Cross-device clock correction method supporting accurate synchronization of meters

Abstract

The invention discloses a cross-device clock correction method supporting accurate synchronization of meters, and aims to improve the data time consistency among multiple meter devices and meet the requirements of high-precision electric power measurement and event analysis. The method comprises the following steps: acquiring a network time protocol timestamp and a first sampling time sequence in a reference meter, generating a clock reference signal, and broadcasting the clock reference signal to a corrected meter; comparing the corrected meter with a local second sampling time sequence, and calculating a clock skew coarse value vector; further performing wavelet decomposition on the offset coarse value, extracting to form a drift feature vector, fusing the drift feature vector with a historical template, and calculating a unified correction parameter set by using a Bayesian recursion method; and generating a correction pulse sequence carrying a delay residual compensation code based on the parameter set, sending the correction pulse sequence to a corrected meter, calling an interpolation compensation function to construct a piecewise polynomial drift compensation table, and finally realizing dynamic correction and cross-device accurate synchronization of a local real-time clock.

Description

Technical Field

[0001] The present invention relates to the technical field of electric meters, and in particular to a cross-device clock correction method supporting precise synchronization of meters. Background Art

[0002] In existing technologies, electricity meters, especially various types of single-phase and three-phase meters, generally rely on local crystal oscillator clocks or low-precision network time synchronization to update device time during the collection and upload of power parameters. Some systems use the Network Time Protocol (NTP) to periodically synchronize the master control node, but most lower-level meters still operate with relatively independent local clocks and lack a unified time coordination mechanism across devices. In power distribution environments with low-cost deployment and limited communication resources, traditional methods often only achieve coarse-grained time synchronization and cannot support high-precision analysis of load behavior, event chains, or timing anomalies.

[0003] However, existing technologies have obvious shortcomings in the following aspects: First, the NTP protocol is easily affected by network delay fluctuations, resulting in uncertainty in the reference time and cannot be used for fine time correction at the meter level; second, traditional time correction methods usually do not take into account the long-term frequency drift and instantaneous fluctuation characteristics of the meter crystal oscillator, and lack the ability to model dynamic drift; third, the current method fails to achieve high-frequency comparison and drift compensation linkage between sampling time series, and clock correction across devices cannot achieve closed-loop feedback and continuous optimization, which seriously affects the timing integrity and precise synchronization capabilities of the distribution network side.

[0004] In view of the problems existing in the above-mentioned existing technologies, it is urgent to propose a new method to improve the cross-device clock consistency and dynamic correction accuracy between meters. Summary of the Invention

[0005] The present application provides a cross-device clock correction method that supports precise synchronization of meters, so as to improve the cross-device clock consistency and dynamic correction accuracy between meters.

[0006] This application provides a cross-device clock calibration method that supports precise meter synchronization, including: Acquire a network time protocol timestamp and a first sampling time sequence in a reference meter, and generate a clock reference signal by combining the network time protocol timestamp and the first sampling time sequence; broadcasting the clock reference signal to the meter to be calibrated via a communication link, recording a second sampling time series in the meter to be calibrated, and calculating a clock offset coarse value vector based on the clock reference signal and the second sampling time series; Performing multi-scale wavelet decomposition based on the clock offset coarse value vector to extract drift characteristic components, wherein the drift characteristic components include an instantaneous drift component, a trend drift component, and a random jitter component, and combining the drift characteristic components to form a drift characteristic vector; The drift feature vector is fused with the drift history template window by window, and a unified correction parameter set is calculated using a Bayesian recursive method, wherein the unified correction parameter set includes a drift amplitude coefficient and a phase compensation coefficient; generating a correction pulse sequence based on the unified correction parameter set, the correction pulse sequence carrying a delay residual compensation code, and sending the correction pulse sequence to the corrected meter via a meter communication bus; The correction pulse sequence is received in the corrected meter, and an interpolation compensation function is called based on the delay residual compensation code to generate a drift compensation table, where the drift compensation table has a piecewise polynomial structure. The local real-time clock is dynamically corrected according to the drift compensation table to achieve cross-device clock correction.

[0007] The beneficial effects of the technical solution provided by this application include: (1) By combining the network time protocol timestamp with the sampling time series to generate a highly stable clock reference signal, and uniformly synchronizing multiple meters, the time drift and sampling error between multiple devices are effectively reduced, and the accuracy of event alignment and data comparison is significantly improved. (2) By extracting the instantaneous drift, trend drift and random jitter characteristics through multi-scale wavelet decomposition, and constructing the drift feature vector, the correction process can perceive the error sources at different levels, and enhance the model's adaptability to nonlinear drift phenomena. (3) The Bayesian recursive method is used to fuse historical template information and dynamically generate a unified correction parameter set, supporting personalized corrections for different operating environments and equipment aging conditions, ensuring continuous optimization of time calibration accuracy during long-term operation. (4) By encapsulating the correction pulse sequence into a data packet carrying a compensation code and using the existing communication bus for transmission, the introduction of high-frequency communication or additional hardware synchronization lines is avoided, making it suitable for the deployment of large-scale low-power meter systems in edge distribution scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 This is a flowchart of a cross-device clock correction method that supports precise synchronization of meters, provided in the first embodiment of the present application. DETAILED DESCRIPTION

[0009] The following description sets forth many specific details to facilitate a thorough understanding of the present application. However, the present application can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of the present application. Therefore, the present application is not limited to the specific implementations disclosed below.

[0010] The first embodiment of the present application provides a cross-device clock calibration method that supports precise synchronization of meters. Figure 1 , which is a schematic diagram of the first embodiment of this application. Figure 1 The first embodiment of the present application provides a cross-device clock correction method that supports precise synchronization of meters, which is described in detail.

[0011] Step S101: obtaining a Network Time Protocol timestamp and a first sampling time sequence in a reference meter, and generating a clock reference signal by combining the Network Time Protocol timestamp and the first sampling time sequence.

[0012] The process of obtaining the Network Time Protocol timestamp and the first sampling time series in the reference meter, as the starting point of the entire cross-device clock correction method, must establish a unified time base with extremely high accuracy and stability, thereby providing a reliable basis for the subsequent synchronization control between the meters. In specific implementation, it is first necessary to ensure that the reference meter is in a stable network state, ensuring that it can periodically interact with a public or private network time server via the UDP protocol and successfully parse standard NTP data packets. Each interaction includes at least four pieces of information: request timestamp, response timestamp, received timestamp, and sent timestamp. The system calculates the current Network Time Protocol timestamp using the standard NTP algorithm, that is, a Coordinated Universal Time (UTC) time value with a relative error of no more than 10 milliseconds.

[0013] At the same time, a high-precision sampling device operating within the reference meter samples the local clock in real time at a set fixed sampling period (e.g., 10ms or 20ms), recording data to form a first sampling time series. This sampling time series is a set of ordered time points, each of which precisely marks the local crystal oscillator counter value at that moment or a microsecond-level time value derived from it. During the generation of this first sampling time series, the timing mechanism within the reference meter must maintain a certain level of stability, for example by using a temperature-compensated crystal oscillator (TCXO) or digital phase-locked loop (DPLL) to reduce jitter and timebase drift.

[0014] Next, to generate a globally consistent clock reference signal, the acquired Network Time Protocol timestamps must be synchronized with the first sampling time series. Specifically, the most recent NTP synchronization point can be selected as the anchor point, and the NTP timestamps can be mapped to the local sampling time axis, thereby establishing a set of "sampling point-network time" correspondences. Through least squares fitting, Kalman filtering, or interpolation algorithms, a time conversion function with global mapping capabilities can be constructed on the sampling time series. This mathematical expression maps the local sampling time to a globally unified reference time.

[0015] This time conversion function, the core component of the clock reference signal, combines the original NTP timestamp, the aligned sampling time reference, the estimated network delay, and the time uncertainty measurement information into a structured data object, forming the final clock reference signal. In actual transmission, this signal can be packaged using a binary compression structure, such as Protobuf or CBOR. The combination of the reference time point, mapping slope, offset parameters, and reference sampling point data is then sent over a standard communication channel. In subsequent steps, this clock reference signal will be broadcast to multiple calibrated meters, serving as an absolute reference source for synchronization.

[0016] Therefore, by accurately extracting NTP timestamps and combining them with high-frequency sampling data to generate a stable and reliable clock reference signal, we not only lay the foundation for clock consistency across devices, but also provide key input for subsequent drift modeling and dynamic correction.

[0017] Furthermore, the obtaining of a network time protocol timestamp and a first sampling time sequence in a reference meter, and combining the network time protocol timestamp and the first sampling time sequence to generate a clock reference signal includes: In the reference meter, a first network timestamp and a second network timestamp are obtained from a public network NTP server and a GNSS receiver respectively, and the local time of reception and the corresponding time value are recorded; Compare and analyze the time difference between the first network timestamp and the second network timestamp with the historical offset change trend, assign weights based on their stability, jitter range, and delay symmetry, perform a weighted average operation, and generate a fused timestamp as the reference time value for the current period; Collecting multiple clock count values ​​of the local crystal oscillator within a preset sampling period to form a first sampling time series, and linearly corresponding each sampling point in the series to the fused timestamp to generate a set of calibration sampling time series aligned with a unified reference time; The calibration sampling time sequence and the fusion timestamp of the current cycle are encapsulated together to form a clock reference signal including a time reference type identification field, a synchronization error assessment field and a sampling frequency stability flag, which is subsequently broadcast to the calibrated meter.

[0018] In this invention, to achieve high-precision clock synchronization between meters, a clock reference signal must first be generated in the reference meter. This reference signal must not only provide a highly consistent mapping to a unified time source but also be adaptable to timing uncertainties introduced by complex factors such as network fluctuations, signal jitter, and multipath errors in public networks. To this end, this embodiment proposes a reference time generation method that integrates multi-source time information and incorporates local sampling characteristics. This design balances time accuracy, jitter robustness, and synchronization availability for upstream and downstream devices.

[0019] Specifically, the reference meter obtains two sets of timestamp data via parallel channels, one from a public NTP (Network Time Protocol) server and the other from a GNSS (Global Navigation Satellite System) receiver. The first network timestamp originates from a UDP request-response session with the public NTP server and contains the server's current standard time and the round-trip delay of the request. The second network timestamp, based on the UTC (Coordinated Universal Time) solution received by the GNSS receiver, is highly accurate but can be affected by signal obstruction. To leverage the advantages of these two time sources, the system records the values ​​of the first and second network timestamps, along with the local times at which they were received, during each clock generation cycle and calculates the absolute time difference between them.

[0020] Based on this, the system assigns weighting coefficients to the two types of timestamps, taking into account the time difference fluctuation range (e.g., standard deviation or maximum-minimum difference) in historical statistical information, the signal stability of each type of timestamp (e.g., the number of consecutive timestamps available), and the symmetry of request-response delays (i.e., the consistency of NTP round-trip times). These weighting coefficients are not fixed but dynamically generated to reflect the relative credibility of each time source in the current period. For example, when GNSS signal reception quality degrades or obstruction occurs, the corresponding weighting coefficient is automatically reduced, increasing the influence of NTP time on the fusion result. The system then fuses the two timestamps using a weighted average to generate a fused timestamp, which is used as the reference time value for the current period.

[0021] Afterward, the reference meter calls on the local crystal oscillator signal and continuously reads multiple clock counts within a complete preset sampling period (for example, 200ms, 500ms, or 1s). Each count represents the number of beats counted by the crystal oscillator at that instant, thus forming a first sampling time series consisting of multiple time points. Due to slight frequency or temperature drift of the crystal oscillator, this sampling series, while time-continuous, cannot be directly used for precise synchronization and requires mapping and calibration with a unified reference time. To this end, the system linearly maps each sampling point in the sampling time series onto the time axis constructed by the fused timestamps based on its relative position in the current sampling period, thereby forming a set of calibrated sampling time series that are fully aligned with the reference time. This mapping process takes into account the relationship between the actual crystal oscillator frequency and the spacing between sampling points, ensuring that the mapped time intervals are equivalent to the reference time distribution.

[0022] After completing the sampling calibration, the system encapsulates the fusion timestamp generated in this cycle and the calibration sampling time sequence to form an information structure containing multiple fields. This structure is defined as a clock reference signal, in which the specific fields include: a time reference type identification field, which is used to indicate the time source composition based on the current fusion timestamp, such as "GNSS+NTP"; a synchronization error assessment field, which records the time difference between the two original timestamps during the fusion process and its weighted standard deviation; a sampling frequency stability flag, which characterizes the stability index of the local crystal oscillator frequency, such as the frequency fluctuation rate or temperature drift coefficient. By introducing the above-mentioned structured fields, not only is the downstream device's ability to judge the reliability of the source data when receiving the clock reference signal enhanced, but it also provides a verifiable reference basis for subsequent delay compensation and phase correction.

[0023] Ultimately, this clock reference signal will serve as a carrier of a unified time base and be broadcast to multiple calibrated meters through a point-to-multipoint communication mechanism, serving as the original basis for all subsequent clock offset detection and compensation steps, laying the technical foundation for the precise alignment of clocks across devices in the present invention.

[0024] Step S102: broadcasting the clock reference signal to the meter to be calibrated via a communication link, recording a second sampling time series in the meter to be calibrated, and calculating a clock offset coarse value vector based on the clock reference signal and the second sampling time series.

[0025] After establishing the clock reference signal in the reference meter, this signal needs to be reliably transmitted to multiple calibrated meters, enabling them to accurately identify and process the reference signal in their local environment, thereby providing a time alignment basis for estimating clock differences across devices. To achieve this, the clock reference signal must be broadcast from the reference meter to the target meters using an existing communication link. This link can be an RS485 bus, power line carrier, LoRa, or other industrial wireless protocol, or a forwarding channel from a concentrator in a configuration master system. The broadcast method should support low-latency and high-reliability transmission, ensuring that the receiving end completes parsing and storage of the reference signal within a reasonable time window. To improve clock comparison accuracy, it is recommended to add a reference timestamp number field and a hash check field to the broadcast data packet, allowing the calibrated meters to identify synchronization timing and data integrity.

[0026] Upon receiving the clock reference signal, the meter being calibrated must immediately enter data comparison mode and activate its local sampling module to continuously sample the current system time, forming a second sampling time series. This second sampling time series is structurally consistent with the first sampling time series, also consisting of a continuous sequence of high-precision time points. Its sampling period should be as consistent as possible with that of the reference meter, for example, a 10ms sampling interval. To ensure alignment of the sampling start points, the mapping relationship between the second sampling time series and the first sampling time series should be initialized using the reference anchor time embedded in the received clock reference signal.

[0027] After acquiring the sampled data, the time offset estimation phase begins. The system compares the first sampling time series from the reference meter with the second, locally generated sampling time series. The system calculates the time difference for each sampling point, generating a vector of multiple time differences known as the coarse clock offset vector. Each element of this coarse value vector represents the absolute error between the reference time and the local time at a specific sampling instant. In actual calculations, if there are slight sampling time misalignments, window alignment techniques can be used to weightedly match several preceding and succeeding sampling points, thereby avoiding point-to-point error amplification.

[0028] Furthermore, when constructing the coarse clock offset vector, the impact of communication delays and interruptions on the time comparison results should also be considered. To this end, it is recommended to introduce a threshold filtering mechanism during the offset vector construction process to remove outlier time differences caused by link jitter or packet delays, retaining offset results within a stable range, and thus improving the representativeness and usability of the offset estimate.

[0029] Furthermore, broadcasting the clock reference signal to the calibrated meter via a communication link, recording a second sampling time series in the calibrated meter, and calculating a coarse clock offset vector based on the clock reference signal and the second sampling time series includes: Splitting the clock reference signal into two parts including a first time anchor point field and a first sampling alignment parameter field, encapsulating each part into two communication data frames, and broadcasting the two parts to each calibrated meter via the communication link, so that the calibrated meter can parse and restore the complete clock reference signal according to a preset logic; After the calibrated meter receives the first time anchor point field, it starts a real-time collection task for a second sampling time series, records multiple system current times locally at fixed time intervals, and timestamps each recorded point to correspond to the received first time anchor point field, generating a second sampling time series with an anchor point index identifier; Performing a one-to-one matching of the first sampling alignment parameter field in the clock reference signal with the anchor point index in the second sampling time series, and extracting a set of successfully matched sampling pairs using a sequential window comparison mechanism for subsequent offset calculation; For each pair of time points in the set of successfully matched sampling pairs, the time difference between the sampling point in the clock reference signal and the corresponding sampling point in the second sampling time series is calculated to obtain multiple offset samples, and the samples are arranged in chronological order to form a coarse clock offset value vector. The coarse clock offset value vector serves as input for extracting the drift feature component in subsequent steps.

[0030] In the implementation of the present invention, in order to achieve clock alignment between the reference meter and each calibrated meter, it is necessary to first ensure that the clock reference signal can be accurately broadcast to all calibrated meters, and enable each meter to independently calculate its own clock offset relative to the reference meter based on this signal.

[0031] First, after the reference meter generates a clock reference signal, it is logically split into two parts: the first is a time anchor field, used to calibrate key moments on the unified reference timeline; the second is a sampling alignment parameter field, primarily used to record the temporal distribution characteristics and timing structure encoding of the sampling sequence within the reference meter. These two parts are encapsulated into independent data frames. Each data frame uses a predefined communication protocol format and includes a synchronization header, frame identifier, payload length, and a checksum field to ensure data consistency and error detection during communication. These two data frames are then transmitted sequentially to all calibrated meters via a multicast, broadcast, or master-slave bus transmission mechanism through the communication module in the reference meter.

[0032] Once the calibrated meter receives a data frame containing the first time anchor point field, it immediately initiates the local time acquisition task. This acquisition task, based on the internal crystal oscillator signal or system clock, continuously records multiple time points at fixed intervals (e.g., 1ms or 5ms) to form a preliminary second sampling time series. Each recorded time point is timestamped with the received first time anchor point field. This means that the system records each sampling point and associates it with the most recently received time anchor point, thus providing the local sampling series with a unified reference timing identifier. This approach effectively addresses timing drift issues caused by network latency, transmission jitter, or inconsistent sampling starts.

[0033] At the same time, in the meter being calibrated, upon receiving the second data frame, that is, the data frame containing the first sampling alignment parameter field, the system matches the content of this field with each anchor point index in the aforementioned second sampling time series. This matching process uses a sequential window comparison mechanism, that is, searching for segments that are consistent with the structure of the reference sampling sequence within each sampling time period to ensure that the matching results are both accurate and have temporal continuity. Each successful match of a set of sampling points constitutes a sampling pair, which contains one reference sampling time point and one calibrated sampling time point. Ultimately, the system obtains a set of sampling pairs, and this set has eliminated discrete points or mismatched points caused by signal jitter, and has high consistency and representativeness.

[0034] For each successfully matched sampling pair, the system calculates the absolute time difference between the two time points. This time difference is the instantaneous offset between the reference time and the time being corrected. By arranging all the sampling pair differences in chronological order, a complete offset sequence is formed. This sequence not only reflects the overall deviation trend of the corrected meter but also reveals possible characteristics such as periodic jitter and instantaneous fluctuations. Therefore, this offset sequence is defined as a coarse clock offset vector, which serves as an important input variable in subsequent steps for wavelet decomposition, drift feature extraction, and correction parameter estimation.

[0035] Step S103: performing multi-scale wavelet decomposition based on the clock offset coarse value vector to extract drift characteristic components, wherein the drift characteristic components include instantaneous drift components, trend drift components and random jitter components, and combining the drift characteristic components to form a drift characteristic vector.

[0036] After obtaining the coarse clock offset vector, to further model inter-device clock drift characteristics, multi-scale wavelet decomposition is required to effectively identify and extract error components at different time scales that affect clock stability. This process is not only an extension of gross error processing but also a fundamental step in refined drift modeling. Its output directly determines the effectiveness and adaptability of subsequent correction parameters.

[0037] First, a wavelet basis function suitable for processing nonstationary time-domain signals should be selected, such as orthogonal wavelet functions such as Daubechies (db4 or db6), Coiflet, or Symlet, to ensure that both the time and frequency domain information of the signal are preserved during the decomposition process. The goal of wavelet decomposition is to decompose the coarse clock offset vector into a sequence of coefficients at multiple scale levels, with each scale level representing the varying components of a specific frequency band. In practice, it is recommended to use a discrete wavelet transform (DWT) to perform a three- or four-level decomposition of the coarse value vector, sequentially extracting high-frequency detail coefficients and low-frequency approximation coefficients.

[0038] By analyzing coefficients at different scales, representative drift characteristic components can be isolated. The coefficient sequences extracted from the high-frequency layer typically represent instantaneous drift components in the device clock. This type of drift is often caused by sudden temperature changes, power supply fluctuations, or external interference, manifesting as sudden changes or localized jitter in the signal. The coefficients in the mid-frequency layer can correspond to trend drift components, generally reflecting linear or nonlinear drift trends caused by crystal oscillator aging or accumulated system operation time. The components contained in the low-frequency layer are more likely to be related to random jitter components. Although their fluctuation range is small, they pose a continuous threat to system stability over long periods of operation.

[0039] After extracting the various drift components, they should be reorganized, encapsulating instantaneous drift, trend drift, and random jitter in vector form into a uniformly structured data entity, known as the drift feature vector. This vector can be defined as a multi-channel time series matrix, with each column representing a drift type and each row corresponding to the drift state at a sampling moment. To facilitate subsequent algorithm processing, it is recommended to normalize the vector to unify the dimensions and constrain the value range, for example, through Z-score normalization or Min-Max scaling, to make the different types of drift comparable.

[0040] Furthermore, to ensure the representativeness of the drift feature vector, outliers should be identified and filtered from the coarse clock offset vector before decomposition. Robust statistical methods based on the median absolute deviation (MAD) can be used to eliminate extreme offset points caused by communication desynchronization or single-point errors. For time periods with a large number of missing values, adjacent window interpolation or historical pattern-based filling strategies can be used to fill in missing values ​​and prevent discontinuities during the decomposition process.

[0041] The resulting drift eigenvector serves as a key input for unified correction parameter calculation. It must not only characterize temporal drift at high dimensions and resolution, but also exhibit good timing alignment and statistical stability. This stage of processing transforms inter-device clock errors from raw offsets to structural causes, providing a complete data support chain for subsequent parameter estimation and dynamic compensation.

[0042] Furthermore, performing multi-scale wavelet decomposition based on the clock offset coarse value vector to extract drift characteristic components, the drift characteristic components including instantaneous drift components, trend drift components and random jitter components, and combining the drift characteristic components to form a drift characteristic vector, including: Performing three-layer discrete wavelet decomposition on the coarse value vector of the clock offset to extract the first detail component, the second detail component and the third approximate component respectively, and obtaining a high-frequency drift subsequence, a medium-frequency drift subsequence and a low-frequency trend drift subsequence accordingly; Divide the high-frequency drift subsequence into a number of sliding segments of equal width, calculate the maximum fluctuation rate and the number of reversals in each segment, and extract the instantaneous drift component reflecting the strength of the crystal oscillator jump; Performing normalized fitting on the low-frequency trend drift subsequence, calculating the linear fitting slope and the sum of squares of the fitting residuals in each segment, and dynamically assigning confidence weights based on the sum of squares of the residuals to obtain the trend drift component; performing mean envelope processing on the variance and power spectrum density distribution extracted from the intermediate frequency drift subsequence, performing short-time energy normalization calculation on them, and extracting random jitter components used to express non-structural error disturbances; The instantaneous drift component, the trend drift component and the random jitter component are concatenated at the vector level to construct a one-dimensional multi-channel drift feature vector for use in the subsequent calculation of the unified correction parameter set.

[0043] In the proposed cross-device clock calibration method for precise meter synchronization, in-depth time series feature analysis of the coarse clock offset vector is required to achieve refined modeling and classification compensation of the meter's clock errors. The core of this step is to structurally deconstruct the offset vector using multi-scale wavelet decomposition technology, accurately extracting the three main drift characteristic components of the meter's crystal oscillator clock error: instantaneous drift, trend drift, and random jitter. Finally, this combination constructs a drift feature vector, which serves as the basic data input for the subsequent calculation of a unified set of calibration parameters.

[0044] First, a three-layer discrete wavelet decomposition is performed on the constructed coarse clock offset vector using a fixed wavelet basis function (such as the Daubechies wavelet db4). This process decomposes the original offset vector into three refined levels, each corresponding to the data representation of a frequency band. The detail coefficients obtained from the first-layer wavelet decomposition correspond to the high-frequency components, reflecting the rapid fluctuations of the crystal oscillator on short time scales, namely the high-frequency drift subsequence. The detail coefficients of the second-layer decomposition reflect the variation patterns within the medium-frequency band, defined as the medium-frequency drift subsequence, which primarily includes device temperature drift and local timing instability. The approximate coefficients in the third layer are low-frequency trend terms, representing the overall drift trend on long time scales, defined as the low-frequency trend drift subsequence.

[0045] The high-frequency drift subsequence is segmented using a sliding window. Specifically, the entire high-frequency sequence is evenly divided into smaller segments of fixed length (e.g., 50 points). The maximum fluctuation rate (the difference between the maximum and minimum values ​​within the segment divided by the average) is calculated for each segment. The number of reversals (the number of changes in the sequence's direction) is also counted. These two quantitative metrics are used together to assess the intensity and frequency of short-term crystal oscillator transitions within that segment, extracting the instantaneous drift component that reflects the severity of device jitter and forming a one-dimensional vector structure in its original order.

[0046] For the low-frequency trend drift subsequence, the maximum and minimum values ​​are first normalized to remove dimensional differences between devices. Next, the series is divided into multiple equally spaced segments (for example, 10). A linear fit is performed on each segment to obtain the fitting slope and residual sum of squares for each segment. The fitting slope measures the increasing or decreasing drift trend within that segment, while the residual sum of squares reflects the degree of fitting deviation; smaller values ​​indicate more stable trend characteristics. Based on this, confidence weights are assigned to each segment according to the inverse of the residual sum of squares. This is then used in subsequent weighted processing to construct an overall trend drift component, ensuring that this component accounts for both the temporal evolution direction and the confidence level of the change.

[0047] To capture the characteristics of unstructured noise disturbances, a short-term statistical energy analysis is performed on the intermediate frequency drift subsequence. First, the local variance and power spectral density curve of the sequence are extracted, and a sliding mean is used to generate an envelope curve to describe the energy concentration and frequency characteristics. Based on this, the envelope values ​​within each sliding window are subjected to short-term energy normalization. This involves dividing the variance within the window by the mean. This results in a series of numerical values ​​that stably express the degree of disturbance, thereby constructing a random jitter component that describes the intensity and distribution of the noise interference.

[0048] Finally, the three drift characteristic components mentioned above—instantaneous drift, trend drift, and random jitter—are concatenated according to vector dimensions to construct a one-dimensional, multi-channel drift characteristic vector. Each sub-dimension in this vector clearly represents a source of clock offset, providing both good interpretability and high data compression and reconstruction capabilities. This characteristic vector serves as direct input for the subsequent calculation of a unified correction parameter set, ensuring the accuracy, stability, and responsiveness of correction parameter estimation and providing a reliable timing modeling foundation for the entire cross-device synchronization process.

[0049] Step S104: fusing the drift feature vector and the drift history template window by window, and calculating a unified correction parameter set using a Bayesian recursive method. The unified correction parameter set includes a drift amplitude coefficient and a phase compensation coefficient.

[0050] After completing the multi-scale wavelet decomposition of the coarse clock offset vector and constructing the drift eigenvector, the system further models and estimates the parameters of the currently collected drift eigenvector to achieve dynamic prediction and compensation control of the meter's current clock drift state. The key to this process is to fully utilize the statistical laws of historical drift patterns and, in combination with the multidimensional time series structure of the current eigenvector, calculate real-time adaptive correction parameters through recursive reasoning, thereby improving the stability and predictive capabilities of the entire system.

[0051] First, a set of structured drift history templates should be pre-set within the meter system. These templates, extracted from a large amount of historical operating data under stable conditions, contain multiple annotated drift feature subsequences and their corresponding known correction parameters. Each entry in the template includes three standard drift components: instantaneous drift, trend drift, and random jitter. Their value ranges are normalized, and the drift amplitude coefficient and phase compensation coefficient, which have been verified to be effective within the corresponding time period, are stored.

[0052] After inputting the currently acquired drift feature vector, it needs to be segmented into equal-time windows and compared with the historical segments of the corresponding dimensions in the template. This matching method can use metrics such as Euclidean distance, dynamic time warping (DTW), or Mahalanobis distance to score the similarity between the feature vector and the template vector in each sliding window. In practical systems, it is recommended to use a sliding window length of 64 to 128 sampling points and a weighted sliding window mechanism, giving the window center a higher weight than the edges, thereby enhancing the matching effect of local temporal structures.

[0053] After obtaining the similarity results between the current drift feature vector and multiple template fragments, the Bayesian recursive method can be used to perform fusion reasoning on them. The core of the Bayesian method is to combine the prior probability and the observation likelihood to form a posterior probability distribution, and select the most likely drift compensation parameters based on the maximum a posteriori estimation (MAP) principle. Specifically, the system first determines the prior probability of each matching fragment based on the frequency of occurrence or usage in the template library, and constructs a likelihood function based on the current input drift feature vector and its similarity result. The calculation of the posterior probability can be completed using the Bayesian formula: ; in, Represents the local probability (Posterior), that is, in the observed data Given a certain drift compensation parameter set The probability that is the true parameter; Represents the likelihood function (Likelihood), that is, when the parameter is Under the assumption that the current observed drift eigenvector Probability of occurrence; Represents the prior probability (Prior), that is, the parameter obtained based on historical template statistics without observation data Initial estimated probability of occurrence; Represents the marginal probability of the observed data (Marginal Likelihood), that is, the current data is observed under all possible parameters This value is a constant term in the comparison of different parameters and can usually be ignored in the maximum a posteriori estimation (MAP).

[0054] Therefore, in actual calculation, in order to simplify the reasoning process, the present invention adopts the following proportional form to compare parameters: ; This formula indicates that the posterior probability is proportional to the product of the likelihood function and the prior probability, and the omitted denominator does not affect the ranking or selection when comparing multiple parameters.

[0055] In practical applications, the drift compensation parameter set It consists of two core sub-parameters, namely the drift amplitude coefficient and phase compensation coefficient , so the entire posterior inference process is essentially a two-dimensional parameter space The system ultimately determines the most appropriate correction parameter combination for the meter being corrected based on the maximum posterior probability. This mechanism not only ensures the dynamic adaptability of the drift compensation strategy but also enables parameter evolution and convergence control driven by historical data.

[0056] Combining the posterior probabilities of all available window segments, a weighted average or confidence-weighted aggregation method is used to generate a final unified correction parameter set. This set contains at least two core parameters: the drift amplitude coefficient, which indicates the amplification or reduction factor of the current drift compared to the standard state, and the phase compensation coefficient, which indicates the dominant direction and relative phase position of the time offset.

[0057] To improve model stability, a smoothing mechanism, such as an exponentially weighted moving average (EWMA), can be introduced into the final output parameter set to prevent drastic parameter jumps caused by short-term jitter. Furthermore, during long-term operation, the drift history template library should be periodically updated to remove outdated or long-invalid matching segments, ensuring that the model always reflects the actual drift behavior of the system under the current operating state.

[0058] The final unified correction parameter set generated will serve as the basic input for the subsequent construction of the correction pulse sequence, and its accuracy directly determines the effectiveness of clock synchronization compensation control.

[0059] For example, a smart energy meter completes the real-time sampling and modeling of its own crystal oscillator drift within a certain time period, and obtains a drift feature vector consisting of a set of data points with a length of 128. This vector reflects the dynamic offset trend between the meter master clock and the GNSS reference clock within the current cycle, including linear drift, nonlinear perturbations, and sudden phase jitter. Based on this drift feature vector and combined with historical templates, the system performs matching, parameter estimation, and outputs a unified set of clock correction parameters.

[0060] First, the system calls the drift history template library stored in the local NAND memory, which contains 100 structured template sequences. Each template consists of three parts: (1) A normalized historical drift feature subsequence , the length is 128; (2) the drift amplitude coefficient of the corresponding template , the value range is usually (3) Phase compensation coefficient of the corresponding template , the value range is usually between.

[0061] Then, the system converts the current drift feature vector With each template library Perform sliding window alignment. Since the current feature vector length is the same as the template, direct window alignment is used. Use Mahalanobis distance As a matching metric, the covariance weight of the drift data in each dimension is taken into account to enhance the sensitivity to local changes. The calculation formula is as follows: ; in, is the empirical covariance matrix of the historical template data.

[0062] After the matching is completed, the system selects the first 5 template fragments with the smallest distance, which are recorded as , and its corresponding parameters are .

[0063] Next, we enter the Bayesian recursion stage and first calculate the prior probability of each four-matched segment. Assume that the system records the frequency of each template being called historically , it can be normalized to the prior distribution: ; Then, the system is based on the Mahalanobis distance Construct the likelihood function for each template .definition: ; in It is the standard deviation factor determined based on the historical matching error, and the typical value can be set to 0.15.

[0064] Next, calculate the local verification probability of each candidate parameter combination: ; For all The results are normalized to obtain the standardized posterior probability distribution , satisfied Finally, the posterior confidence weighted aggregation method is used to calculate the unified correction parameter set: ; Here That is, the current meter drift amplitude correction coefficient, which is used to fine-tune the local main crystal frequency; is the phase compensation coefficient, which is used to correct the reference offset of the master clock between sampling intervals.

[0065] To ensure the stability of the output parameters, the system introduces an exponentially weighted moving average mechanism to the above estimation results: ; in, is the smoothing factor, and its value range is usually It is the parameter output from the previous calibration cycle.

[0066] In this example, it is assumed that: Matching fragments to The frequency is ; The corresponding Mahalanobis distance is ; The corresponding parameter combination is ,wait; The system finally calculates: Posterior weight ; The unified correction parameter set is .

[0067] This parameter will be directly used in the dynamic adjustment logic of the current meter clock to fine-tune the frequency control loop so that the clock synchronization error between devices is maintained within Within microseconds.

[0068] Furthermore, the drift feature vector and the drift history template are fused window by window, and a unified correction parameter set is calculated using a Bayesian recursive method. The unified correction parameter set includes a drift amplitude coefficient and a phase compensation coefficient, including: The drift feature vector obtained in the current period is divided into multiple overlapping sliding window segments in chronological order, with the length of each segment being consistent with the corresponding segment in the drift history template. The mean and fluctuation range of the instantaneous drift, trend drift, and random jitter components are extracted within each window segment. In each window segment, the Euclidean distance, maximum offset point difference, and component trend direction consistency index between the current drift feature vector and each reference segment in the drift history template are calculated, and multiple similarity score sets are generated accordingly; Confidence weights are assigned based on the cumulative number of matches for each historical template segment in the past operating cycle, the target device type, and the communication conditions. The similarity score of the current window segment is weightedly fused with the historical confidence weights point by point to form a window segment drift matching confidence sequence. The drift matching confidence sequences of all window segments are merged and accumulated, and the reference correction parameters corresponding to the optimal matching segment are extracted, including the drift amplitude factor and the phase offset bias. Dynamic smoothing is performed in the form of recursive averaging to finally generate a unified correction parameter set for the subsequent construction of the correction pulse sequence.

[0069] In a cross-device clock correction method that supports precise meter synchronization, in order to achieve dynamic correction and drift compensation of local real-time clock errors, it is necessary to obtain a representative and timely correction parameter set in a stable and traceable manner. The core of this step is to compare and analyze the drift feature vector extracted in the current period with the pre-built drift history template window by window, dynamically fuse multi-source drift features through the Bayesian recursive method, and finally calculate a set of unified correction parameters. The set specifically includes a drift amplitude coefficient for characterizing the clock drift intensity and a phase compensation coefficient for fine-tuning the sampling reference phase. The entire process emphasizes multi-window information fusion, dynamic updating of historical confidence weights, and statistical stability of time series to ensure that the output parameters can be used for high-precision synchronization adjustment of different clock systems in actual deployment.

[0070] In its implementation, the drift feature vector constructed in the current cycle is first divided into several sliding window segments of fixed length in chronological order. A certain overlap ratio (e.g., 50%) is set between each sliding window segment to enhance sensitivity to continuous temporal changes. The length of these sliding windows must be consistent with the length of each reference segment in the historical template to ensure comparability in subsequent similarity calculations across feature dimensions and duration ranges. Within each sliding window segment, three core components are further extracted: statistical features such as the mean, variance, maximum value, and variation range of instantaneous drift, trend drift, and random jitter. These features serve as input for matching analysis.

[0071] For each extracted feature in the sliding window segment, the system compares it one by one with all reference segments in the drift history template. This comparison is evaluated using a combination of multi-dimensional quantitative metrics, including Euclidean distance (which measures the numerical difference between the overall feature vectors), maximum offset point difference (which characterizes the absolute difference between the drift mutation points in the two windows), and component trend direction consistency metrics (for example, using cosine similarity to determine the consistency of the direction of change of the trend vector). These three metrics are normalized and combined to form a set of similarity scores between the window segment and each historical template segment, indicating the degree of closeness between the current observation and the historical drift pattern across different dimensions.

[0072] To further improve the credibility and adaptability of the matching results, dynamic confidence weights need to be assigned to historical template segments. These confidence weights not only reflect the frequency of successful matches of the template segment in the past actual operation cycle, but also comprehensively consider its empirical score for the current target device type and the impact of current communication conditions on the drift manifestation. The system performs a point-by-point weighted fusion of the confidence weight of each historical template segment and the similarity score corresponding to the current sliding window segment, outputting a window segment drift matching confidence sequence, which represents the strength of each historical segment's explanation of the current drift pattern.

[0073] The drift matching confidence sequences generated by all sliding window segments are accumulated and integrated in chronological order. The system extracts the corresponding historical reference correction parameters from the matching segment with the highest cumulative confidence score, including the drift amplitude factor (an overall adjustment for the error rate of change) and the phase offset (used to fine-tune the relative relationship between the sampling boundary and the external time base). To avoid drastic fluctuations in output parameters due to local transient anomalies or template drift, a smoothing mechanism based on a recursive averaging algorithm is adopted. The reference parameters obtained in consecutive cycles are fused by exponentially weighted averaging. This ensures that the final unified correction parameter set has temporal continuity, mutation robustness, and structural interpretability.

[0074] This unified set of correction parameters will be used in subsequent steps to construct and broadcast the correction pulse sequence, thereby supporting dynamic compensation adjustments for the real-time clocks of the meters being corrected. The establishment and implementation of this step effectively accommodates complex clock drift differences between meters in various application environments, enhancing the adaptability and stability of the entire cross-device clock synchronization system and ensuring the system's ability to maintain precise alignment and self-convergence of errors over the long term.

[0075] Step S105: generating a correction pulse sequence based on the unified correction parameter set, wherein the correction pulse sequence carries a delay residual compensation code and is sent to the meter to be corrected via a meter communication bus.

[0076] After calculating the unified set of correction parameters, the system must convert this set into control signals that provide real-time guidance to the meters being corrected, enabling high-precision clock synchronization across devices. To achieve this, a structured set of correction pulse sequences must be constructed and sent to the target meters via the existing meter communication bus, enabling them to perform the clock correction locally. This step is the key to converting the output of the parameter model into control commands that are perceptible, parseable, and applicable to the meters.

[0077] The construction of the correction pulse sequence must be based on the unified set of correction parameters obtained in the previous stage. This set contains at least two core quantities: the drift amplitude coefficient and the phase compensation coefficient. The drift amplitude coefficient primarily characterizes the time unit expansion and contraction offset between the current meter clock and the reference benchmark, that is, the degree of error in the timing speed per unit time. The phase compensation coefficient is used to represent the overall forward and backward offset of the meter clock relative to the reference time, that is, the absolute deviation from the synchronization point. When constructing the correction pulse sequence, these two parameters must be encoded as timing control information, supplemented with the necessary identification fields and integrity check fields.

[0078] To enhance the real-time and robustness of the correction, it is recommended to design the correction pulse sequence with a frame structure. Each frame includes at least a timestamp identifier, a parameter field, a delay residual compensation code, a device identification code, and a data check segment. The delay residual compensation code is dynamically estimated based on the inevitable delay changes generated in the current communication link, and is used to correct the slight difference between the actual reception time and the ideal time. For example, during the communication process, if the one-way average delay from the reference meter to the meter being corrected is estimated to be , a residual compensation value can be calculated by interpolation prediction or communication feedback RTT (round trip time) , and finally encode it into a 16-bit or 32-bit data segment and insert it into the correction pulse frame.

[0079] The transmission method for the entire correction pulse sequence should be compatible with the meter's existing data communication mechanism, such as through the DL / T645 bus protocol, MODBUS protocol, or master station polling structure. Within the protocol frame, existing reserved fields or vendor-extension fields can be used to carry correction data. To reduce additional bandwidth overhead, it is recommended to use periodic broadcasts or acknowledgment transmissions at a low frequency (e.g., every 1-5 minutes) rather than real-time high-frequency transmissions to balance system load and synchronization accuracy.

[0080] Furthermore, the generation of the correction pulse sequence should support redundancy protection and sequence control mechanisms. Specifically, each set of correction parameters should carry a unique sequence number and a retransmission flag to prevent drift errors caused by packet loss or mismatch in the network. Before each transmission, the reference meter should query the sequence number of the most recently successfully received signal from the meter being corrected, and based on this, decide whether to send updated data or resend the previous frame.

[0081] Once received by the meter being calibrated, the generated calibration pulse sequence becomes the sole basis for adjusting its local real-time clock. Therefore, this sequence must not only possess the fundamental properties of structural integrity, accurate parameters, and delay compensation, but also adhere to engineering specifications for communication security and fault tolerance in the energy meter system, ensuring stable cross-device synchronization even in harsh environments or with weak links.

[0082] Furthermore, the generating of a correction pulse sequence based on the unified correction parameter set, the correction pulse sequence carrying a delay residual compensation code, and sending the correction pulse sequence to the meter to be corrected via a meter communication bus, includes: Using the drift amplitude coefficient and the phase compensation coefficient included in the unified correction parameter set, a corresponding clock offset function is constructed, wherein the clock offset function is used to calculate the theoretical clock error residual of each sampling point in a future time period; Performing difference compression processing on the theoretical clock error residual, using segment extreme value extraction and incremental coding algorithm to generate a delayed residual compression vector, and mapping the compression vector into a correction codeword sequence; The correction codeword sequence is embedded into the correction pulse, synchronized with the start bit of the time pulse, and the output scheduling instruction configured by the dynamic instruction control module is scheduled and sent in time-sharing according to the current load status of the communication bus; Under the time-sharing scheduling strategy, the channel status changes of the meter communication bus are continuously monitored. When it is detected that the communication channel bandwidth decreases or the frame conflict risk increases, the correction pulse transmission is temporarily suspended and the next group of pulses is automatically queued for transmission until the communication conditions are met and the transmission process of the entire correction pulse sequence is completed.

[0083] In the cross-device clock correction method that supports precise meter synchronization described in the present invention, the construction and transmission of the correction pulse sequence is the key bridge between the unified correction parameter set and the actual clock correction action. Its accuracy and stability directly determine the effectiveness of the clock correction result and the robustness of the system operation. In order to achieve efficient and reliable correction pulse broadcasting in a multi-device, complex communication environment, it is necessary to design a complete generation and transmission process that includes a multi-layer processing mechanism, so that the correction data is not only highly compressible and time-adaptive, but also can maintain stable transmission under conditions of changing channel states. This step, based on the structural characteristics of the unified correction parameter set, unfolds the generation, encoding, embedding and scheduling processes of the correction pulse sequence layer by layer to ensure that the sequence has recoverability, low latency and strong anti-interference properties.

[0084] Specifically, the first core process of this step involves constructing a clock offset function. This function takes as input a unified set of correction parameters, consisting of two key parameters: a drift amplitude coefficient, which expresses the estimated rate of change of the local clock offset of the current device over a certain period in the future; and a phase compensation coefficient, which characterizes the difference in reference synchronization caused by hardware phase delay or system response offset during the current correction period. Based on these two parameters, a piecewise-defined clock offset function is constructed. Within each control period, this function predicts the theoretical error residual value of each future sampling point of the device using either a linear increment or a low-order polynomial recursion. This function must fully consider crystal oscillator stability, environmental noise, and the offset variation trend of the previous period. Where necessary, higher-order compensation terms are introduced to simulate nonlinear drift. For example, the drift amplitude can be reflected by a slope term, while the phase deviation can be corrected by using an offset term to modify the function's initial state. Ultimately, a theoretical clock offset curve is obtained for correction calculations during the current period.

[0085] After obtaining the theoretical clock error residual curve, the second stage is to perform compression coding to improve data transmission efficiency. In distributed metering systems, communication bus resources are usually limited, and the original error residual sequence is still large even after sampling. If it is transmitted directly, it will not only cause redundant occupation, but also increase the risk of conflict. To this end, the present invention adopts a two-layer compression mechanism: first, the segment extreme value extraction is performed, the entire residual curve is divided into several equal time segments, and the maximum residual, minimum residual and average slope change value of each segment are extracted as the compression basic tuple; then the incremental coding algorithm is used to encode the extreme value differences of adjacent segments, reducing the number of redundant bits and improving the compression ratio. In incremental coding, in order to prevent high-frequency mutation points from causing excessive code expansion, a dynamic weight adjustment strategy based on a sliding reference window is also introduced, and extra bits are used to mark points that exceed the set change threshold. After compression is completed, a structured delay residual compression vector is obtained.

[0086] This compressed vector cannot be used directly for transmission, so it is necessary to enter the third stage: mapping it into a codeword and embedding it into a correction pulse frame. Specifically, the system designs a frame structure for the correction codeword that includes a start flag bit, an error segment coding bit, and a check bit. Each codeword frame represents the error change state within a certain segment and contains an anchor point index used to identify its position in the entire correction sequence. After all codeword frames are arranged in sequence according to the generation order, a complete correction codeword sequence is formed. At the output layer, the system embeds the codeword sequence into a standard frame structure with a time pulse trigger bit, and adds a time pulse channel in a synchronous form to achieve dual logical and physical alignment. During embedding, the dynamic instruction control module is used to configure the communication scheduling parameters to generate a scheduling instruction set including the start transmission time, inter-frame interval, and the maximum cycle allowed, thereby providing a structured control basis for subsequent scheduling.

[0087] During the fourth phase, the actual transmission of correction pulses, the present invention takes into account the fact that the meter communication bus is often in a concurrent state, inevitably subject to bandwidth jitter and conflict risks. To adapt to this environment, the system introduces a time-sharing scheduling strategy based on channel state awareness. Under this strategy, the system continuously monitors the current state of the communication bus, including indicators such as channel idle time, channel collision rate in recent cycles, and transmission success rate. When the current communication load is detected to be low or high bandwidth is predicted to be available within a small period in the future, the system immediately triggers the output instruction for the currently pending pulse frame and sends the next set of codeword frames. If a significant decrease in channel bandwidth is detected, or the risk of frame transmission conflict exists (e.g., an increase in retransmissions or response timeout ratio), the transmission of the current correction pulse frame is automatically suspended, and the unsent codeword frames are temporarily buffered in the transmit queue, awaiting the next scheduling cycle when trigger conditions are restored. This strategy significantly enhances the communication system's load awareness and effectively mitigates the risk of interference with critical correction pulses during periods of high conflict.

[0088] The entire calibration pulse sequence construction and transmission process demonstrates a high degree of adaptability and structural rigor. By employing a highly compressed, structured encoding mechanism at the data generation end and a dynamic perception scheduling mechanism at the transmission scheduling end, the system ensures stable transmission and low-error recovery of critical data required for cross-device clock calibration under various communication conditions. The system is not only compatible with existing meter communication protocols but also enables rapid deployment across multiple devices through firmware upgrades, significantly improving meter network synchronization capabilities and timing consistency control.

[0089] It's worth noting that during the construction of the correction pulse, additional fields can be introduced, such as drift signal-to-noise ratio estimates, pulse signal quality indicators, and statistics of the previous cycle's drift prediction error. This information can be used to determine the effectiveness and confidence of the correction pulse in the meter being corrected. This information can serve as an auxiliary control variable, guiding the corrected device to adopt more targeted compensation strategies or feedback mechanisms, thereby forming a cross-device clock collaborative control system with closed-loop learning capabilities.

[0090] In summary, this step plays a key role in connecting the previous and next steps in the present invention. It not only performs a highly structured conversion on the parameter set extracted in the previous stage, but also provides a high-precision, low-error input basis for the dynamic correction of the local real-time clock in the subsequent stage.

[0091] Step S106: The correction pulse sequence is received in the meter to be corrected, and an interpolation compensation function is called based on the delay residual compensation code to generate a drift compensation table, wherein the drift compensation table is a piecewise polynomial structure, and the local real-time clock is dynamically corrected according to the drift compensation table to achieve cross-device clock correction.

[0092] After receiving the calibration pulse sequence sent by the reference meter, the meter being calibrated must immediately perform targeted processing to ensure that its local clock accurately adjusts to the time correction information carried in the sequence. The delay residual compensation code contained in the calibration pulse sequence is a highly accurate estimate of the dynamic jitter and transmission delay uncertainty in the actual communication link. It provides a critical basis for compensating the actual reception time against the expected arrival time. Therefore, the primary task in the processing flow of the meter being calibrated is to correctly parse this compensation code from the received data frame and effectively decode and interpret it.

[0093] After parsing the compensation code, the meter being calibrated must invoke a set of preloaded or dynamically generated interpolation compensation functions to calculate a clock drift compensation table using the current compensation parameters. This compensation table should construct a set of mappable time correction paths through mathematical interpolation, ensuring that the local real-time clock running within the subsequent system can continuously and specifically adjust the offset based on the characteristics of different time periods. To meet the actual needs of system deployment in large-scale industrial sites, the compensation function must be adaptable to nonlinear drift behavior. It is recommended to use methods such as piecewise cubic polynomial interpolation, cubic spline interpolation, or piecewise Lagrange interpolation. The specific function form should strike a balance between the number of interpolation nodes, computational complexity, and memory overhead.

[0094] The generated drift compensation table should be organized in a segmented structure. Each segment corresponds to a sampling time interval. Each segment consists of key fields such as the start and end time, the corresponding drift correction coefficient, and the function form (such as polynomial coefficients). For example, in a compensation segment, it may be specified that the time arrive The time output of the real-time clock needs to be corrected as follows: ; in , Indicates the time point of the current correction.

[0095] in, Indicates the corrected real-time clock value, Represents the current local clock raw value. a, b, c, and d are the coefficients of the interpolation function within this interval, calculated from the parameters in the received correction pulse.

[0096] Once constructed, this drift compensation table is written to the local time buffer or clock control register interface of the meter being calibrated, serving as the primary correction reference for the real-time clock control module. During each control cycle, the clock control module reads the interpolation expression corresponding to the current time interval from the compensation table and applies it in real time to the output logic of the system's master clock. This enables continuous, dynamic, and nonlinear compensation for errors caused by long-term crystal oscillator aging, short-term jitter, or sudden drift, going beyond simple time alignment or fixed offset calibration.

[0097] It's worth noting that this compensation mechanism isn't limited to instantaneous corrections at the current moment; it supports continuous tracking and adjustment over a period of minutes to hours. To achieve this, the control logic within the meter being calibrated must support a cyclical, rolling compensation table refresh mechanism. This mechanism automatically replaces expired or no longer applicable segments in the compensation table with newly generated segments upon receipt of each new calibration pulse sequence, implementing a continuous, progressive drift control strategy.

[0098] Furthermore, implementation must account for abnormalities such as communication anomalies, missing compensation codes, or interpolation failures. Therefore, default fault-tolerance mechanisms should be implemented, such as falling back to the last successfully received compensation table, enabling linear simplified compensation mode, or automatically reporting to the master station for recalibration. All processing flows must possess deterministic execution logic, timestamp tracking, and boundary condition assessment capabilities to ensure stable operation and easy maintenance of the entire calibration mechanism in large-scale, multi-site deployments.

[0099] Through the above process, the calibrated meter can independently complete segmented, multi-scale, and dynamic adjustment of its real-time clock without relying on external hardware synchronization signals, thereby achieving precise time consistency across multiple devices. This not only solves the problem of data timing misalignment caused by asynchronous device sampling, but also provides a solid time foundation for subsequent power data processing steps such as event reconstruction, load identification, and anomaly location.

[0100] For example, in a typical power distribution network scenario, reference meters are aligned with the Universal Coordinated Time (UTC) via the Network Time Protocol (NTP) and periodically broadcast calibration pulse sequences to multiple calibrated meters. Suppose a calibrated meter receives the following calibration pulse sequence, which contains valid information for the current calibration period: Correction segment start time: =1000 ms; Correction segment end time: =2000 ms; Delay residual compensation code: =0.56 ms; Drift compensation reference node (calculated and embedded by the system according to the previous steps): ; After receiving the data frame, the meter being calibrated immediately performs the following operations: First, the correction period time interval and the reference node sequence are parsed, and the built-in cubic spline interpolation algorithm is called to generate the current correction interval Interpolation compensation function on Cubic spline interpolation is a function fitting method that defines a third-order polynomial on each adjacent node interval and ensures that the entire function has continuous first-order and second-order derivatives on the entire interval. Its form is as follows: ; in, is the current real-time clock value, independent variable, in milliseconds; The polynomial coefficients of the spline interpolation in each small interval; Current time The corresponding correction offset value in the interval; all The standard cubic spline interpolation algorithm is used according to the node values The solution is that the first-order and second-order derivatives between the interpolation segments are continuous, ensuring that the overall correction curve is smooth.

[0101] The meter being corrected will then delay the residual compensation code Add a constant term to each interpolation segment , and perform uniform drift and floating processing on it to construct a complete interpolation compensation function: ; Next, the system constructs a drift compensation table based on the interpolation compensation function. Each record contains: Start and end time interval (e.g. [1000, 1250] ms); Corresponding polynomial coefficients ; Time period identifier and version information; Time calibration flag and check field.

[0102] After the compensation table is built, the meter writes the table into the dedicated buffer of the local real-time clock control module. The control logic obtains the current system time in real time during each sampling cycle or control cycle. , determine the compensation segment it is in, and substitute the interpolation function to complete the real-time execution of the following clock correction formula: ; in, Indicates the corrected time, used for data upload, event recording, etc. Indicates the current uncorrected real-time clock value; Indicates the offset that should be applied at the current time point, which comes from the interpolation function.

[0103] For example, in , assuming the meter's current local time is: ; The system calculates the correction term at this point as: (from node); Adding the 0.56 ms delay residual compensation code, the total offset is: ; Final calibration time: ; This value will serve as the standard time output for data writing, synchronization reference, or upload to the master station within the system. The system repeats the above process every control cycle and refreshes the compensation table when a new correction pulse arrives, forming a closed-loop adaptive update mechanism.

[0104] It can be seen from this embodiment that: the interpolation compensation function It is formed by fitting a set of discrete drift offsets through the spline method, and has a continuous, differentiable, and interpolable mathematical structure; each They represent high-order nonlinear drift, trend offset, linear offset, and static compensation, respectively. This function is ultimately used for real-time clock correction and exhibits excellent computational stability, physical rationality, and system compatibility. It can be directly embedded in existing meter firmware or edge time management chips for implementation.

[0105] Furthermore, the correction pulse sequence is received in the corrected meter, an interpolation compensation function is called based on the delay residual compensation code to generate a drift compensation table, the drift compensation table is a piecewise polynomial structure, and a local real-time clock is dynamically corrected according to the drift compensation table to achieve cross-device clock correction, including: Receive the correction pulse sequence and perform structural analysis on the delayed residual compensation code, extract the compression vector index, drift change amplitude and phase offset term therein, and construct a residual reconstruction node sequence; According to the time distribution density of the residual reconstruction node sequence, an adaptive interpolation compensation function selection mechanism is called to dynamically determine the optimal interpolation strategy among piecewise linear, polynomial and spline function families to form a target interpolation function for fitting the error drift change trend in each time segment; The target interpolation function is segmented, and the error distribution time axis is divided into several correction segments. A corresponding piecewise polynomial expression is constructed based on the node fitting results in each segment to form a drift compensation table with a time series index. The drift compensation table uses the segment number as a key value and records the coefficient vector and applicable time domain of each piecewise polynomial; Based on the drift compensation table, a dynamic correction cycle threshold is set within the local real-time clock operation cycle, the current correction segment is located according to the segment number, the corresponding piecewise polynomial is called to solve the error compensation value corresponding to this moment, and the master clock output frequency and phase are adjusted through the internal clock frequency fine-tuning unit to achieve precise alignment of clocks across devices.

[0106] In the cross-device clock correction method for supporting precise meter synchronization described in the present invention, the final key step is to receive the correction pulse sequence generated and transmitted via the communication bus within the meter being corrected. Based on the delay residual compensation code carried in the sequence, a drift compensation table that can be independently executed on the device side is constructed to dynamically correct the local real-time clock and achieve high-precision clock alignment with the host system or reference device. This process not only requires effective decoding and parsing of compressed correction data, but also requires a precise interpolation fitting mechanism and an efficient error compensation application strategy to ensure the real-time and stability of the correction process. It is suitable for various business scenarios requiring cross-device timing consistency, such as electricity metering, synchronous sampling, and power monitoring.

[0107] First, the meter being calibrated receives a correction pulse sequence transmitted from an external control device via a communication interface module. This pulse sequence is typically encapsulated in a standardized clock synchronization protocol format and carries key information, including a compressed sequence of delay residual compensation codewords. After receiving this sequence, the meter system uses a local decoding module to perform structured parsing of the data segments within it, extracting the index information, residual values, and time period identifiers contained in each correction codeword. This is used to construct a chronological sequence of residual reconstruction nodes. Each node contains its position within the sampling period, the residual change amplitude (i.e., the expected deviation from the reference clock at the current time point), and the corresponding phase compensation term, which is used as a reference for subsequent clock frequency adjustments.

[0108] After obtaining the sequence of residual reconstruction nodes, the system initiates an adaptive interpolation compensation function selection mechanism based on the node distribution density and change trend along the time axis. This mechanism is designed to select the most appropriate interpolation strategy based on the distribution characteristics of the error nodes, ensuring high fitting accuracy and computational efficiency of the error function throughout the entire cycle. The interpolation strategy library primarily includes piecewise linear interpolation functions, fixed-order polynomial interpolation functions (such as cubic or quintic polynomials), and spline functions (such as cubic splines and B-splines), which offer higher fitting accuracy but relatively higher computational complexity. During the selection process, the system first analyzes the average spacing between the current nodes and the standard deviation of the error slope. If the node spacing is large and the change trend is approximately linear, piecewise linear interpolation is preferred. If the node density is high and the error trend is nonlinear but smooth overall, polynomial interpolation is preferred. If there are obvious inflection points or localized error abrupt changes, spline functions are preferred for fitting. The core of this mechanism is to automatically balance computational complexity and compensation accuracy, ensuring that the interpolation function accurately reflects the actual drift changes without causing overfitting or numerical oscillation.

[0109] After determining the interpolation strategy, the system performs an error curve fitting operation on each time segment. That is, based on the set of reconstructed nodes in the segment, the selected target interpolation function is called to solve the coefficients to obtain a complete error fitting expression. In order to improve operational efficiency and the convenience of subsequent calls, the system divides the entire error timeline into several continuous but non-overlapping correction segments. Each segment is identified by a start and end timestamp, and the fitted target interpolation function is used internally to express the error changes within the segment. The interpolation function expressions in all segments are in standard polynomial form, for example: ; Each segment number is recorded, along with the corresponding polynomial coefficient vector and applicable time interval, forming a table entry. This ultimately forms a drift compensation table. This table is typically structured as a key-value pair, where the key is the segment number and the value is a composite data structure containing the coefficient vector and time boundaries. This table can be directly loaded into the meter's local RAM or FLASH storage area for subsequent dynamic corrections.

[0110] After completing the construction of the drift compensation table, the actual time calibration phase begins. The meter system will set several dynamic correction cycle thresholds within the operating cycle of its internal real-time clock. For example, a correction operation will be triggered every hundred clock increments, or a compensation query will be performed before each data sampling point. Each time a correction is triggered, the system first locates the error correction segment based on the current local clock count value, that is, it searches the segment number corresponding to the current moment in the drift compensation table and obtains the polynomial coefficient vector of the segment. Subsequently, the system calculates the current moment value. Substitute it into the polynomial function of this segment and solve for the error compensation value δt at this moment, that is, the time offset that the current clock should theoretically correct.

[0111] After obtaining the compensation value, the system performs error correction via an internal clock frequency fine-tuning unit. This fine-tuning unit typically consists of a programmable digitally controlled oscillator (DCO) or clock correction registers, capable of adjusting the master clock's output frequency or phase by microseconds or even nanoseconds based on the input offset. When the compensation value is positive, the system slightly increases the oscillation frequency within the current cycle to catch up with the reference clock. When the compensation value is negative, the pulse output is appropriately delayed to align the local time with the reference time. If the system supports a more refined time adjustment mechanism, a cycle-by-cycle phase interpolation mechanism can be used to smoothly transition between two sampling points, further improving correction accuracy. This clock correction process continues throughout the entire correction cycle until the error compensation corresponding to all correction segments has been applied, and the drift compensation table is updated after the next round of correction pulses arrives.

[0112] Furthermore, to prevent error compensation information anomalies caused by communication errors or unusual interference, the system has designed a correction pulse validity verification mechanism. Each time the drift compensation table is reconstructed or called, the system performs CRC checks, codeword redundancy correction, and timestamp integrity checks on the received delay residual compensation code. Furthermore, if the error trend significantly deviates from the normal drift pattern, an abnormal drift warning mechanism is triggered, prompting the system to enter protection mode and suspend the application of the current correction sequence to avoid inaccurate corrections to the local clock.

[0113] In summary, this step plays a key role in the overall cross-device clock calibration process, converting external correction information into dynamic adjustment parameters for the local real-time clock. Through structured decoding of compressed information, selection of adaptive interpolation functions, construction of segmented polynomials, and real-time application of high-precision error compensation, this solution not only balances compensation accuracy and system operational efficiency, but also exhibits strong communication adaptability and error tolerance in complex environments. This provides a highly engineered and deployable technical path for achieving precise cross-device clock synchronization in power meter systems, exhibiting excellent scalability and versatility, and adapting to a variety of network structures and synchronization accuracy requirements.

[0114] A second embodiment of the application provides an electronic device, comprising: processor; The memory is used to store a program. When the program is read and executed by the processor, it executes a cross-device clock correction method that supports precise synchronization of meters provided in the first embodiment of the present application.

[0115] The third embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, a cross-device clock correction method supporting precise synchronization of meters provided in the first embodiment of the present application is executed.

[0116] Although the present application is disclosed as above with the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art may make possible changes and modifications without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be based on the scope defined by the claims of the present application.

Claims

1. A cross-device clock calibration method supporting precise meter synchronization, characterized in that: include: Acquire a network time protocol timestamp and a first sampling time sequence in a reference meter, and generate a clock reference signal by combining the network time protocol timestamp and the first sampling time sequence; broadcasting the clock reference signal to the meter to be calibrated via a communication link, recording a second sampling time series in the meter to be calibrated, and calculating a clock offset coarse value vector based on the clock reference signal and the second sampling time series; Performing multi-scale wavelet decomposition based on the clock offset coarse value vector to extract drift characteristic components, wherein the drift characteristic components include an instantaneous drift component, a trend drift component, and a random jitter component, and combining the drift characteristic components to form a drift characteristic vector; The drift feature vector is fused with the drift history template window by window, and a unified correction parameter set is calculated using a Bayesian recursive method, wherein the unified correction parameter set includes a drift amplitude coefficient and a phase compensation coefficient; generating a correction pulse sequence based on the unified correction parameter set, the correction pulse sequence carrying a delay residual compensation code, and sending the correction pulse sequence to the corrected meter via a meter communication bus; The correction pulse sequence is received in the corrected meter, and an interpolation compensation function is called based on the delay residual compensation code to generate a drift compensation table, where the drift compensation table has a piecewise polynomial structure. The local real-time clock is dynamically corrected according to the drift compensation table to achieve cross-device clock correction.

2. The cross-device clock correction method supporting precise meter synchronization according to claim 1, characterized in that: The obtaining of a network time protocol timestamp and a first sampling time sequence in a reference meter, and combining the network time protocol timestamp and the first sampling time sequence to generate a clock reference signal includes: In the reference meter, a first network timestamp and a second network timestamp are obtained from a public network NTP server and a GNSS receiver respectively, and the local time of reception and the corresponding time value are recorded; Compare and analyze the time difference between the first network timestamp and the second network timestamp with the historical offset change trend, assign weights based on their stability, jitter range, and delay symmetry, perform a weighted average operation, and generate a fused timestamp as the reference time value for the current period; Collecting multiple clock count values ​​of the local crystal oscillator within a preset sampling period to form a first sampling time series, and linearly corresponding each sampling point in the series to the fused timestamp to generate a set of calibration sampling time series aligned with a unified reference time; The calibration sampling time sequence and the fusion timestamp of the current cycle are encapsulated together to form a clock reference signal including a time reference type identification field, a synchronization error assessment field and a sampling frequency stability flag, which is subsequently broadcast to the calibrated meter.

3. The cross-device clock correction method supporting precise meter synchronization according to claim 1, characterized in that: Broadcasting the clock reference signal to the calibrated meter via a communication link, recording a second sampling time series in the calibrated meter, and calculating a clock offset coarse value vector based on the clock reference signal and the second sampling time series, includes: Splitting the clock reference signal into two parts including a first time anchor point field and a first sampling alignment parameter field, encapsulating each part into two communication data frames, and broadcasting the two parts to each calibrated meter via the communication link, so that the calibrated meter can parse and restore the complete clock reference signal according to a preset logic; After the calibrated meter receives the first time anchor point field, it starts a real-time collection task for a second sampling time series, records multiple system current times locally at fixed time intervals, and timestamps each recorded point to correspond to the received first time anchor point field, generating a second sampling time series with an anchor point index identifier; Performing a one-to-one matching of the first sampling alignment parameter field in the clock reference signal with the anchor point index in the second sampling time series, and extracting a set of successfully matched sampling pairs using a sequential window comparison mechanism for subsequent offset calculation; For each pair of time points in the set of successfully matched sampling pairs, the time difference between the sampling point in the clock reference signal and the corresponding sampling point in the second sampling time series is calculated to obtain multiple offset samples, and the samples are arranged in chronological order to form a coarse clock offset value vector. The coarse clock offset value vector serves as input for extracting the drift feature component in subsequent steps.

4. The cross-device clock correction method supporting precise meter synchronization according to claim 1, characterized in that: The method of performing multi-scale wavelet decomposition based on the coarse clock offset value vector to extract drift characteristic components, wherein the drift characteristic components include an instantaneous drift component, a trend drift component, and a random jitter component, and combining the drift characteristic components to form a drift characteristic vector includes: Performing three-layer discrete wavelet decomposition on the coarse value vector of the clock offset to extract the first detail component, the second detail component and the third approximate component respectively, and obtaining a high-frequency drift subsequence, a medium-frequency drift subsequence and a low-frequency trend drift subsequence accordingly; Divide the high-frequency drift subsequence into a number of sliding segments of equal width, calculate the maximum fluctuation rate and the number of reversals in each segment, and extract the instantaneous drift component reflecting the strength of the crystal oscillator jump; Performing normalized fitting on the low-frequency trend drift subsequence, calculating the linear fitting slope and the sum of squares of the fitting residuals in each segment, and dynamically assigning confidence weights based on the sum of squares of the residuals to obtain the trend drift component; performing mean envelope processing on the variance and power spectrum density distribution extracted from the intermediate frequency drift subsequence, performing short-time energy normalization calculation on them, and extracting random jitter components used to express non-structural error disturbances; The instantaneous drift component, the trend drift component and the random jitter component are concatenated at the vector level to construct a one-dimensional multi-channel drift feature vector for use in the subsequent calculation of the unified correction parameter set.

5. The cross-device clock correction method supporting precise meter synchronization according to claim 1, characterized in that: The drift feature vector and the drift history template are fused window by window, and a unified correction parameter set is calculated using a Bayesian recursive method. The unified correction parameter set includes a drift amplitude coefficient and a phase compensation coefficient, including: The drift feature vector obtained in the current period is divided into multiple overlapping sliding window segments in chronological order, with the length of each segment being consistent with the corresponding segment in the drift history template. The mean and fluctuation range of the instantaneous drift, trend drift, and random jitter components are extracted within each window segment. In each window segment, the Euclidean distance, maximum offset point difference, and component trend direction consistency index between the current drift feature vector and each reference segment in the drift history template are calculated, and multiple similarity score sets are generated accordingly; Confidence weights are assigned based on the cumulative number of matches for each historical template segment in the past operating cycle, the target device type, and the communication conditions. The similarity score of the current window segment is weightedly fused with the historical confidence weights point by point to form a window segment drift matching confidence sequence. The drift matching confidence sequences of all window segments are merged and accumulated, and the reference correction parameters corresponding to the optimal matching segment are extracted, including the drift amplitude factor and the phase offset bias. Dynamic smoothing is performed in the form of recursive averaging to finally generate a unified correction parameter set for the subsequent construction of the correction pulse sequence.

6. The cross-device clock correction method supporting precise meter synchronization according to claim 1, characterized in that: The generating of a correction pulse sequence based on the unified correction parameter set, wherein the correction pulse sequence carries a delay residual compensation code and is sent to the meter to be corrected via a meter communication bus, comprises: Using the drift amplitude coefficient and the phase compensation coefficient included in the unified correction parameter set, a corresponding clock offset function is constructed, wherein the clock offset function is used to calculate the theoretical clock error residual of each sampling point in a future time period; Performing difference compression processing on the theoretical clock error residual, using segment extreme value extraction and incremental coding algorithm to generate a delayed residual compression vector, and mapping the compression vector into a correction codeword sequence; The correction codeword sequence is embedded into the correction pulse, synchronized with the start bit of the time pulse, and the output scheduling instruction configured by the dynamic instruction control module is scheduled and sent in time-sharing according to the current load status of the communication bus; Under the time-sharing scheduling strategy, the channel status changes of the meter communication bus are continuously monitored. When it is detected that the communication channel bandwidth decreases or the frame conflict risk increases, the correction pulse transmission is temporarily suspended and the next group of pulses is automatically queued for transmission until the communication conditions are met and the transmission process of the entire correction pulse sequence is completed.

7. The cross-device clock correction method supporting precise meter synchronization according to claim 1, characterized in that: The corrected meter receives the correction pulse sequence, calls an interpolation compensation function based on the delay residual compensation code to generate a drift compensation table, wherein the drift compensation table is a piecewise polynomial structure, and dynamically corrects the local real-time clock according to the drift compensation table to achieve cross-device clock correction, including: Receive the correction pulse sequence and perform structural analysis on the delayed residual compensation code, extract the compression vector index, drift change amplitude and phase offset term therein, and construct a residual reconstruction node sequence; According to the time distribution density of the residual reconstruction node sequence, an adaptive interpolation compensation function selection mechanism is called to dynamically determine the optimal interpolation strategy among piecewise linear, polynomial and spline function families to form a target interpolation function for fitting the error drift change trend in each time segment; The target interpolation function is segmented, and the error distribution time axis is divided into several correction segments. A corresponding piecewise polynomial expression is constructed based on the node fitting results in each segment to form a drift compensation table with a time series index. The drift compensation table uses the segment number as the key value and records the coefficient vector and applicable time domain of each piecewise polynomial. Based on the drift compensation table, a dynamic correction cycle threshold is set within the local real-time clock operation cycle, the current correction segment is located according to the segment number, the corresponding piecewise polynomial is called to solve the error compensation value corresponding to the current moment, and the master clock output frequency and phase are adjusted through the internal clock frequency fine-tuning unit to achieve precise alignment of clocks across devices.

Images