack the ability to accurately quantify through multi-feature fusion. The internal state of a smart meter involves multi-dimensional features such as voltage, frequency, ripple, and signal strength. Existing technologies mostly monitor single parameters in isolation, ignoring the inherent correlation between various feature parameters. Small changes in a single parameter may not be significant, but specific combinations of deviations in multiple parameters often indicate potential fault modes. Existing technologies struggle to accurately calculate the overall health deviation through multi-dimensional feature fusion. Finally, existing interval division models suffer from boundary jump defects. While some improved technologies introduce environmental interval divisions, they typically employ hard, either-or boundary divisions. When environmental factors fluctuate slightly near the boundaries of adjacent intervals, the baseline model parameters can experience drastic changes, leading to unstable diagnostic outputs and severely impacting the robustness of the diagnostic results. Therefore, there is an urgent need for a smart energy meter condition self-diagnosis method that can adapt to complex environmental changes, integrate multi-dimensional features, possess robustness assessment capabilities, and demonstrate safe self-repair capabilities. Summary of the Invention
[0004] The purpose of this invention is to provide a method for self-diagnosing the status of a smart energy meter based on dynamic sensing, and an energy meter in general, to solve the problems mentioned in the background art.
[0005] To address the aforementioned technical problems, this invention provides a method for self-diagnosing the status of a smart energy meter based on dynamic sensing, comprising:
[0006] S100: Through the sensing module inside the smart energy meter, it collects state characteristic data and environmental factor data during operation, and reads the historical state characteristic data and historical environmental factor data stored in the storage module.
[0007] Smart meters contain different types of hardware modules, including at least a sensing module, a storage module, a metering module, a communication module, and an MCU microcontroller.
[0008] The sensing module includes at least a temperature sensor, a humidity sensor, a voltage sampling circuit, and a frequency capture circuit, used to collect environmental factor data and state characteristic data.
[0009] Environmental factor data should include at least the internal temperature and humidity of the smart meter. Status characteristic data should include at least the reference voltage offset of the metering chip, the ripple coefficient of the internal DC power supply, the micro-offset of the crystal oscillator frequency, and the signal noise floor of the communication module.
[0010] The storage module is used to store historical state characteristic data, historical environmental factor data, and cached data of the MCU microcontroller.
[0011] The historical state feature data and historical environmental factor data are analyzed. The collected raw data are cleaned and normalized to eliminate differences in the dimensions of different features. The processed multidimensional historical data is then used to construct a time-series aligned feature matrix.
[0012] The Min-Max normalization method is used to normalize the data, mapping the original data to the [0,1] interval.
[0013] Because the values of the reference voltage offset of the metering chip in the status characteristic data are usually in the millivolt range, while the values of the micro offset of the crystal oscillator frequency may be in the Hertz range or smaller, and the values of the ambient temperature may be in the tens of degrees Celsius range.
[0014] If the weighted deviation is calculated directly, features with larger values will mask the influence of features with smaller values.
[0015] The normalization formula is as follows:
[0016] ;
[0017] In the formula, The original feature data, This represents the minimum value of this type of feature in historical data. This represents the maximum value of this type of feature in historical data. These are the normalized eigenvalues.
[0018] By normalizing the data, the differences between different monitoring indicators due to their different dimensions and orders of magnitude are eliminated, making the feature dimensions in the subsequently constructed health baseline model comparable, thereby ensuring the accuracy of the weighted deviation calculation.
[0019] S200. Based on historical environmental factor data and historical state characteristic data, environmental intervals are divided, and feature distribution parameters are extracted for different environmental intervals to construct a health baseline model that dynamically changes with the environment. Specifically, this includes:
[0020] S201. Divide the temperature and relative humidity in the historical environmental factor data into several continuous environmental intervals, each environmental interval corresponding to a specific combination of temperature and humidity ranges.
[0021] S202. For each environmental interval, select the corresponding historical state feature data, and calculate the expected vector and covariance matrix of each internal state feature under that environmental interval.
[0022] S203. The expected vector and covariance matrix are used as the health baseline parameters for this environmental interval. The health baseline parameters of all environmental intervals together constitute the energy meter health baseline model that changes dynamically with the environment.
[0023] By binding environmental ranges with characteristic distribution parameters, the interference of normal fluctuations in component parameters caused by seasonal or diurnal temperature variations on self-diagnostic results is eliminated, enabling the health baseline to accurately reflect the true health status under specific environmental conditions.
[0024] S300. Input the real-time collected state characteristic data and environmental factor data into the health baseline model, calculate the weighted deviation between the real-time state characteristics and the health baseline, and combine time series analysis algorithms to predict the drift trend of the state characteristics. Specifically, this includes:
[0025] S301. Locate the current environmental interval based on the real-time collected environmental factor data, and retrieve the expected vector and covariance matrix corresponding to the environmental interval.
[0026] When the real-time collected environmental factor data is at the boundary of an environmental interval, a Gaussian kernel-based dynamic membership degree fusion algorithm for environmental intervals is used to obtain the fusion expectation vector and the fusion covariance matrix, specifically:
[0027] Calculate real-time environmental data points For adjacent Membership weights of each environmental interval :
[0028] ;
[0029] In the formula, This represents real-time environmental data points based on the current temperature and humidity. For the first The center temperature and humidity coordinates of each adjacent environmental zone. For the first The central temperature and humidity coordinates of each adjacent environmental zone, among which For summation index variables.
[0030] For environmental data points With the center point The Euclidean distance between them. This is the environmental sensitivity coefficient, preset to a constant greater than 0, used to control the rate at which membership decreases with increasing distance. This represents the total number of adjacent environmental intervals.
[0031] Environmental sensitivity coefficient The value of is related to the density of the environmental interval division.
[0032] If the center points of adjacent environmental zones are close to each other, then Taking a larger value makes the membership degree decay faster with distance, so as to distinguish subtle environmental differences.
[0033] If the span between adjacent environmental intervals is large, then Choose a smaller value to smoothly transition to larger environmental changes.
[0034] The algorithm parameters can be flexibly adjusted according to different actual deployment environments to ensure the adaptability and optimal smoothing effect of the dynamic membership degree fusion algorithm in different application scenarios.
[0035] Based on the calculated membership weights For adjacent The expected vectors and covariance matrices of each environmental interval are weighted and fused to obtain the fused expected vector currently in actual use. and fusion covariance matrix :
[0036] ;
[0037] In the formula, and The first The expected vector and covariance matrix of each adjacent environmental interval.
[0038] By introducing a Gaussian kernel membership fusion mechanism, a smooth transition of baseline parameters between adjacent environmental intervals is achieved, eliminating the phenomenon of baseline parameter jumps at interval boundaries, and greatly improving the robustness and accuracy of state diagnosis under complex alternating environments.
[0039] S302. Construct a real-time feature vector from the real-time collected internal state feature data, and substitute it into the formula to calculate the weighted deviation between the real-time state features and the healthy baseline. :
[0040] ;
[0041] Weighted deviation This represents the overall degree to which the internal state of the smart meter deviates from the healthy baseline at the current moment. The larger the value, the more severe the abnormality or degradation of the meter's state.
[0042] In the formula, is the real-time feature vector, and is a column vector.
[0043] It consists of multidimensional internal state feature data collected at the current moment and processed by normalization.
[0044] is the expected vector corresponding to the current environmental range, and is a column vector. is the mean vector of historical health status characteristic data under the current temperature and humidity environmental range, representing the typical benchmark value when the electricity meter is in a healthy state under this specific environment.
[0045] This is the matrix transpose symbol, which converts column vectors into row vectors to meet the dimensionality requirements of matrix multiplication.
[0046] Let be the inverse of the covariance matrix corresponding to the current environmental interval, where It is the covariance matrix of historical health status characteristic data under the current environmental interval, reflecting the correlation between each characteristic variable and its own variance.
[0047] The purpose of inverting the matrix is to eliminate the interference of linear correlation that may exist between multidimensional features and to automatically standardize and scale features with different dimensions.
[0048] It is a diagonal weight matrix, where the diagonal elements correspond to the weights of each internal state feature. The weights of the crystal oscillator frequency micro-offset and the metering chip reference voltage offset are greater than the weights of the power supply ripple coefficient and the signal noise floor intensity.
[0049] weight matrix The weighting coefficients in the model are determined using the analytic hierarchy process (AHP).
[0050] With metrological accuracy as the core indicator, the slight deviation of the crystal oscillator frequency directly affects the time metrological accuracy, and the deviation of the reference voltage of the metrological chip directly affects the sampling accuracy of the ADC. Therefore, these parameters are given high weight.
[0051] The power supply ripple coefficient and signal noise floor intensity have a relatively small indirect impact on measurement accuracy and are therefore assigned lower weights.
[0052] In practical applications, the MCU microcontroller automatically loads the diagonal weight matrix according to a preset weight configuration table. .
[0053] By clarifying the specific source and assignment logic of the weights, and through differentiated assignment, the diagnostic algorithm can focus on the key parameters that affect the core metering performance of the electricity meter, and filter out the random fluctuations of non-key parameters.
[0054] It not only considers the correlation between multidimensional features, but also highlights the feature deviations that have the most significant impact on metering accuracy through a weight matrix, thus reflecting the minute performance deviations inside the electricity meter more sensitively and accurately.
[0055] S303, First-order difference exponential smoothing algorithm based on time series for weighted deviation Perform trend prediction and calculate drift trend values. Use trend prediction to remove transient noise interference and extract the long-term evolution direction of performance degradation.
[0056] The formula for calculating the drift trend value is as follows:
[0057] ;
[0058] In the formula, and Each represents the current time. and the previous moment The weighted deviation, and Each represents the current time. and the previous moment The drift trend value. The initial drift trend value is preset to zero.
[0059] Indicates the weighted deviation at the current time. The direction and rate of change. A positive value indicates that the state is continuously deteriorating, a negative value indicates that the state is recovering or improving, and zero indicates that the state is stable.
[0060] The trend smoothing coefficient is a constant with a value range of (0, 1), used to control the algorithm's sensitivity to the latest deviation changes.
[0061] The closer the value is to 1, the more sensitive the algorithm is to recent instantaneous changes and the faster it can detect sudden degradation. The closer the value is to 0, the stronger the smoothing effect of the algorithm, the better it can filter out short-term random fluctuations and reflect long-term slow degradation trends.
[0062] This indicates that the decay of historical trends is retained with weight. This ensures that trend predictions rely not only on current single fluctuations but also on a comprehensive judgment combining historical evolutionary inertia, thus improving the robustness of predictions.
[0063] S400 comprehensively assesses the health status of the energy meter based on weighted deviation and drift trend, performs self-diagnosis, outputs diagnostic results, and adjusts the compensation parameters of the internal metering and communication modules of the energy meter according to the diagnostic results. Specifically, this includes:
[0064] S401, Set the first warning threshold and the second fault threshold, when the weighted deviation... Values greater than the first warning threshold and drift trend values A value greater than zero indicates an early performance degradation state. When the weighted deviation... When the value exceeds the second fault threshold, it is diagnosed as a local hardware fault state.
[0065] S402. When diagnosed as an early performance degradation state, compensation parameters are adaptively generated based on the direction and magnitude of the deviation in each dimension of the real-time feature vector. The compensation formula is as follows:
[0066] ;
[0067] In the formula, The adjusted compensation parameter can be either a scalar or a vector. This parameter is ultimately sent to the gain calibration register of the metering module or the transmit power control unit of the communication module to compensate for the offset caused by hardware degradation.
[0068] The initial baseline compensation parameter represents the default basic compensation value of each module in the smart energy meter under factory settings or the last health calibration state.
[0069] This is a proportional compensation coefficient matrix / vector whose dimensions match the eigenvector, used to adjust the current feature bias. The magnitude of the deviation is compensated proportionally in real time. The larger the deviation, the greater the correction amount generated by the proportional compensation term, achieving rapid correction.
[0070] This is an integral compensation coefficient matrix / vector used to accumulate compensation for long-term steady-state small deviations. Even if the current single deviation is very small, as long as the deviation direction persists, the integral term will continuously accumulate over time, eventually generating sufficient compensation to eliminate the steady-state error.
[0071] This is the current feature deviation vector, i.e., the real-time feature vector. With health expectation vector The difference reflects the direction and magnitude of the deviation of various indicators from the normal benchmark.
[0072] This is the moment when early performance degradation is detected. From From time to the present time The discrete-time cumulative sum of the eigenvalues, where and The first The real-time feature vector and expected vector at each time step.
[0073] The accumulation of the integral term is not unlimited. When the state is detected to have returned to normal, the integral accumulator is automatically cleared to zero and the compensation is stopped, so as to avoid the historical accumulated deviation from interfering with the normal working state.
[0074] S403. Employ a dynamic attenuation mechanism for compensation effectiveness to calculate the final compensation parameters to be issued. The final compensation parameters will be issued. The parameters are sent to the metering module and communication module to adjust and compensate for metering gain and communication transmission power, suppressing accuracy deviations and implementing a self-healing closed-loop performance mechanism. The final calculation formula for the compensation parameters is as follows:
[0075] ;
[0076] In the formula, For the current moment The weighted deviation, This is the set second fault threshold. is the decay curvature exponent, which is a positive integer. This is the function for finding the maximum value.
[0077] Decay Curvature Index Determine the speed of the compensation reduction. When At that time, the attenuation factor decreases linearly with the deviation, and the reduction rate is relatively slow.
[0078] when When the deviation is greater than or equal to the fault threshold, the attenuation factor decreases sharply and nonlinearly, meaning that the compensation capability is quickly cut off when the deviation approaches the fault threshold.
[0079] This approach avoids masking the risk of failure through overcompensation when hardware is on the verge of failure, while ensuring that software compensation capabilities are maximized to extend the equipment's lifespan in the early stages of a failure.
[0080] The ratio of deviation to fault threshold is incorporated into the calculation of the attenuation factor. When the deviation is small, the attenuation factor approaches 1, and the compensation is fully distributed.
[0081] When the deviation approaches the fault threshold, the attenuation factor drops rapidly to 0 according to the exponential curve, actively reducing and terminating the compensation, thereby preventing software overcompensation from masking the real hardware fault and achieving a precise balance between self-repair closed loop and safety early warning.
[0082] When the S403 compensation efficiency dynamic attenuation mechanism is executed, if the attenuation factor calculation result is 0, the integral accumulator will be forcibly reset, marking the end of this self-repair closed loop and entering the standby monitoring or fault alarm state.
[0083] After hardware failure is eliminated or the system is reset, historical accumulated errors should be cleared in a timely manner to ensure the real-time performance and effectiveness of the compensation mechanism and avoid overcompensation or system oscillation caused by historical residual integral values.
[0084] The present invention also provides a smart energy meter, which has different types of functional modules inside, including at least a sensing module, a storage module, a communication module, a metering module and an MCU microcontroller;
[0085] The sensing module includes at least a temperature sensor, a humidity sensor, a voltage sampling circuit, and a frequency capture circuit, used to collect environmental factor data and state characteristic data;
[0086] The storage module is used to store historical state characteristic data, historical environmental factor data, and cached data of the MCU microcontroller.
[0087] The communication module has different preset data transmission and reception frequencies in different working modes, and adaptively adjusts the transmission power compensation according to the MCU instructions; the metering module dynamically adjusts its internal gain according to the compensation parameters issued by the MCU.
[0088] The MCU microcontroller is connected to the sensing module, storage module, communication module and metering module to execute trend prediction and compensation parameter generation algorithms.
[0089] Compared with the prior art, the beneficial effects achieved by the present invention are:
[0090] 1. Existing technologies typically use hard boundaries to divide environmental intervals. When temperature and humidity fluctuate slightly at the interval boundaries, the invoked health baseline parameters can change drastically, leading to false alarms in self-diagnosis results. This invention introduces a Gaussian kernel membership fusion mechanism to achieve a smooth transition of baseline parameters between adjacent environmental intervals, eliminating the phenomenon of baseline parameter jumps at interval boundaries, and greatly improving the robustness and accuracy of state diagnosis in complex alternating environments.
[0091] 2. Traditional compensation algorithms continuously increase the compensation amount during early hardware degradation. However, when degradation approaches the hardware failure threshold, excessive software compensation not only fails to restore physical performance but also masks the true hardware fault, potentially leading to loss of control. This invention incorporates the ratio of deviation to the fault threshold into the attenuation factor calculation. When the deviation is small, the attenuation factor approaches 1, and full compensation is applied. As the deviation approaches the fault threshold, the attenuation factor rapidly decreases to 0 according to an exponential curve, proactively reducing and terminating compensation application. This prevents excessive software compensation from masking the true hardware fault, achieving a precise balance between self-healing closed loop and safety early warning. Attached Figure Description
[0092] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0093] Figure 1 This is a flowchart illustrating a self-diagnosis method for the status of a smart energy meter based on dynamic sensing, according to the present invention. Detailed Implementation
[0094] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0095] Example 1: Please refer to Figure 1 This invention provides a method for self-diagnosing the status of a smart energy meter based on dynamic sensing, comprising:
[0096] S100: Through the sensing module inside the smart energy meter, it collects state characteristic data and environmental factor data during operation, and reads the historical state characteristic data and historical environmental factor data stored in the storage module.
[0097] In practice, smart meters are equipped with different types of hardware modules, including at least a sensing module, a storage module, a metering module, a communication module, and an MCU microcontroller.
[0098] The sensing module includes at least a temperature sensor, a humidity sensor, a voltage sampling circuit, and a frequency capture circuit, used to collect environmental factor data and state characteristic data.
[0099] Environmental factor data should include at least the internal temperature and humidity of the smart meter. Status characteristic data should include at least the reference voltage offset of the metering chip, the ripple coefficient of the internal DC power supply, the micro-offset of the crystal oscillator frequency, and the signal noise floor of the communication module.
[0100] The storage module is used to store historical state characteristic data, historical environmental factor data, and cached data of the MCU microcontroller.
[0101] The historical state feature data and historical environmental factor data are analyzed. The collected raw data are cleaned and normalized to eliminate differences in the dimensions of different features. The processed multidimensional historical data is then used to construct a time-series aligned feature matrix.
[0102] In the specific implementation process, the Min-Max standardization method is used for normalization processing, mapping the original data to the [0,1] interval.
[0103] Because the values of the reference voltage offset of the metering chip in the status characteristic data are usually in the millivolt range, while the values of the micro offset of the crystal oscillator frequency may be in the Hertz range or smaller, and the values of the ambient temperature may be in the tens of degrees Celsius range.
[0104] If the weighted deviation is calculated directly, features with larger values will mask the influence of features with smaller values.
[0105] The normalization formula is as follows:
[0106] ;
[0107] In the formula, The original feature data, This represents the minimum value of this type of feature in historical data. This represents the maximum value of this type of feature in historical data. These are the normalized eigenvalues.
[0108] By normalizing the data, the differences between different monitoring indicators due to their different dimensions and orders of magnitude are eliminated, making the feature dimensions in the subsequently constructed health baseline model comparable, thereby ensuring the accuracy of the weighted deviation calculation.
[0109] S200. Based on historical environmental factor data and historical state characteristic data, environmental intervals are divided, and feature distribution parameters are extracted for different environmental intervals to construct a health baseline model that dynamically changes with the environment. Specifically, this includes:
[0110] S201. Divide the temperature and relative humidity in the historical environmental factor data into several continuous environmental intervals, each environmental interval corresponding to a specific combination of temperature and humidity ranges.
[0111] S202. For each environmental interval, select the corresponding historical state feature data, and calculate the expected vector and covariance matrix of each internal state feature under that environmental interval.
[0112] S203. The expected vector and covariance matrix are used as the health baseline parameters for this environmental interval. The health baseline parameters of all environmental intervals together constitute the energy meter health baseline model that changes dynamically with the environment.
[0113] In the specific implementation process, by binding the environmental range with the characteristic distribution parameters, the interference of normal fluctuations in component parameters caused by seasonal or diurnal temperature variations on the self-diagnosis results is eliminated, so that the health baseline accurately reflects the real health status under specific environments.
[0114] S300. Input the real-time collected state characteristic data and environmental factor data into the health baseline model, calculate the weighted deviation between the real-time state characteristics and the health baseline, and combine time series analysis algorithms to predict the drift trend of the state characteristics. Specifically, this includes:
[0115] S301. Locate the current environmental interval based on the real-time collected environmental factor data, and retrieve the expected vector and covariance matrix corresponding to the environmental interval.
[0116] In the specific implementation process, when the real-time collected environmental factor data is at the boundary of an environmental interval, a Gaussian kernel-based dynamic membership degree fusion algorithm for environmental intervals is used to obtain the fusion expectation vector and the fusion covariance matrix, specifically as follows:
[0117] Calculate real-time environmental data points For adjacent Membership weights of each environmental interval :
[0118] ;
[0119] In the formula, This represents real-time environmental data points based on the current temperature and humidity. For the first The center temperature and humidity coordinates of each adjacent environmental zone. For the first The central temperature and humidity coordinates of each adjacent environmental zone, among which For summation index variables.
[0120] For environmental data points With the center point The Euclidean distance between them. This is the environmental sensitivity coefficient, preset to a constant greater than 0, used to control the rate at which membership decreases with increasing distance. This represents the total number of adjacent environmental intervals.
[0121] In the specific implementation process, the environmental sensitivity coefficient The value of is related to the density of the environmental interval division.
[0122] If the center points of adjacent environmental intervals are close to each other (i.e., the intervals are divided into smaller sections), then Take the larger value (e.g.) This allows membership to decay more rapidly with distance, thus distinguishing subtle environmental differences.
[0123] If the span between adjacent environmental intervals is large, then Take the smaller value (e.g.) This allows for a smoother transition from significant environmental changes. For example, when the temperature range is divided into 5°C intervals, A value of 0.5 can be taken. When divided into intervals of 10°C, 0.2 is acceptable.
[0124] The algorithm parameters are flexibly adjusted according to the actual deployment environment (such as temperature-controlled computer rooms or variable outdoor environments) to ensure the adaptability and optimal smoothing effect of the dynamic membership degree fusion algorithm in different application scenarios.
[0125] Based on the calculated membership weights For adjacent The expected vectors and covariance matrices of each environmental interval are weighted and fused to obtain the fused expected vector currently in actual use. and fusion covariance matrix :
[0126] ;
[0127] In the formula, and The first The expected vector and covariance matrix of each adjacent environmental interval.
[0128] In the specific implementation process, by introducing a Gaussian kernel membership fusion mechanism, a smooth transition of baseline parameters between adjacent environmental intervals is achieved, eliminating the phenomenon of baseline parameter jumps at interval boundaries, which greatly improves the robustness and accuracy of state diagnosis under complex alternating environments.
[0129] S302. Construct a real-time feature vector from the real-time collected internal state feature data, and substitute it into the formula to calculate the weighted deviation between the real-time state features and the healthy baseline. :
[0130] ;
[0131] Weighted deviation This represents the overall degree to which the internal state of the smart meter deviates from the healthy baseline at the current moment. The larger the value, the more severe the abnormality or degradation of the meter's state.
[0132] In the formula, is the real-time feature vector, and is a column vector.
[0133] It consists of multi-dimensional internal state feature data collected and normalized at the current moment (such as the reference voltage offset of the metering chip, the DC power supply ripple coefficient, the micro offset of the crystal oscillator frequency, and the signal noise floor intensity of the communication module).
[0134] is the expected vector corresponding to the current environmental range, and is a column vector. is the mean vector of historical health status characteristic data under the current temperature and humidity environmental range, representing the typical benchmark value when the electricity meter is in a healthy state under this specific environment.
[0135] This is the matrix transpose symbol, which converts column vectors into row vectors to meet the dimensionality requirements of matrix multiplication.
[0136] Let be the inverse of the covariance matrix corresponding to the current environmental interval, where It is the covariance matrix of historical health status characteristic data under the current environmental interval, reflecting the correlation between each characteristic variable and its own variance.
[0137] The purpose of inverting the matrix is to eliminate the interference of linear correlation that may exist between multidimensional features and to automatically standardize and scale features with different dimensions.
[0138] It is a diagonal weight matrix, where the diagonal elements correspond to the weights of each internal state feature. The weights of the crystal oscillator frequency micro-offset and the metering chip reference voltage offset are greater than the weights of the power supply ripple coefficient and the signal noise floor intensity.
[0139] In the specific implementation process, the weight matrix The weighting coefficients in the model are determined using the Analytic Hierarchy Process (AHP).
[0140] With measurement accuracy as the core indicator, the slight deviation of the crystal oscillator frequency directly affects the time measurement accuracy, and the deviation of the reference voltage of the measurement chip directly affects the sampling accuracy of the ADC. Therefore, they are given high weights (e.g., set to 0.35 and 0.35 respectively).
[0141] The power supply ripple coefficient and signal noise floor intensity have a relatively small indirect impact on measurement accuracy, so they are given low weights (e.g., set to 0.15 and 0.15 respectively).
[0142] In practical applications, the MCU microcontroller automatically loads the diagonal weight matrix according to a preset weight configuration table. .
[0143] By clarifying the specific source and assignment logic of the weights, and through differentiated assignment, the diagnostic algorithm can focus on the key parameters that affect the core metering performance of the electricity meter, and filter out the random fluctuations of non-key parameters.
[0144] It not only considers the correlation between multidimensional features, but also highlights the feature deviations that have the most significant impact on metering accuracy through a weight matrix, thus reflecting the minute performance deviations inside the electricity meter more sensitively and accurately.
[0145] S303, First-order difference exponential smoothing algorithm based on time series for weighted deviation Perform trend prediction and calculate drift trend values. Use trend prediction to remove transient noise interference and extract the long-term evolution direction of performance degradation.
[0146] The formula for calculating the drift trend value is as follows:
[0147] ;
[0148] In the formula, and Each represents the current time. and the previous moment The weighted deviation, and Each represents the current time. and the previous moment The drift trend value. The initial drift trend value is preset to zero.
[0149] Indicates the weighted deviation at the current time. The direction and rate of change. A positive value indicates that the state is continuously deteriorating (the deviation is increasing), a negative value indicates that the state is recovering or improving, and zero indicates that the state is stable.
[0150] The trend smoothing coefficient is a constant with a value range of (0, 1), used to control the algorithm's sensitivity to the latest deviation changes.
[0151] The closer the value is to 1, the more sensitive the algorithm is to recent instantaneous changes and the faster it can detect sudden degradation. The closer the value is to 0, the stronger the smoothing effect of the algorithm, the better it can filter out short-term random fluctuations and reflect long-term slow degradation trends.
[0152] This indicates that the decay of historical trends is retained with weight. This ensures that trend predictions rely not only on current single fluctuations but also on a comprehensive judgment combining historical evolutionary inertia, thus improving the robustness of predictions.
[0153] S400 comprehensively assesses the health status of the energy meter based on weighted deviation and drift trend, performs self-diagnosis, outputs diagnostic results, and adjusts the compensation parameters of the internal metering and communication modules of the energy meter according to the diagnostic results. Specifically, this includes:
[0154] S401, Set the first warning threshold and the second fault threshold, when the weighted deviation... Values greater than the first warning threshold and drift trend values A value greater than zero indicates an early performance degradation state. When the weighted deviation... When the value exceeds the second fault threshold, it is diagnosed as a local hardware fault state.
[0155] S402. When diagnosed as an early performance degradation state, compensation parameters are adaptively generated based on the direction and magnitude of the deviation in each dimension of the real-time feature vector. The compensation formula is as follows:
[0156] ;
[0157] In the formula, The adjusted compensation parameter can be either a scalar or a vector (depending on the number of modules to be adjusted). This parameter is ultimately sent to the gain calibration register of the metering module or the transmit power control unit of the communication module to compensate for the offset caused by hardware degradation.
[0158] The initial baseline compensation parameter represents the default basic compensation value of each module in the smart energy meter under factory settings or the last health calibration state.
[0159] This is a proportional compensation coefficient matrix / vector whose dimensions match the eigenvector, used to adjust the current feature bias. The magnitude of the deviation is compensated proportionally in real time. The larger the deviation, the greater the correction amount generated by the proportional compensation term, achieving rapid correction.
[0160] This is an integral compensation coefficient matrix / vector used to accumulate compensation for long-term steady-state small deviations. Even if the current single deviation is very small, as long as the deviation direction persists, the integral term will continuously accumulate over time, eventually generating sufficient compensation to eliminate the steady-state error.
[0161] This is the current feature deviation vector, i.e., the real-time feature vector. With health expectation vector The difference reflects the direction (positive or negative) and magnitude (size) of the deviation of various indicators from the normal benchmark.
[0162] This is the moment when early performance degradation is detected. From From time to the present time The discrete-time cumulative sum of the eigenvalues, where and The first The real-time feature vector and expected vector at each time step.
[0163] In practice, the accumulation of the integral term is not infinite; it continues until the state is detected to have returned to normal (i.e., ...). When the value is less than the first warning threshold, the integral accumulator is automatically cleared to zero and compensation is stopped to avoid interference from historical accumulated deviations on normal operation.
[0164] S403. Employ a dynamic attenuation mechanism for compensation effectiveness to calculate the final compensation parameters to be issued. The final compensation parameters will be issued. The parameters are sent to the metering module and communication module to adjust and compensate for metering gain and communication transmission power, suppressing accuracy deviations and implementing a self-healing closed-loop performance mechanism. The final calculation formula for the compensation parameters is as follows:
[0165] ;
[0166] In the formula, For the current moment The weighted deviation, This is the set second fault threshold. is the decay curvature exponent, which is a positive integer. This is the function for finding the maximum value.
[0167] In practical implementation, the attenuation curvature index Determine the speed of the compensation reduction. When At that time, the attenuation factor decreases linearly with the deviation, and the reduction rate is relatively slow.
[0168] when When the deviation is greater than or equal to the fault threshold, the attenuation factor decreases sharply and nonlinearly, meaning that the compensation capability is quickly cut off when the deviation approaches the fault threshold.
[0169] This approach avoids masking the risk of failure through overcompensation when hardware is on the verge of failure, while ensuring that software compensation capabilities are maximized to extend the equipment's lifespan in the early stages of a failure.
[0170] The ratio of deviation to fault threshold is incorporated into the calculation of the attenuation factor. When the deviation is small, the attenuation factor approaches 1, and the compensation is fully distributed.
[0171] When the deviation approaches the fault threshold, the attenuation factor drops rapidly to 0 according to the exponential curve, actively reducing and terminating the compensation, thereby preventing software overcompensation from masking the real hardware fault and achieving a precise balance between self-repair closed loop and safety early warning.
[0172] When the S403 compensation efficiency dynamic attenuation mechanism is executed, if the attenuation factor calculation result is 0, the integral accumulator will be forcibly reset, marking the end of this self-repair closed loop and entering the standby monitoring or fault alarm state.
[0173] In the specific implementation process, after the hardware fault is eliminated or the system is reset, the historical accumulated error is cleared in a timely manner to ensure the real-time performance and effectiveness of the compensation mechanism and avoid overcompensation or system oscillation caused by historical residual integral values.
[0174] The present invention also provides a smart energy meter, which has different types of functional modules inside, including at least a sensing module, a storage module, a communication module, a metering module and an MCU microcontroller;
[0175] The sensing module includes at least a temperature sensor, a humidity sensor, a voltage sampling circuit, and a frequency capture circuit, used to collect environmental factor data and state characteristic data;
[0176] The storage module is used to store historical state characteristic data, historical environmental factor data, and cached data of the MCU microcontroller.
[0177] The communication module has different preset data transmission and reception frequencies in different working modes, and adaptively adjusts the transmission power compensation according to the MCU instructions; the metering module dynamically adjusts its internal gain according to the compensation parameters issued by the MCU.
[0178] The MCU microcontroller is connected to the sensing module, storage module, communication module and metering module to execute trend prediction and compensation parameter generation algorithms.
[0179] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0180] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for self-diagnosing the status of a smart energy meter based on dynamic sensing, characterized in that: The method includes: S100: Through the sensing module inside the smart energy meter, it collects state characteristic data and environmental factor data during operation, and reads the historical state characteristic data and historical environmental factor data stored in the storage module. S200. Based on historical environmental factor data and historical state characteristic data, environmental intervals are divided, and feature distribution parameters are extracted for different environmental intervals to construct a health baseline model that changes dynamically with the environment. S300. Input the real-time collected state characteristic data and environmental factor data into the health baseline model, calculate the weighted deviation between the real-time state characteristics and the health baseline, and combine the time series analysis algorithm to predict the drift trend of the state characteristics. S400 comprehensively assesses the health status of the energy meter based on the weighted deviation and drift trend, performs status self-diagnosis and outputs the diagnostic results, and adjusts the compensation parameters of the internal metering and communication modules of the energy meter according to the diagnostic results.
2. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 1, characterized in that: In the S100, the smart energy meter has different types of hardware modules, including at least a sensing module, a storage module, a metering module, a communication module, and an MCU microcontroller. The sensing module includes at least a temperature sensor, a humidity sensor, a voltage sampling circuit, and a frequency capture circuit, used to collect environmental factor data and state characteristic data; Environmental factor data includes at least the temperature and humidity inside the smart meter; state characteristic data includes at least the reference voltage offset of the metering chip, the ripple coefficient of the internal DC power supply, the micro offset of the crystal oscillator frequency, and the signal noise floor intensity of the communication module. The storage module is used to store historical state characteristic data, historical environmental factor data, and cached data of the MCU microcontroller. The historical state feature data and historical environmental factor data are analyzed. The collected raw data are cleaned and normalized to eliminate differences in the dimensions of different features. The processed multidimensional historical data is then used to construct a time-series aligned feature matrix.
3. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 2, characterized in that: S200 includes: S201. Divide the temperature and relative humidity in the historical environmental factor data into several continuous environmental intervals, with each environmental interval corresponding to a combination of temperature and humidity ranges; S202. For each environmental interval, select the corresponding historical state feature data, and calculate the expected vector and covariance matrix of each internal state feature under that environmental interval. S203. The expected vector and covariance matrix are used as the health baseline parameters for this environmental interval. The health baseline parameters of all environmental intervals together constitute the energy meter health baseline model that changes dynamically with the environment.
4. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 3, characterized in that: The S300 includes: S301. Locate the current environmental interval based on the real-time collected environmental factor data, and retrieve the expected vector and covariance matrix corresponding to the environmental interval. S302. Construct a real-time feature vector from the real-time collected internal state feature data, and substitute it into the formula to calculate the weighted deviation between the real-time state features and the healthy baseline. : ； In the formula, For real-time feature vectors, This represents the expected vector corresponding to the current environmental interval. This is the matrix transpose symbol. It is the inverse of the covariance matrix corresponding to the current environmental interval. It is a diagonal weight matrix; diagonal weight matrix The diagonal elements in the diagram correspond to the weights of each internal state feature, with the weights of the crystal oscillator frequency micro-offset and the metering chip reference voltage offset being greater than the weights of the power supply ripple coefficient and the signal noise floor intensity. S303, First-order difference exponential smoothing algorithm based on time series for weighted deviation Perform trend prediction and calculate drift trend value; use trend prediction to remove transient noise interference and extract the long-term evolution direction of performance degradation.
5. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 4, characterized in that: In S301, when the real-time collected environmental factor data is at the boundary of an environmental interval, a Gaussian kernel-based dynamic membership degree fusion algorithm for environmental intervals is used to obtain the fusion expectation vector and the fusion covariance matrix, specifically: Calculate real-time environmental data points For adjacent Membership weights of each environmental interval : ； In the formula, These are real-time environmental data points consisting of current temperature and humidity. For the first The center temperature and humidity coordinates of each adjacent environmental zone; For the first The central temperature and humidity coordinates of each adjacent environmental zone, among which For summation index variables; For environmental data points With the center point The Euclidean distance between them; This is the environmental sensitivity coefficient, preset to a constant greater than 0, used to control the rate at which membership decreases with increasing distance; This represents the total number of adjacent environmental intervals; Based on the calculated membership weights For adjacent The expected vectors and covariance matrices of each environmental interval are weighted and fused to obtain the fused expected vector currently in actual use. and fusion covariance matrix : ； In the formula, and The first The expected vector and covariance matrix of each adjacent environmental interval.
6. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 4, characterized in that: In S303, the formula for calculating the drift trend value is as follows: ； In the formula, and Each represents the current time. and the previous moment The weighted deviation, and Each represents the current time. and the previous moment The drift trend value; The trend smoothing coefficient is a constant with a value range of (0, 1).
7. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 4, characterized in that: The S400 includes: S401, Set the first warning threshold and the second fault threshold, when the weighted deviation... Values greater than the first warning threshold and drift trend values When the deviation is greater than zero, it is diagnosed as an early performance degradation state; when the weighted deviation is greater than zero, it is diagnosed as an early performance degradation state. When the value exceeds the second fault threshold, it is diagnosed as a local hardware fault state; S402. When diagnosed as an early performance degradation state, compensation parameters are adaptively generated based on the direction and magnitude of the deviation in each dimension of the real-time feature vector; the compensation formula is as follows: ； In the formula, The adjusted compensation parameters, These are the initial reference compensation parameters. This is the proportional compensation coefficient. This is the integral compensation coefficient. This is the current feature deviation vector; This is the moment when early performance degradation is detected; From From time to the present time The discrete-time cumulative sum of the eigenvalues, where and The first Real-time feature vector and expected vector at time step; S403. Employ a dynamic attenuation mechanism for compensation effectiveness to calculate the final compensation parameters to be issued. The final compensation parameters will be issued. The data is sent to the metering module and communication module to adjust and compensate the metering gain and communication transmission power, suppress accuracy deviation, and perform a self-healing closed loop for performance.
8. The method for self-diagnosing the status of a smart energy meter based on dynamic sensing according to claim 7, characterized in that: In S403, the final calculation formula for the compensation parameters is as follows: ； In the formula, For the current moment The weighted deviation, The second fault threshold is set; The decay curvature exponent is a positive integer. This is the function for finding the maximum value.
9. A smart energy meter, characterized in that: The smart energy meter is equipped with different types of functional modules to implement the self-diagnosis method for the status of a smart energy meter based on dynamic sensing as described in any one of claims 1 to 8; the functional modules include at least a sensing module, a storage module, a communication module, a metering module, and an MCU microcontroller; The sensing module includes at least a temperature sensor, a humidity sensor, a voltage sampling circuit, and a frequency capture circuit, used to collect environmental factor data and state characteristic data. The storage module is used to store historical state feature data, historical environmental factor data, and cached data of the MCU microcontroller. The communication module has different preset data transmission and reception frequencies in different working modes, and adaptively adjusts the transmission power compensation according to MCU instructions; the metering module dynamically adjusts its internal gain according to the compensation parameters issued by the MCU. The MCU microcontroller is connected to the sensing module, storage module, communication module and metering module, and is used to execute trend prediction and compensation parameter generation algorithms.