A method and system for evaluating the life of a meter relay

Abstract

The application relates to the field of measuring electric variables, in particular to a meter relay service life evaluation method and system, which comprises the following steps: obtaining current data of a relay under different historical loads, calculating current distortion damage degrees of different loads, and obtaining a current distortion damage sequence sorted according to the sizes of the load currents; adjusting the current distortion damage sequence to obtain a load current-distortion damage degree curve, obtaining a threshold value preset at a load grade boundary point, and then obtaining load damage amounts of different load grades; calculating a service life loss degree of the relay according to the load damage amounts of different load grades; obtaining historical data of a plurality of relays to train a preset LSTM model, optimizing model parameters through a preset loss function, and obtaining a service life estimation model; obtaining real-time data of the relay, and inputting the real-time data into the service life estimation model to complete real-time remaining service life evaluation of the relay. The application has the effect of accurately evaluating the service life of the relay.

Description

Technical Field

[0001] This application relates to the field of measuring electrical variables, and in particular to a method and system for assessing the lifespan of an electricity meter relay. Background Technology

[0002] Against the backdrop of the accelerated penetration of smart grids and the transformation of electricity management towards lean management, the electricity meter relay, as the core execution unit for electricity metering and circuit control, has emerged and continues to iterate. Its appearance breaks the limitations of traditional electricity meters that "only measure, without control," becoming a crucial hub connecting the electricity metering terminal and the user's electricity circuit. Its lifespan (actual remaining cycles) directly determines the meter's operational reliability and maintenance costs.

[0003] Currently, the industry's assessment of relay lifespan largely relies on the traditional empirical value of "fixed number of opening and closing cycles," without considering the varying impacts of different load levels on relay damage during actual use.

[0004] Using only a fixed number of opening and closing cycles for evaluation can lead to serious errors in the assessment of relay life (actual remaining cycles). For example, high loads cause far more damage to relays than low loads. Existing technologies lack data-based objectivity and accuracy in classifying load levels, and the construction of prediction models is not well-suited to complex sample scenarios with multiple load levels and differentiated life prediction sensitivity, resulting in low accuracy in relay maintenance. Summary of the Invention

[0005] To accurately assess the lifespan of relays, this application provides a method and system for assessing the lifespan of electricity meter relays.

[0006] In a first aspect, this application provides a method for evaluating the lifespan of an electricity meter relay, including:

[0007] Obtain historical relay current data under different loads, fit a current distortion curve based on the current data, extract distortion parameters, and calculate the current distortion damage degree; among which, the distortion parameters include: distortion time and peak current.

[0008] The current distortion damage degree is sorted based on the magnitude of the load current to obtain a current distortion damage sequence. The current distortion damage sequence is then cross-validated to calculate initial smoothing parameters. The load current is divided into multiple level intervals, and the initial smoothing parameters are adjusted based on each level interval to obtain adjusted parameters. A load current-distortion damage degree curve is fitted based on the adjusted parameters. The second derivative of the load current-distortion damage degree curve is used to analyze the threshold of the load level boundary. Points where the absolute value of the second derivative is greater than the threshold are selected as the boundary points of the load levels, and load levels are defined. The mean value of all current distortion damage degrees in each load level is calculated to obtain the load damage amount for the corresponding load level. The relay life loss degree is obtained based on the load damage amounts for all load levels.

[0009] The load level, load damage amount, and life loss degree are used to construct feature vectors. The historical feature vectors of several relays are obtained to train the preset LSTM model. The model parameters are optimized by the preset loss function to obtain the life prediction model.

[0010] The real-time feature vector of the relay is obtained as input to the life prediction model, and the output prediction normalized remaining number of times is used to obtain the remaining life based on the prediction normalized remaining number of times, thus completing the evaluation of the real-time remaining life of the relay.

[0011] Optionally, the current distortion curve is calculated as follows: the current range of the meter is evenly divided into several small segments, and the maximum current value within the range is used as the metering point to obtain multiple load metering points. The timing data of the relay from open to closed under different loads is collected to obtain multiple current timing sequences. The steady-state current value of each current timing sequence is collected. The rise amplitude parameter, rise time constant, decay amplitude parameter, and decrease amplitude parameter are solved by the least squares method. Based on each parameter, the multiple timing data are fitted with a double exponential function, and the current distortion curve is obtained by combining the steady-state current value.

[0012] The optional method for calculating the distortion parameter is as follows: obtain the fitted current value based on the current distortion curve, and take the time interval from the relay closing moment to the moment when the fitted current value tends to stabilize as the distortion time; select any load as the target load, repeatedly collect the current data of the relay closing under the target load to obtain multiple candidate current peaks, and calculate the ratio of the standard value to the mean of the multiple candidate current peaks to obtain the coefficient of variation; set an initial noise tolerance threshold, and obtain the target load dynamic noise tolerance threshold by multiplying the initial noise tolerance threshold with the hyperbolic tangent function of the target load; select the candidate current peak with a coefficient of variation less than the dynamic noise tolerance threshold as the true peak, select the maximum value among the true peaks as the current peak of the target load, and similarly obtain the current peaks of all loads.

[0013] Optionally, the current distortion damage degree is calculated as follows: the product of the square of the peak current of the target load and the corresponding distortion time is taken as the total damage amount, and the product of the square of the minimum peak current and the minimum distortion time among the actual peak currents of the target load is taken as the reference damage amount; the ratio of the total damage amount to the reference damage amount is calculated to obtain the current distortion damage degree of the target load; similarly, the current distortion damage degree of all loads is obtained.

[0014] Optionally, the calculation method of the load current-distortion damage curve is as follows: the data points in the current distortion damage sequence are cross-validated to calculate the initial smoothing parameter, the load current is divided into multiple level intervals, any level interval is selected as the target level interval, the ratio of the current distortion damage variance of the target level interval to the variance of all current distortion damage is taken as the first ratio, the product of the first ratio and the initial smoothing parameter is calculated to obtain the smoothing parameter of the target level interval, and the smoothing parameters of all level intervals are obtained in the same way. Based on the multiple smoothing parameters, the current distortion damage sequence is fitted with a cubic spline smoothing curve function to obtain the load current-distortion damage curve.

[0015] Optionally, the load level threshold is calculated as follows: calculate the absolute median difference of the absolute values ​​of the second derivatives of all data in the load current-distortion damage curve, and calculate the sum of twice the absolute median difference and the median of the absolute values ​​of the second derivatives of all data to obtain the load level threshold.

[0016] Optionally, the calculation method for the life loss of the relay is as follows: according to the different load currents, the data points in the current distortion damage sequence are divided into corresponding load levels, any load level is selected as the target load level, and the load damage amount of the target load level is obtained by summing and averaging the current distortion damage of all loads in the target load level.

[0017] The lifetime wear factor of the target load level is obtained by multiplying the load damage amount of the target load level by the sum of the distortion time of the target load level. The lifetime wear factor of the relay is obtained by summing the lifetime wear factors of all load levels.

[0018] Optionally, the loss function is calculated as follows:

[0019] Record the number of closures in units of one closure. Select any relay as the target relay and collect the different cumulative number of closures of the target relay under different loads. Calculate the load damage and life loss corresponding to different load levels. Use the different cumulative number of closures of the target relay under different loads, the load damage and life loss corresponding to different load levels as historical feature vectors.

[0020] The remaining number of times is obtained by subtracting the current number of closures from the rated total number of mechanical closures. The ratio of the remaining number of times to the rated total number of mechanical closures is taken as the actual normalized remaining number of times. The high damage weighting factor and the remaining number of times weighting factor are calculated.

[0021] The square of the difference between the predicted normalized remaining number of times and the actual normalized remaining number of times is taken as the first difference. The product of the first difference, the remaining number of times weight factor and the high damage weight factor is taken as the first product. The loss function is obtained by averaging the first products of several relays.

[0022] By setting a weighting factor for remaining cycles and a weighting factor for high damage, the errors caused by samples nearing the end of life and relays about to fail are reduced, thereby improving the accuracy of the assessment.

[0023] Optionally, the remaining lifespan can be calculated by multiplying the predicted normalized remaining number of times by the rated total number of times of mechanical operation and then rounding down to obtain the actual remaining number of times of the relay.

[0024] Secondly, an electricity meter relay life assessment system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned electricity meter relay life assessment method is implemented.

[0025] This application has the following technical effects:

[0026] 1. Since different loads cause different damage to relays, for example, high loads cause greater damage to relays, an improved cubic spline smoothing fitting function is used to classify the load levels to ensure that the load levels are continuous and without discreteness, thereby improving the robustness of the fitting. After classifying the load levels, the life loss rate is calculated, making the calculation results more objective and accurate, and reducing errors.

[0027] 2. This application fits the current distortion curve using a double exponential function and combines it with the damage calculation logic of "peak current squared × distortion time" to accurately characterize the energy impact and time accumulation effect of current distortion on relay contacts, thereby improving the damage accuracy of a single relay closure. Furthermore, by setting a high damage weighting factor and a remaining number of cycles weighting factor in the loss function, the prediction error of relay samples nearing failure is reduced. Simultaneously, the prediction bias caused by the complex decay law of relay life (actual remaining cycles) under high load is reduced, ensuring that the life prediction model maintains high accuracy in high-load industrial scenarios. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of a method for evaluating the lifespan of an electricity meter relay according to an embodiment of this application.

[0029] Figure 2 This is a structural block diagram of a meter relay life assessment system according to an embodiment of this application. Detailed Implementation

[0030] This application discloses a method for evaluating the lifespan of an electricity meter relay, including:

[0031] S1: Obtain the current data of the relay under different historical loads, fit the current distortion curve based on the current data, extract the distortion parameters and calculate the current distortion damage degree; where the distortion parameters include: distortion time and current peak value.

[0032] In one embodiment, the method for fitting the current distortion curve is as follows: the current range of the meter is evenly divided into several small segments, and the maximum current value within the range is used as the metering point to obtain multiple load metering points. The timing data of the relay from open to closed under different loads is collected to obtain multiple current timing sequences. The steady-state current value of each current timing sequence is collected. The rise amplitude parameter, rise time constant, decay amplitude parameter, and decrease amplitude parameter are solved using the least squares method. Based on each parameter, the multiple timing data are fitted with a double exponential function, and the current distortion curve is obtained by combining the steady-state current value.

[0033] Specifically, multiple load (current) test groups are set up, and the current range is subdivided: according to the current range of the meter, it is evenly divided into several small segments, and the maximum current value in each segment is used as the measurement point to obtain... There are 1 current metering points, among which It can be dynamically adjusted according to the meter's accuracy; the higher the meter's measurement accuracy, the more accurate the meter. The larger. Divided by... Each current metering point serves as the load reference.

[0034] The current timing data of the relay during the process from opening to closing is collected by the current sensor built into the meter and recorded as a sequence. An example sampling frequency is set to 1kHz.

[0035] Because the current exhibits a "rapid surge → slow stabilization" characteristic when the relay is closed, a double exponential function is used to fit this curve (compared to a single exponential function, it can more accurately fit the different rates of change between the "rapid surge" and the "stabilization" segments). The fitting formula is as follows:

[0036] ;

[0037] in, To fit the current distortion curve Current value at any given time. The steady-state current value is the value when the current finally stabilizes, which can be obtained by collecting the average current value within 2 seconds after the relay is closed and the "waiting for stabilization time" has elapsed.

[0038] This is the amplitude parameter (in A) for the current surge segment, characterizing the magnitude of the current surge from its initial value to near its steady-state value. The larger the value, the more significant the current increase during the surge phase; This is the attenuation amplitude parameter (in A) during the current stabilization phase, characterizing the magnitude of current attenuation as it converges to its steady-state value after reaching its peak. The larger the value, the more pronounced the fluctuations during the stabilization phase.

[0039] The time constant (in seconds) for the sudden rise in current reflects the rate of current increase. The smaller the value, the faster the rapid increase. The steady-state time constant (in seconds) reflects the rate at which the current converges from its peak value to its steady state. The larger the value, the slower the stabilization process.

[0040] The stabilization time is the time interval obtained from the experiment when the relay is closed and the current tends to stabilize under various load currents.

[0041] All of them automatically solve the curve by curve fitting of the collected current time series data through the least squares method. The fitting objective is to minimize the mean square error between the calculated value and the actual collected value.

[0042] The least squares method is an existing technique and will not be elaborated upon here.

[0043] Beneficial effects: By using a double exponential function to fit the current curve, it can be adapted to the current change patterns in the two stages of "rapid rise" and "slow stabilization", ensuring the accuracy of distortion parameter extraction.

[0044] Obtain the distortion time and peak current from the distortion parameters.

[0045] In one embodiment, the time interval from the moment the relay closes to the moment when the fitted current value tends to stabilize is taken as the distortion time.

[0046] A longer distortion time indicates a slower process from the current surge to a stable state, resulting in longer arcing time at the relay contacts and greater damage. The distortion time is measured from the start of relay closure to the fitted current value. The time interval at which things tend to stabilize.

[0047] Distortion start time: The moment when the current value first exceeds the relay's operating threshold current (e.g., 0.1A, which can be adjusted according to the specific model). At this time, the relay contacts begin to make contact, and the current begins to rise sharply.

[0048] Distortion time endpoint: After the current drops from its peak, the condition that "the current fluctuation amplitude within 0.5s is ≤5% of the steady-state current value" is met for the first time. At the moment when the current stabilizes, the distortion effects such as arc burning basically end.

[0049] Peak current This is defined as the maximum current value during the distortion time, and it needs to be verified. To determine if a point is a noise point, follow these steps:

[0050] Multiple candidate current peak values ​​were obtained by repeatedly collecting current data under the target load and collecting data when the relay was closed. The coefficient of variation was then calculated. The calculation method is existing technology and will not be elaborated here. Then, a dynamic noise tolerance threshold is set. First, the initial noise tolerance threshold is set according to the specific implementation scenario. (For example Because the peak current amplitude range generated by the closed relay varies under different loads, and the peak current fluctuation is significantly higher under high loads, the noise tolerance threshold dynamically changes according to the load: the dynamic noise tolerance threshold is equal to the initial noise tolerance threshold multiplied by the hyperbolic tangent of the load (current). ).

[0051] like If it is less than the noise tolerance threshold, then If the current value is the true peak value, it is considered noise, and the maximum current value of the fitted curve is taken as the current peak value.

[0052] By repeatedly collecting current data from relay closure under the target load, noise points can be eliminated, effectively removing false peaks caused by sensor interference, ensuring the accuracy of current peak values, and effectively avoiding noise interference with subsequent damage calculations. Repeating the above steps yields the distortion time and current peak values ​​for all loads.

[0053] Thus, the core characteristic of current distortion when the relay is closed is obtained: distortion time. Peak current .

[0054] Current distortion damage This refers to the core indicator for quantifying the damage caused by a single closure of a relay.

[0055] The method for calculating the above-mentioned current distortion damage degree can be as follows:

[0056] ;

[0057] in, The degree of damage is determined by current distortion. This represents the peak current. For distortion time; This is the minimum peak current value among the actual peak values; The distortion time is the load current at which the minimum current peak occurs.

[0058] The total damage energy of a single closure was quantified. Since the damage during relay closure mainly comes from the contact arc energy and Joule heat, and both arc energy and Joule heat are proportional to the square of the current, the square of the current peak reflects the instantaneous energy impact intensity. Multiplying this by the distortion time gives a comprehensive characterization of the total damage amount during a single closure process, which is calculated as "energy impact × duration".

[0059] It is used to set the baseline damage amount. The damage multiplier obtained by dividing the total damage energy of a single closure by the baseline damage amount can intuitively represent the degree of damage.

[0060] Compared to using a single indicator or other calculation relationships, this calculation logic reflects the physical law that "energy is proportional to the square of the current" by using the square of the current, and combines the distortion time to reflect the cumulative effect of time, thus more accurately depicting the process of energy impact from current distortion to electric shock injury.

[0061] Thus, by fitting the current change curves generated when the relay is closed under different loads, and combining this with the noise point judgment method, the core feature related to current distortion, namely the distortion time, is obtained. Peak current Then, the damage caused by current distortion when the relay is closed is calculated by combining the two characteristics mentioned above.

[0062] S2: Sort the current distortion damage degree based on the magnitude of the load current to obtain the current distortion damage sequence; perform cross-validation on the current distortion damage sequence to calculate the initial smoothing parameter; divide the load current into multiple level intervals, adjust the initial smoothing parameter based on each level interval to obtain the adjustment parameter, fit the load current-distortion damage degree curve based on the adjustment parameter, analyze the boundary threshold of the load level according to the second derivative of the load current-distortion damage degree curve, select the point where the absolute value of the second derivative is greater than the boundary threshold as the boundary point of the load level, divide the load level, calculate the mean of all current distortion damage degrees in each load level, and obtain the load damage amount of the corresponding load level; obtain the relay life loss degree based on the load damage amount of all load levels.

[0063] In one embodiment, the method for obtaining the load current-distortion damage degree curve is as follows: The data points in the current distortion damage sequence are cross-validated to calculate initial smoothing parameters. The load current is divided into multiple level intervals. Any level interval is selected as the target level interval. The ratio of the variance of the current distortion damage degree in the target level interval to the variance of all current distortion damage degrees is used as the first ratio. The product of the first ratio and the initial smoothing parameters is calculated to obtain the smoothing parameters for the target level interval. Similarly, the smoothing parameters for all level intervals are obtained. Based on the multiple smoothing parameters, a cubic spline smoothing curve function is fitted to the current distortion damage sequence to obtain the load current-distortion damage degree curve.

[0064] Specifically, the damage caused to the relay by the current distortion when the relay is closed under different loads is used as the data benchmark for classifying load levels (low, medium and high). When classifying load levels for multiple ordered load currents, it is necessary to ensure the continuity of the load level classification and avoid discrete situations (for example, 2A is classified as a high load level while 3A is classified as a low load level).

[0065] Therefore, based on the idea of ​​ordered clustering, different loads (currents) and their corresponding current distortion damage are fitted, and the load levels are classified in combination with the inflection point properties.

[0066] Arrange all data points in ascending order of load size (current): ,in And a cubic spline smoothing curve function was applied to the current distortion damage sequence. The cubic spline smoothing is adapted and improved as follows:

[0067] Adaptive smoothing parameter determination: The initial smoothing parameters are calculated using GCV (Generalized Cross Validation). Then, the current range is divided into three equal-level intervals, and the smoothing parameters for each level interval are adjusted to obtain the adjusted parameters. The adjusted parameters are equal to the initial smoothing parameters multiplied by the ratio of the variance of the current distortion damage degree in each level interval to the variance of the global current distortion damage degree.

[0068] Cubic spline smoothing is commonly used for data fitting to achieve curve smoothing and noise suppression. However, in relay damage-current relationships, the smoothing parameter is usually fixed or manually adjusted, lacking adaptability and unable to handle differences in noise levels across different current ranges. It fails to address insufficient noise suppression in high-current regions while also failing to avoid over-smoothing in low-current regions. Therefore, combining cross-validation and dynamic adjustment of interval characteristics is crucial. Traditional cross-validation provides global smoothing, while interval adjustment addresses noise inhomogeneity (e.g., higher noise levels in high-current regions require larger smoothing parameters, and vice versa in low-current regions), effectively improving the robustness and accuracy of the fitting.

[0069] Then, the load level threshold was analyzed using the load current-distortion damage curve to classify the load levels.

[0070] Specifically, calculate the second derivative of the load current-distortion damage curve points, calculate the absolute median difference of the absolute values ​​of the second derivatives of all fitted functions, and add twice the absolute median difference to the median of the absolute values ​​of the second derivatives of all data to obtain the load level threshold. :

[0071] Threshold is calculated here. Using the median, compared to the mean, avoids interference from outliers and ensures that the threshold reflects the typical rate of change of the curve. At the same time, the median absolute difference is more robust to outliers than the standard deviation, and twice the median absolute difference corresponds to a higher confidence level, ensuring that only significant changes are identified as inflection points.

[0072] All that satisfy the condition that the second derivative is greater than The point is used as the dividing point for load levels, thus obtaining... Each load level boundary point. At this point, the current range can be divided into... Each of the following consecutive intervals corresponds to... Types of load levels.

[0073] get After dividing the load level into intervals, the average value of the current distortion damage degree corresponding to all loads within each load level is summed to obtain the load damage amount corresponding to the load level. .

[0074] Thus, by using the improved clustering algorithm with the data itself as the driving force and current distortion damage degree as the clustering core, the load level can be classified more objectively and accurately.

[0075] After classifying the load levels, the lifespan loss component is calculated based on each load level, and then the lifespan loss of the relay is calculated.

[0076] Lifespan loss This represents the cumulative equivalent total lifespan loss across various load levels, used to characterize relay lifespan loss. It is calculated and characterized using the relay closing duration at each load level and the corresponding load damage at each load level.

[0077] Lifespan loss It equals the sum of the lifetime loss components under each load level, where the lifetime loss component equals the sum of the relay closing duration under each load level multiplied by the load damage amount under the corresponding load level.

[0078] S3: Construct feature vectors from load level, load damage amount, and life loss degree, obtain historical feature vectors of several relays to train a preset LSTM model, optimize model parameters through a preset loss function, and obtain a life prediction model.

[0079] Specifically, a loss function is constructed, recording data in units of one closure. Any relay is selected as the target relay, and the cumulative number of closures under different loads is collected. The load damage and lifespan loss corresponding to different load levels are calculated. The cumulative number of closures under different loads, the load damage, and the lifespan loss corresponding to different load levels are used as historical feature vectors. The remaining number of closures is obtained by subtracting the current number of closures from the total rated mechanical count. The ratio of the remaining number of closures to the total rated mechanical count is used as the actual normalized remaining number of closures. A high-damage weighting factor and a remaining number of closures weighting factor are calculated. The square of the difference between the predicted normalized remaining number of closures and the actual normalized remaining number of closures is used as the first difference. The product of the first difference, the remaining number of closures weighting factor, and the high-damage weighting factor is used as the first product. The average of the first products of several relays is then used to obtain the loss function.

[0080] First, prepare training samples by selecting historical data of the entire lifecycle of several (e.g., 1000) relays of the same model (covering the complete process from brand new to the end of their lifespan). Record the following core parameters for each relay:

[0081] Timestamp: Recorded per closure;

[0082] Load level data: Cumulative number of closures under each load ;

[0083] Load impairment data: Load impairment at each load level ;

[0084] Lifetime loss data: Lifetime loss rate ;

[0085] Remaining lifespan data: The remaining number of cycles is obtained by subtracting the current number of cycles from the total number of cycles closed for each relay. .

[0086] For the historical data of each relay, multiple samples are divided according to "time nodes":

[0087] Time node: Starting from the time the relay is put into use, the time is taken up to the time when the rated number of closures is accumulated (e.g., 1000 times, which can be adjusted according to the total lifespan). That is, the time from the time the relay is put into use to the time of the cut-off is one time node.

[0088] Input normalized features ( ): No. Historical feature vectors at each node.

[0089] ;

[0090] in, They are respectively The number of times the relay closes under a certain load level; They are respectively The load damage caused by relay closure at various load levels; Indicates the degree of lifespan loss; subscript Indicates the first The normalized value of the cumulative values ​​at each time point (for easy input to the prediction model), such as For the first The normalized value of the total number of closures under the first load level within a given time point; The normalization can be obtained by dividing the original number of times by the rated total number of mechanical times (factory parameter) when the relay is not connected to a load; All can be normalized by dividing by the maximum current distortion damage degree calculated from the experimental data in step S1; The normalization can be obtained by dividing the lifetime damage degree by the maximum lifetime damage degree in the historical data sample.

[0091] No. The actual normalized remainder count of the nth node, i.e., the nth At each node, the remaining number of cycles is obtained by subtracting the current number of cycles from the rated total number of cycles. This is then divided by the rated total number of cycles (factory parameter) under no-load conditions to obtain the actual normalized remaining number of cycles. .

[0092] The loss function can be designed as follows:

[0093] ;

[0094] in, The loss function; The number of samples; For the first One sample; For the first The weighting factor for the remaining frequency of each sample; For the first High damage weighting factor for each sample For the first The number of prediction normalization remainder times for each sample. These are the actual normalized residual parameters.

[0095] The value varies with the sample The remaining number of times decreases and increases. Therefore, the first time is used. The remaining lifetime (remaining number of times) of the sample divided by the number of times. The percentage of remaining cycles is obtained by calculating the total number of closures (total rated mechanical cycles) throughout the entire lifecycle of each sample; the exponential function of the negative of the percentage of remaining cycles is then used to obtain the result. This causes the remaining count weight factor to increase in the range [0, 1] as the remaining count of the sample decreases.

[0096] The calculation logic is as follows: The value varies with the sample The proportion of damage from medium to high loads increases. Therefore, samples are extracted first. The cumulative damage for each load level is calculated by multiplying the number of closures at a load level by the corresponding load damage. Then, the cumulative load damage for the highest load level is divided by the sum of the cumulative load damage for all load levels to obtain the high-load damage percentage. Finally, the hyperbolic tangent function of the high-load damage percentage is calculated, thus yielding the... This causes the high-damage weighting factor to increase within the range of [0, 1] as the proportion of high-damage samples increases. Its purpose is to intensify the penalty for "high-damage samples".

[0097] pass and The combined effect of these factors increases the penalty for samples whose lifespan (actual remaining cycles) is closer to the end (the situation is more sensitive and urgent) and whose load damage is greater (the relay quality is more damaged).

[0098] By setting weight factors for remaining cycles and high damage in the loss function, the model not only reduces prediction errors for relay samples nearing the end of their lifespan, but also enhances the accuracy of lifespan prediction for relays about to fail compared to traditional loss functions, reducing the risk of missed maintenance reports. Furthermore, the model specifically optimizes the fitting effect for high-load samples, effectively reducing prediction bias caused by the complex lifespan decay patterns under high loads, ensuring the lifespan prediction model maintains high accuracy in high-load industrial scenarios.

[0099] This completes the construction and training of the relay life prediction model.

[0100] S4: Obtain the real-time feature vector of the relay and input it into the life prediction model to predict the normalized remaining number of times. Based on the predicted normalized remaining number of times, obtain the remaining life and complete the evaluation of the real-time remaining life of the relay.

[0101] In one embodiment, relay data is acquired in real time, and the lifespan estimation frequency can be set or the estimation can be manually triggered according to actual conditions. Input data is then acquired and calculated. Input the dataset into the lifespan prediction model and output the predicted normalized remaining number of times. .

[0102] The actual remaining life of the relay is obtained by rounding down the product of the normalized remaining number of cycles and the rated total number of cycles.

[0103] A lifetime prediction model that integrates the remaining number of attempts weight factor and the high damage weight factor is constructed to achieve a lifetime prediction function that is more in line with the scenario.

[0104] This application also discloses a life assessment system for electricity meter relays, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a life assessment method for electricity meter relays according to this application.

[0105] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0106] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method of evaluating the life of a metering relay, characterized in that, The method comprises the following steps: Obtain current data of the relay under different historical loads, fit a current distortion curve according to the current data, extract distortion parameters and calculate current distortion damage degree; wherein the distortion parameters include distortion time and current peak value; Sort the current distortion damage degree based on the size of the load current to obtain a current distortion damage sequence; cross-verify the current distortion damage sequence to calculate an initial smoothing parameter; divide the load current into multiple grade intervals, adjust the initial smoothing parameter based on each grade interval to obtain an adjusted parameter, fit a load current-distortion damage degree curve based on the adjusted parameter, analyze the boundary point threshold of the load grade based on the second derivative of the load current-distortion damage degree curve, select a point with an absolute value of the second derivative greater than the boundary point threshold as the boundary point of the load grade, and divide the load grade, calculate the average of all current distortion damage degrees in each load grade to obtain the load damage amount corresponding to the load grade; obtain the life consumption degree of the relay according to the load damage amount of all load grades; Construct a feature vector with the load grade, load damage amount, and life consumption degree, obtain historical feature vectors of several relays to train a preset LSTM model, optimize the model parameters through a preset loss function, and obtain a life estimation model; Obtain a real-time feature vector of the relay, input it into the life estimation model, output a predicted normalized remaining number, obtain the remaining life based on the predicted normalized remaining number, and complete the evaluation of the real-time remaining life of the relay.

2. The method of claim 1, wherein, The calculation method of the current distortion curve is as follows: uniformly divide the current range of the electric meter into several small segments, take the maximum current value in the range as the measurement point, obtain multiple load measurement points, collect time sequence data of the relay from disconnection to closure under different loads, obtain multiple current time sequences, collect steady-state current values of each current time sequence, solve the sudden rise amplitude parameter and the sudden rise time constant, the decay amplitude parameter and the sudden decay amplitude parameter by the least square method, fit a double exponential function based on each parameter to multiple time sequence data, and obtain the current distortion curve combined with the steady-state current value.

3. The method of claim 1, wherein, The calculation method of the distortion parameter is as follows: Obtain the fitted current value according to the current distortion curve, and take the time interval from the relay closing time to the time when the fitted current value tends to be stable as the distortion time; Select any load as a target load, repeatedly collect current data of the relay under the target load multiple times to obtain multiple candidate current peak values, calculate the ratio of the standard deviation to the mean of the multiple candidate current peak values to obtain the coefficient of variation; Set an initial noise tolerance threshold, and obtain a target load dynamic noise tolerance threshold according to the product of the initial noise tolerance threshold and the hyperbolic tangent function of the target load; Select the candidate current peak value with a coefficient of variation less than the dynamic noise tolerance threshold as the true peak value, select the maximum value in the true peak value as the current peak value of the target load, and similarly obtain the current peak value of all loads.

4. The method of claim 3, wherein, The calculation method of the current distortion damage degree is as follows: take the product of the square of the current peak value of the target load and the corresponding distortion time as the total damage amount, and take the product of the square of the minimum current peak value in the true peak value of the target load and the minimum distortion time as the reference damage amount. The ratio of the total damage amount to the reference damage amount is calculated to obtain the current distortion damage degree of the target load; Similarly, the current distortion damage degrees of all loads are obtained.

5. The method of claim 1, wherein, The calculation method of the load current-distortion damage degree curve is: the data points in the current distortion damage sequence are used to calculate the initial smoothing parameter by cross-validation, the load current is divided into multiple level intervals, any level interval is selected as a target level interval, the ratio of the current distortion damage degree variance of the target level interval to the variance of all current distortion damage degrees is taken as a first ratio, the product of the first ratio and the initial smoothing parameter is calculated to obtain the smoothing parameter of the target level interval, and the smoothing parameters of all level intervals are obtained in the same way, and the current distortion damage sequence is fitted by a cubic spline smoothing curve function based on the multiple smoothing parameters to obtain the load current-distortion damage degree curve.

6. The method of claim 1, wherein, The calculation method of the demarcation point threshold of the load level is: the absolute median difference of the absolute value of the second derivative of all data in the load current-distortion damage degree curve is calculated, and the sum of twice the absolute median difference and the median of the absolute value of the second derivative of all data is taken as the demarcation point threshold of the load level.

7. The method of claim 1, wherein, The calculation method of the life consumption degree of the relay is: according to the different load currents, the data points in the current distortion damage sequence are divided into corresponding load levels, any load level is selected as a target load level, the current distortion damage degrees of all loads in the target load level are summed and averaged to obtain the load damage amount of the target load level; The product of the load damage amount of the target load level and the sum of the distortion time of the target load level is taken as the life consumption degree component of the target load level, and the sum of the life consumption degree components of all load levels is taken as the life consumption degree of the relay.

8. The method of claim 1, wherein, The calculation method of the loss function is: In the unit of closed times, any relay is selected as a target relay, the different cumulative closing times of different loads of the target relay are collected, the load damage amount and the life consumption degree corresponding to different load levels are calculated, and the different cumulative closing times of different loads of the target relay, the load damage amount and the life consumption degree corresponding to different load levels are taken as the historical feature vector; The residual number is obtained by subtracting the current closing time from the rated total mechanical number, and the ratio of the residual number to the rated total mechanical number is taken as the actual normalized residual number; the high damage weight factor and the residual number weight factor are calculated; The square of the difference between the predicted normalized residual number and the actual normalized residual number is taken as a first difference, the product of the first difference, the residual number weight factor and the high damage weight factor is taken as a first product, the first products of several relays are calculated, and the average value is taken to obtain the loss function.

9. The method of claim 1, wherein, The calculation method of the residual life is: the product of the predicted normalized residual number and the rated total mechanical number is taken as an integer to obtain the actual residual number of the relay.

10. An electric meter relay life assessment system characterized by, The method comprises: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a meter relay life evaluation method according to any one of claims 1-9 is realized.

Images