A method and system for synchronizing data acquisition of multiple sensors

Abstract

The application discloses a kind of data acquisition synchronization method and system of multipath sensor, it is related to power system automation and data acquisition technical field.The method includes: based on main clock source establishes global reference time, for each sensor node configuration local soft clock;Acquire sensor data and attach local time stamp, synchronous record clock offset estimation value and data quality identification, form data frame;According to multidimensional time information, different sensor data is mapped to uniform virtual time axis, and the preprocessed data set of time sequence consistency is generated;Correlation analysis is carried out to preprocessed data set, and the clock drift trend and synchronization error of each sensor node are dynamically evaluated;Based on evaluation result, generate feedback instruction, dynamically adjust the compensation logic and data fusion strategy of local soft clock.By embedding the physical law of power system into the synchronization error detection link, combined with global reference time and closed-loop feedback mechanism, the whole-link adaptive cooperation from data acquisition to fusion is realized.

Description

Technical Field

[0001] This invention relates to the field of power system automation and data acquisition technology, specifically to a method and system for synchronizing data acquisition from multiple sensors. Background Technology

[0002] With the deepening of smart grid construction, the number and types of sensor nodes deployed in power systems have increased dramatically. Various intelligent electronic devices (IEDs), such as smart meters, substation monitoring units, merging units, fault recorders, and phasor measurement units (PMUs), are widely distributed, undertaking the task of collecting multi-dimensional power data including voltage, current, frequency, power, switch status, harmonics, and fault alarms. This data is the cornerstone supporting power grid situational awareness, fault diagnosis, precise dispatching, and equipment status assessment. Since electrical quantities such as voltage, current, and phase angle are all time-dependent, the synchronization accuracy of data acquisition directly affects the reliability of relay protection actions, the accuracy of fault location, and the confidence level of system stability analysis. Therefore, ensuring high-precision synchronous acquisition of data from multiple sensors distributed in different geographical locations, with different sampling frequencies, and different triggering methods under a unified time reference is one of the core technical requirements for the safe and stable operation of power systems.

[0003] In existing power system multi-sensor data acquisition systems, time synchronization solutions mainly rely on the following technical approaches: Time synchronization based on satellite navigation systems (GPS / BeiDou). This method uses satellite time signals as a synchronization clock to control the sampling pulses of each acquisition terminal. However, satellite signals are easily attenuated by obstruction in substations or underground facilities, and the stability of the time receiving equipment faces severe challenges when electromagnetic interference is strong at the power site. After the satellite signal is lost, each terminal relies on its local crystal oscillator to keep time, and the clock drift accumulates over time, causing the synchronization accuracy to continuously deteriorate.

[0004] Network synchronization is based on Network Time Protocol (NTP) or IEEE 1588 Precision Time Protocol (PTP). This method achieves frequency and phase synchronization between master and slave clocks through network message exchange and has been widely used in smart substations. However, synchronization accuracy is significantly affected by uncertainties such as network topology, switching delay, and message queuing. When the network load is high, the transmission delay jitter of synchronization messages directly affects the accuracy of clock offset estimation. In addition, the PTP protocol requires network device hardware support, resulting in complex system deployment and high modification costs.

[0005] Centralized synchronization based on unified time synchronization commands. This method involves the master station uniformly issuing time synchronization commands to each terminal. It places extremely high demands on the real-time performance and reliability of the communication link; communication interruptions or command delays will lead to global synchronization failure. For example, the "A Steady-State Data Synchronization Acquisition and Transmission Method and System" (publication number pending) applied for by XJ Electric Co., Ltd. in 2025, while avoiding modifications to the protocol and master station software, has its synchronization granularity limited by a preset period multiple, making it difficult to meet the microsecond-level synchronization accuracy requirements of sudden events such as transient faults and harmonic anomalies.

[0006] Furthermore, the above solutions all share the following common shortcomings: First, data alignment remains at the protocol layer rather than the physical layer—even if each node completes synchronization with the master clock source, due to differences in sampling time, transmission delay, and crystal oscillator drift characteristics, the multi-source data arriving at the fusion center is still not truly aligned on the time axis, forming an "asynchronous after synchronization" problem, directly affecting the accuracy of upper-layer applications such as fault diagnosis and state estimation. Second, they lack closed-loop feedback and adaptive adjustment capabilities—existing solutions mostly adopt an "open-loop" design, where each node operates independently according to fixed parameters after synchronization configuration, unable to dynamically monitor and differentiate compensation for clock deviation changes caused by factors such as network environment changes, equipment aging, and temperature drift. Third, the three stages of acquisition, synchronization, and fusion are isolated from each other, lacking an end-to-end global coordination mechanism, making it difficult to consistently maintain optimal synchronization accuracy and fusion quality.

[0007] In summary, existing multi-sensor data acquisition and synchronization technologies for power systems suffer from several shortcomings, including highly susceptible-to-environmental-affected synchronization accuracy, failure to achieve physical-layer consistency in data alignment, lack of dynamic adjustment capabilities based on data quality feedback, and the independence of various system components. There is an urgent need for a data acquisition and synchronization method and system capable of achieving high-precision synchronous acquisition, possessing adaptive adjustment capabilities, and a closed-loop feedback mechanism. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for synchronizing data acquisition from multiple sensors, addressing the aforementioned problems. By introducing a global reference time, a local soft clock, and multi-dimensional time information, high-precision synchronous acquisition of data from various sensor data sources can be achieved.

[0009] The technical solution of the present invention is as follows: A method for synchronizing data acquisition from multiple sensors includes the following steps: When a data synchronization command is received, a global reference time is established based on the master clock source; Configure a local soft clock for each sensor node based on the global reference time; Collect sensor data from each sensor node; A local timestamp is added to the sensor data using a local soft clock, and the estimated clock offset and data quality identifier are recorded synchronously to form a data frame; the data frame carries multi-dimensional time information. Based on multidimensional time information, different sensor data are mapped to a unified virtual time axis to generate a time-consistent preprocessed dataset; Correlation analysis is performed on the preprocessed dataset to dynamically evaluate the clock drift trend and synchronization error of each sensor node and obtain the evaluation results; Feedback instructions are generated based on the evaluation results to dynamically adjust the compensation logic of the local soft clock and the data fusion strategy.

[0010] By establishing a global reference time and configuring local soft clocks for each sensor node, each sensor node can maintain synchronization with the global reference time even after network fluctuations or loss of external time signals, reducing dependence on external time sources and improving synchronization stability in complex power environments. By attaching a local timestamp, clock offset estimate, and data quality identifier to each sensor data point to form a data frame carrying multi-dimensional time information, and mapping data from different sensors onto a unified virtual time axis for alignment, timing misalignment caused by differences in sampling time, transmission delay, and crystal oscillator drift characteristics among nodes is eliminated, achieving true time consistency of multi-source data at the physical level. Through correlation analysis, the clock drift trend and synchronization error of each node are dynamically evaluated, and feedback instructions are generated based on the evaluation results to dynamically adjust the compensation logic of the local soft clock and the data fusion strategy. This constructs a closed-loop feedback mechanism of acquisition-synchronization-fusion, ensuring that synchronization accuracy and fusion quality are continuously maintained at optimal levels, significantly reducing system synchronization overhead and data redundancy.

[0011] Furthermore, the association analysis of the preprocessed dataset includes: Identify and extract multi-sensor data fragments describing the same physical event from a preprocessed dataset; Compare the alignment consistency of the data segments from various multi-sensor systems on the virtual time axis, and verify the rationality of their physical logic based on the physical laws of the power system; Physical logic validity checks include at least one of the following: Based on Kirchhoff's current law, a vector sum verification is performed on the three-phase current measurements at the same electrical node: the effective value of the vector sum is calculated. ,when Exceeding the preset first deviation threshold When this occurs, it is determined that there is a physical logic inconsistency, and the deviation is converted into an additional residual synchronization error; Based on three-phase voltage balance constraints, amplitude consistency verification is performed on the three-phase voltage measurements at the same electrical node: the maximum deviation of the three-phase voltage amplitude is calculated. ,in ,when Exceeding the preset second deviation threshold If the physical logic is unreasonable, the deviation is converted into an additional residual synchronization error.

[0012] By introducing a three-phase current vector sum verification based on Kirchhoff's current law and a voltage amplitude consistency verification based on three-phase voltage balance constraints, the inherent physical laws of the power system are used as the evaluation criteria for data alignment quality. When the vector sum of the three-phase currents or the amplitude deviation of the three-phase voltages exceeds a preset threshold, synchronization errors that cannot be detected by timestamp alignment alone can be accurately identified. This technique, which integrates the physical laws of the power system into synchronization error detection, means that the detection of synchronization errors no longer relies solely on statistical methods but incorporates physical layer consistency information, significantly improving detection accuracy and reliability. Simultaneously, the unreasonable deviations in physical logic are converted into additional residual synchronization errors, providing a more accurate quantitative basis for subsequent clock correction and fusion strategy adjustments.

[0013] Furthermore, the step of mapping different sensor data to a unified virtual timeline based on multidimensional time information to generate a time-consistent preprocessed dataset includes: Configure an independent buffer for each sensor data stream to temporarily store data frames carrying multi-dimensional time information; Determine the target reference time point based on the preset fusion time granularity or event trigger signal; Based on the local timestamp and clock offset estimate, the state estimate of each sensor data point at the target reference time point is calculated and mapped onto the virtual time axis. Align and package the state estimates of all data streams at the target reference time point to generate a preprocessed dataset.

[0014] By configuring an independent buffer for each sensor data stream, the problem of inconsistent data arrival times caused by differences in network transmission latency is effectively solved. Determining the target reference time point through a preset fusion time granularity or event-triggered signal can simultaneously accommodate different alignment requirements for periodic steady-state data and sudden transient data. By estimating the state of each sensor at the target reference time point based on local timestamps and clock offset estimates, all data streams are precisely aligned to the same moment. The resulting preprocessed dataset exhibits high temporal consistency, providing a reliable data foundation for subsequent correlation analysis and fault diagnosis.

[0015] Furthermore, the calculation of the state estimate of each sensor data point at the target reference time point includes: Construct a correlation model for sensor data; the correlation model includes an evolution model in the time dimension and a correlation model in the spatial dimension; The evolutionary model in the time dimension is: taking the target reference time point. Adjacent intervals Aligned data points within The least squares method is used to fit the linear relationship. ,in: , , Will Substituting the linear relationship, we obtain the time dimension estimate. ; The spatial correlation model is based on a pre-calibrated spatial correlation matrix derived from the power system's network topology and line impedance parameters. When the first sensor is at the target reference time point When missing data is encountered, the spatial dimension estimate is obtained by multiplying the aligned data from other available sensors at the same time by the corresponding electrical quantity transfer coefficient. ; When the time dimension estimates and spatial dimension estimates are obtained independently, the weighted average is taken as the final state estimate. ,in This is a linear regression estimate over the time dimension. This is an estimate of the spatial dimension correlation matrix. and For the corresponding weights, .

[0016] By constructing a joint correlation model that incorporates a temporal evolution model and a spatial correlation model: in the temporal dimension, historical data near the target time is used to fit the variation of measured values ​​over time, and the state is predicted by fitting a linear relationship using the least squares method; in the spatial dimension, a spatial correlation matrix is ​​pre-calibrated based on the power system's network topology and line impedance parameters, and spatial estimation is performed using data from other correlated sensors at the same time. The final state estimate is obtained by weighted averaging after independent estimations in these two dimensions, integrating information from both time and space. This approach utilizes both the historical evolution of individual sensors and the inherent physical correlations between different sensors, significantly improving the accuracy and robustness of state estimation in cases of missing data or mismatched sampling times.

[0017] Furthermore, weight and Dynamically allocate based on the confidence levels of the two estimates: When the number of historical data points And goodness of fit hour, The value ranges from 0.6 to 0.8; When the amount of calibration data for the spatial correlation matrix is ​​sufficient and the number of available sensors participating in spatial estimation is ≥3, The value ranges from 0.5 to 0.7.

[0018] The system dynamically assigns weights to the time and spatial dimension estimates based on their respective confidence levels: when there are many historical data points and a high goodness of fit, the weight of the time dimension estimate is increased; when the spatial correlation matrix calibration data is sufficient and there are many associated sensors, the weight of the spatial dimension estimate is increased. Through this dynamic weight allocation mechanism, the system can adaptively adjust the fusion strategy according to the actual data situation, always assigning higher confidence estimates higher weights, thereby ensuring that the fusion accuracy remains at a good level under various data conditions.

[0019] Furthermore, the identification and extraction of multi-sensor data fragments describing the same physical event includes: A sliding window is used to scan the preprocessed dataset, with a window length of three to five power frequency cycles; Calculate the rate of change of the magnitude of the data within the window. and root mean square deviation ; When the amplitude change rate or root mean square deviation of any sensor data stream exceeds the preset detection threshold, it is determined that there is a potential event within the window. Then, with the sensor as the center, the search for other data streams is expanded by two to five window lengths before and after along the time axis. Data segments that show synchronous characteristic changes within the same time window are extracted and collected to form a multi-sensor data segment set describing the same event.

[0020] By scanning the preprocessed dataset through a sliding window, the amplitude change rate and root mean square deviation of the data within the window are calculated as dual feature indicators for event detection. When either feature exceeds a preset threshold, a potential event is identified, and data segments in other data streams showing synchronous feature changes are searched along the time axis centered on that sensor. This dual feature indicator joint detection mechanism effectively avoids the problem of misjudgment by a single indicator, improving the accuracy and reliability of event detection. By setting the window length and search range, the complete reflection of the same event in multi-sensor data can be accurately captured, providing an accurate data foundation for subsequent alignment consistency assessment and residual synchronization error calculation.

[0021] Furthermore, comparing the alignment consistency of the various multi-sensor data segments on the virtual time axis includes: Extract the moments of signal abrupt change start points or peak points on the virtual time axis from each data segment; Using the sensor corresponding to the median time as a reference, calculate the time difference between the feature point times of other sensors and the reference time. ; Calculate the standard deviation of the time difference As a quantitative indicator of alignment consistency; The residual synchronization error is calculated using the following formula: ,in The theoretical time difference caused by differences in the event propagation path. The time equivalent of the unreasonable deviation in physical logic.

[0022] By extracting the start or peak times of signal abrupt changes in each data segment, and using the sensor corresponding to the median time as a reference, the standard deviation of the time difference between the characteristic points of other sensors and the reference time is calculated as a quantitative indicator of alignment consistency. A smaller standard deviation indicates higher alignment consistency. When calculating residual synchronization error, the theoretical time difference caused by differences in event propagation paths is deducted, and the time equivalent calculated from physical logical inconsistencies is added. This method ensures that the calculation of residual synchronization error considers both alignment deviations on the time axis and logical inconsistencies at the physical level, resulting in a more comprehensive and accurate evaluation. The quantification using standard deviation allows for an objective measurement of the synchronization accuracy of each node, providing a quantitative basis for subsequent feedback correction.

[0023] Furthermore, the step of generating feedback instructions based on the evaluation results to dynamically adjust the compensation logic and data fusion strategy of the local soft clock includes: The evaluation results include a drift trend index and a synchronization error index; the drift trend index is the slope of the linear regression of the long-term residual synchronization error sequence. The synchronization error index is the standard deviation of the long-term residual synchronization error sequence. ; when Exceeding the preset drift threshold When a systemic drift is detected in the local soft clock, a clock correction command containing updated compensation parameters is generated. After receiving the command, the node replaces the corresponding parameters in the local compensation logic. When all nodes None exceeded But there is at least one node Exceeding the preset quality threshold If the fusion quality fails to meet the preset requirements, a data realignment strategy instruction is generated, which includes at least one of the following: reducing the fusion time granularity to half of the original value, replacing linear interpolation with cubic spline interpolation, or adjusting the confidence weight allocation strategy of the cross-validation step.

[0024] The evaluation results are quantified into two dimensions: a drift trend indicator and a synchronization error indicator. The drift trend indicator reflects the directional drift of the clock leading continuously, while the synchronization error indicator reflects the fluctuation of synchronization accuracy. When the drift trend indicator exceeds a preset threshold, a systematic drift is identified, and a clock correction command is issued to update the compensation parameters, thus suppressing the cumulative effect of clock drift at its source. When the drift trend indicator does not exceed the threshold but the synchronization error indicator exceeds the quality threshold, the fusion quality is identified as degraded, and a data realignment strategy command is issued. This improves the fusion quality by shortening the fusion time granularity, changing the interpolation algorithm, or adjusting the confidence weight allocation. These two feedback mechanisms are executed in parallel and complement each other, enabling the system to adaptively track the clock drift characteristics of each sensor node during long-term operation. High-precision synchronization can be maintained continuously without manual intervention, significantly reducing system maintenance costs and synchronization overhead.

[0025] This application also includes a data acquisition and synchronization system for multiple sensors, and a method for data acquisition and synchronization of multiple sensors, including: The reference establishment module is used to establish a global reference time based on the master clock source when a data synchronization command is received. The soft clock configuration module is used to configure a local soft clock for each sensor node based on the global reference time. The data acquisition module is used to collect sensor data from each sensor node; The data frame forming module is used to add a local timestamp to the sensor data using a local soft clock, synchronously record the estimated clock offset and data quality identifier, and form a data frame; the data frame carries multi-dimensional time information. The preprocessing dataset generation module is used to map different sensor data to a unified virtual time axis based on multidimensional time information, and generate a time-consistent preprocessing dataset. The correlation analysis module is used to perform correlation analysis on the preprocessed dataset, dynamically evaluate the clock drift trend and synchronization error of each sensor node, and obtain the evaluation results. The feedback instruction generation module is used to generate feedback instructions based on the evaluation results, so as to dynamically adjust the compensation logic and data fusion strategy of the local soft clock.

[0026] This application also includes an electronic device, including a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, run a multi-sensor data acquisition synchronization method.

[0027] Compared with existing technologies, the advantages of this invention are: 1. Improve synchronization accuracy and environmental adaptability. Addressing the issues of existing satellite time synchronization being susceptible to obstruction and network synchronization being affected by latency jitter, this application establishes a global reference time and configures each sensor node with a dynamically compensated local soft clock. This enables sensor nodes to maintain high-precision synchronization with the global reference time even after network fluctuations or satellite signal loss, reducing dependence on external time sources and improving synchronization stability in complex power environments.

[0028] 2. Achieving physical-level data alignment. Addressing the issue that existing solutions only align data at the protocol level and that multi-source data still exhibits asynchronous behavior after synchronization before fusion, this application adds a local timestamp, clock offset estimate, and data quality identifier to each sensor data point to form a data frame carrying multi-dimensional time information. It then precisely maps data from different sensors onto a unified virtual time axis for alignment, eliminating timing misalignments caused by differences in sampling times, transmission delays, and crystal oscillator drift characteristics among nodes, thus achieving true time consistency of multi-source data at the physical level.

[0029] 3. Achieve dynamic monitoring and adaptive compensation for synchronization errors. Addressing the lack of continuous monitoring and dynamic adjustment capabilities for synchronization errors in existing open-loop schemes, this application performs correlation analysis on the preprocessed dataset to identify multi-sensor data segments describing the same physical event. By comparing their alignment consistency and physical logic rationality on the virtual time axis, the residual synchronization error and its drift trend of each node are calculated, enabling quantitative evaluation and long-term tracking of the clock synchronization status of each sensor node.

[0030] 4. Construct a closed-loop feedback mechanism for data acquisition, synchronization, and fusion. Addressing the issue of existing solutions where data acquisition, time synchronization, and data fusion are independent and lack global coordination, this application generates feedback instructions based on evaluation results. It dynamically adjusts the compensation logic of the local soft clock and the data fusion strategy, ensuring that synchronization accuracy and fusion quality are consistently maintained at their optimal levels. This achieves adaptive coordination from data acquisition and edge alignment to global evaluation, significantly reducing system synchronization overhead and data redundancy, and improving the timeliness and resource utilization efficiency of multi-source data fusion. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the data acquisition and synchronization method for multiple sensors according to this application.

[0032] Figure 2 This is a detailed flowchart illustrating the data acquisition and synchronization method for multiple sensors according to this application. Figure 3 This is a structural block diagram of a multi-channel sensor data acquisition and synchronization system according to this application.

[0033] Figure 4A structural block diagram of an electronic device provided in this application. Detailed Implementation

[0034] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0035] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0036] Please see Figure 1-4 A method for synchronizing data acquisition from multiple sensors, such as Figure 1 and Figure 2 As shown, it includes the following steps: When a data synchronization command is received, a global reference time is established based on the master clock source; Configure a local soft clock for each sensor node based on the global reference time; Collect sensor data from each sensor node; A local timestamp is added to the sensor data using a local soft clock, and the estimated clock offset and data quality identifier are recorded synchronously to form a data frame; the data frame carries multi-dimensional time information. Based on multidimensional time information, different sensor data are mapped to a unified virtual time axis to generate a time-consistent preprocessed dataset; Correlation analysis is performed on the preprocessed dataset to dynamically evaluate the clock drift trend and synchronization error of each sensor node and obtain the evaluation results; Feedback instructions are generated based on the evaluation results to dynamically adjust the compensation logic of the local soft clock and the data fusion strategy.

[0037] Configure a local soft clock for each sensor node based on the global reference time, including: The relative clock offset sequence and frequency deviation sequence between each sensor node and the master clock source are measured according to a preset cycle. An adaptive filtering algorithm is used to smooth the clock offset sequence and frequency deviation sequence and predict their trends. Dynamic compensation is generated based on the prediction results and applied to the local soft clock to maintain synchronization with the global reference time.

[0038] The adaptive filtering algorithm is a sliding weighted average filtering algorithm, where the sliding window length is... The value is dynamically configured based on the crystal stability of the sensor node and the network environment. The value range is 5 to 20, preferably... The dynamic compensation quantity includes two components: frequency compensation value and phase compensation value.

[0039] Based on multidimensional temporal information, different sensor data are mapped to a unified virtual timeline, generating a time-consistent preprocessed dataset, including: Configure an independent buffer for each sensor data stream to temporarily store data frames with multidimensional time information; Determine the target reference time point based on the preset fusion time granularity or event trigger signal; Based on the local timestamp and clock offset estimate, the state estimate of each sensor data at the target reference time point and mapped onto the virtual time axis is calculated; Align and package the state estimates of all data streams at the target reference time point to generate a preprocessed dataset.

[0040] Based on the local timestamp and clock offset estimate, the state estimate of each sensor data point at the target reference time point and mapped onto the virtual time axis is calculated, including: Construct a correlation model for sensor data; the correlation model is used to describe the evolution of the same sensor data over time, as well as the spatial correlation between different sensor data. When the current sensor node is missing data at the corresponding target reference time point, the correlation model is used to interpolate or predict the state at the time points near the target reference time point to obtain the state estimate. When cross-sensor correlation is required, the correlation model and aligned sensor data are used to cross-validate and correct the current sensor state estimates.

[0041] The correlation model includes a linear regression model in the time dimension and a spatial correlation matrix model in the spatial dimension. The spatial correlation matrix is ​​based on the network topology of the power system and the impedance parameters of each line, and its elements are the electrical quantity transfer coefficients between different electrical nodes or between different electrical quantities in the same node.

[0042] A correlation analysis was performed on the preprocessed dataset to dynamically evaluate the clock drift trend and synchronization error of each sensor node, yielding evaluation results, including: Identify and extract multi-sensor data fragments describing the same physical event or state from the preprocessed dataset; By comparing the alignment consistency and physical logic of multi-sensor data segments on the virtual time axis, the residual synchronization error of each sensor node is calculated. Based on the long-term sequence of residual synchronization error, the corresponding drift patterns and trends are identified to dynamically evaluate clock drift trends and synchronization errors, and the evaluation results are obtained.

[0043] Physical logic rationality includes verification of the three-phase current vector sum based on Kirchhoff's current law and verification of voltage amplitude consistency based on three-phase voltage balance constraints.

[0044] Feedback instructions are generated based on the evaluation results to dynamically adjust the compensation logic of the local soft clock and the data fusion strategy, including: If the evaluation results indicate that there is a systematic drift in the local soft clock, a clock correction instruction containing the parameters of the deviation prediction model is generated to update the compensation logic; If the evaluation results indicate that the fusion quality of the preprocessed dataset does not meet the preset requirements, a data realignment strategy is generated. The strategy is used to indicate whether to change the interpolation algorithm or shorten the fusion time granularity when generating the preprocessed dataset.

[0045] The evaluation results include a drift trend index and a synchronization error index; the drift trend index is the slope of the linear regression of the long-term residual synchronization error series, and the synchronization error index is the standard deviation of the long-term residual synchronization error series.

[0046] This application also includes a data acquisition and synchronization system for multiple sensors, comprising: The reference establishment module is used to establish a global reference time based on the master clock source when a data synchronization command is received. The soft clock configuration module is used to configure a local soft clock for each sensor node based on the global reference time. The data acquisition module is used to collect sensor data from each sensor node; The data frame forming module is used to add a local timestamp to the sensor data using a local soft clock, synchronously record the estimated clock offset and data quality identifier, and form a data frame; the data frame carries multi-dimensional time information; The preprocessing dataset generation module is used to map different sensor data to a unified virtual time axis based on multidimensional time information, and generate a time-consistent preprocessing dataset. The correlation analysis module is used to perform correlation analysis on the preprocessed dataset, dynamically evaluate the clock drift trend and synchronization error of each sensor node, and obtain the evaluation results. The feedback instruction generation module is used to generate feedback instructions based on the evaluation results in order to dynamically adjust the compensation logic and data fusion strategy of the local soft clock.

[0047] The correlation analysis module includes a physical logic rationality verification unit, which is used to perform vector sum verification on the three-phase current measurement values ​​of the same electrical node based on Kirchhoff's current law, and to perform amplitude consistency verification on the three-phase voltage measurement values ​​of the same electrical node based on three-phase voltage balance constraints.

[0048] The technical solutions in this application will now be clearly and completely described with reference to the accompanying drawings.

[0049] Please see Figure 1 This application discloses a data acquisition synchronization method for multiple sensors, comprising: Step S101: When a data synchronization command is received, a global reference time is established based on the master clock source.

[0050] Data synchronization commands can be issued by the system's host computer or data fusion center upon system startup, timed triggering, or when a decrease in synchronization accuracy is detected. Upon receiving the command, the system connects to a high-precision physical time synchronization source (such as GPS, BeiDou Navigation Satellite System, NTP server, or internal high-stability rubidium atomic clock) to logically construct a unified, monotonically increasing global reference time within the power system control level. This serves as the absolute reference for time synchronization of all network devices. The global reference time is based on International Standard Time (UTC) with an accuracy of no less than 1 microsecond.

[0051] Specifically, when the master clock source is GPS or the BeiDou satellite navigation system, the system obtains UTC time by receiving satellite timing signals and achieves sub-microsecond clock synchronization accuracy through pulse-per-second (PPS) signals. When the master clock source is an NTP server, the system obtains UTC time through the Network Time Protocol, with synchronization accuracy at the millisecond level. When the master clock source is an internal high-stability rubidium atomic clock, the system maintains timekeeping accuracy by relying on the high stability of the atomic clock itself without external timing signals, achieving an accuracy better than 1 microsecond per day. The system automatically selects an available master clock source and establishes a global reference time based on the actual deployment scenario and equipment configuration.

[0052] When the system has multiple master clock sources simultaneously, a multi-source redundancy backup strategy is adopted: GPS / BeiDou satellite signals are used as the first priority master clock source, the NTP server as the second priority master clock source, and the internal high-stability rubidium atomic clock as the third priority master clock source. When a high-priority clock source signal is lost or its accuracy drops below a preset threshold, the system automatically switches to the next lower priority backup clock source to ensure the continuity and stability of the global reference time. The switching process uses a smooth transition algorithm, that is, a weighted transition is performed between the old and new clock sources, and the transition time is a preset transition window (e.g., 60 seconds) to avoid global reference time jumps caused by clock source switching.

[0053] Step S102: Configure a local soft clock for each sensor node based on the global reference time.

[0054] Using a global reference time as the standard, network clock synchronization protocols (such as NTP or IEEE 1588 PTP) guide each power smart terminal to complete initial time synchronization and frequency calibration. Specifically, the master clock source periodically sends synchronization messages to each sensor node. Each node calculates the initial deviation between its local clock and the global reference time based on the round-trip time of the messages and corrects its local clock accordingly. Building upon this, a local soft clock, maintained and dynamically adjusted by software, is created within each terminal device. This local soft clock is independent of the device's hardware clock, and its count value can be accelerated or decelerated using software algorithms to achieve dynamic tracking of the global reference time.

[0055] The local soft clock's count is maintained by a software timer, which uses the device's hardware clock as its clock source. Dynamic adjustment of the local soft clock's frequency and phase is achieved by setting the timer interrupt period and the count adjustment logic within the interrupt service routine. Specifically, when the local soft clock needs to be sped up, the count is incremented by a step greater than 1 in each timer interrupt; when the local soft clock needs to be slowed down, the count is incremented by a step less than 1; and when the current speed needs to be maintained, the step is equal to 1. The precision of the step is no less than 10. -6 This allows for fine-tuning of the local soft clock.

[0056] The initial time of the local soft clock is set to the current value of the global reference time, and the initial frequency is set to the nominal frequency of the device's hardware clock. During subsequent operation, the frequency and phase of the local soft clock are continuously adjusted according to the dynamic compensation amount in step S204.

[0057] Step S103: Collect sensor data from each sensor node.

[0058] Each smart power terminal begins collecting target physical quantities based on the sampling strategy or event triggering conditions (such as relay contact action, fault occurrence, harmonic exceedance, etc.) of its built-in sensors (including but not limited to voltage transformers, current transformers, frequency sensors, power sensors, temperature sensors, vibration sensors, switch status sensors, etc.). Each sensor node performs analog-to-digital conversion on the analog signal according to a preset sampling frequency (such as 128 points, 256 points per cycle, or a frequency set according to actual needs), generating a raw data stream that can be transmitted. For status-type data such as switch quantities, collection is triggered when the status changes.

[0059] For analog sensors, each node employs an equally spaced sampling strategy, with the sampling frequency determined based on the highest frequency component of the measured signal, satisfying the Nyquist sampling theorem. For voltage and current signals in power systems, the sampling frequency is no less than 64 points per cycle, preferably 256 points per cycle, to balance measurement accuracy and data transmission overhead. For digital sensors (such as switch status sensors), a position change-triggered acquisition strategy is adopted, recording the position change time and the subsequent state value only when the state changes, thus reducing the generation and transmission of redundant data.

[0060] For different types of sensors, the system supports configuring different sampling frequencies: high sampling frequencies (e.g., 256 or 512 points per cycle) are used for rapidly changing quantities such as voltage and current; low sampling frequencies (e.g., once per second or once per minute) are used for slowly changing quantities such as temperature and pressure; and switch status quantities are triggered by position changes. The data streams with these different sampling frequencies are uniformly aligned in subsequent steps through virtual timeline mapping.

[0061] Step S104: Use a local soft clock to add a local timestamp to the sensor data, synchronously record the estimated clock offset and data quality identifier, and form a data frame.

[0062] At the instant sensor data generation or acquisition is completed, the terminal uses its dynamically compensated local soft clock to timestamp the data point locally. Simultaneously, it records the estimated clock offset of the local clock relative to the global reference time, as well as the data quality identifier of the data frame. A positive clock offset estimate indicates that the node's local soft clock is ahead of the global reference time, while a negative estimate indicates lag. The data quality identifier includes, but is not limited to, data validity flags (valid / invalid), confidence levels (high / medium / low), signal quality indicators (such as signal-to-noise ratio), and acquisition channel status (normal / abnormal). This information, along with the raw sampled data, is encapsulated to form a multi-dimensional data frame carrying complete temporal context.

[0063] The data frame structure comprises three parts: a frame header, a data body, and a frame trailer. The frame header includes a data frame type identifier, sensor node number, data frame length, and frame sequence number. The data body includes a local timestamp (64-bit unsigned integer in microseconds), a clock offset estimate (32-bit signed integer in microseconds, positive values ​​indicating lead and negative values ​​indicating lag), a data quality identifier (8-bit unsigned integer, where bit 0 indicates validity, bits 1-2 indicate confidence level, bits 3-4 indicate signal quality, and bits 5-7 are reserved), raw sampled data (32-bit floating-point number or 16-bit integer, determined according to sensor type), and a checksum (16-bit CRC check). The frame trailer includes a frame end marker. This data frame structure ensures the complete carrying and reliable transmission of multi-dimensional time information.

[0064] The confidence level of the data quality identifier is comprehensively evaluated based on the signal-to-noise ratio (SNR) and the status of the acquisition channel: when the SNR is higher than 40dB and the acquisition channel is in normal condition, the confidence level is "high"; when the SNR is between 20dB and 40dB and the acquisition channel is in normal condition, the confidence level is "medium"; when the SNR is lower than 20dB or the acquisition channel is in abnormal condition, the confidence level is "low". Data frames with a "low" confidence level will be assigned a lower weight or marked as pending verification in subsequent data fusion processing.

[0065] Step S105: Based on multidimensional time information, different sensor data are mapped to a unified virtual time axis to generate a time-consistent preprocessed dataset.

[0066] By analyzing multi-dimensional time information such as local timestamps and clock offset estimates in each data frame, and employing precise time transformation and interpolation algorithms, sensor data from different smart power terminals, different sampling frequencies, and different event triggers are accurately mapped onto a unified virtual time axis based on a global reference time. Specifically, the local timestamp of each data frame is subtracted from the corresponding clock offset estimate to obtain the actual occurrence time of the data under the global reference time. Then, all data are arranged on the virtual time axis according to their actual occurrence time. For cases where there are multiple data points or missing data points for the same target time on the virtual time axis, interpolation or fitting algorithms are used for unified processing. This ensures the temporal consistency of all data, forming a preprocessed dataset that can be used for efficient correlation analysis.

[0067] The time conversion uses the following formula to calculate the global actual occurrence time of the data: , in For the local timestamp in the data frame, This is the estimated clock offset value in the data frame.

[0068] In this process, when multiple data points exist at the target reference time point on the virtual timeline, a weighted average method is used to merge them into a single data point, with the weights determined by the confidence level of each data frame. When data points are missing at the target reference time point, nearest neighbor interpolation or linear interpolation methods are used to supplement the data points. The choice of interpolation method is adaptively determined based on the rate of change of the data: when the rate of change of the data exceeds a preset threshold, linear interpolation is used to track rapid changes; when the rate of change of the data does not exceed the preset threshold, nearest neighbor interpolation is used to reduce computational overhead.

[0069] Step S106: Perform correlation analysis on the preprocessed dataset to dynamically evaluate the clock drift trend and synchronization error of each sensor node and obtain the evaluation results.

[0070] A correlation analysis is performed on the preprocessed dataset to identify the reflection of the same power system event (such as a short-circuit fault, load switching operation, harmonic anomaly, etc.) in multi-sensor data, and to check its precise alignment accuracy on the virtual time axis. Specifically, firstly, multi-sensor data segments describing the same physical event or state are identified and extracted from the preprocessed dataset; then, the alignment consistency of each data segment on the virtual time axis is compared, and the physical logic rationality is verified based on the inherent physical laws of the power system (such as Kirchhoff's laws, power balance relationships, etc.); based on the alignment deviation, the residual synchronization error of the local soft clock of each sensor node is calculated. The residual synchronization error of each node is continuously collected to form a long-term sequence, and the drift mode and change trend of each node are identified through statistical analysis. The clock drift trend and synchronization error of each sensor node are dynamically evaluated to obtain the evaluation results.

[0071] The association analysis specifically includes the following sub-steps: A sliding window scan was used to preprocess the dataset. The window length was set to 3 to 5 power frequency cycles (i.e., 60 to 100 milliseconds) based on the power system frequency cycle, and the step size was set to 1 power frequency cycle (20 milliseconds). Statistical features were calculated for the data within each window. When the statistical feature of any data stream exceeds the preset detection threshold, the window is marked as an event window, and multi-sensor data segments within the window are extracted. Alignment consistency analysis and physical logic rationality analysis are performed on the extracted multi-sensor data segments to calculate residual synchronization error; The residual synchronization errors are stored in chronological order to form a long-term sequence.

[0072] Step S107: Generate feedback instructions based on the evaluation results to dynamically adjust the compensation logic of the local soft clock and the data fusion strategy.

[0073] Based on the evaluation results of step S106, two types of control commands are generated and issued: The first type is the clock correction command, which sends a clock correction command containing deviation prediction model parameters to a specific sensor node that has systematic drift. After receiving the command, the node updates the corresponding parameters in the compensation logic of its local soft clock, thereby improving the time synchronization accuracy at the terminal level from the source.

[0074] The second category is data realignment strategy instructions, which send dynamic data fusion strategy instructions to the data preprocessing module. These instructions specify methods such as changing the interpolation algorithm or shortening the fusion time granularity when generating the preprocessed dataset, in order to improve the fusion quality and analysis reliability of multi-source heterogeneous power data at the application level.

[0075] The two types of feedback mentioned above can be executed in parallel or separately, so that the synchronization accuracy and fusion quality are continuously maintained within the preset range.

[0076] The clock correction command and the data realignment strategy command employ different issuance cycles: the clock correction command takes effect immediately upon issuance by the node, with the effective time being the moment the node receives the command plus a preset processing delay (e.g., 100 milliseconds); the data realignment strategy command takes effect at the start of the next fusion cycle by the preprocessing module, with the effective time being the start of the next preset fusion time granularity. This differentiated effective time setting avoids system jitter caused by the simultaneous activation of clock correction and strategy adjustments.

[0077] Please see Figure 2 The specific implementation methods for each step were further refined. This method includes: Step S201: When a data synchronization command is received, a global reference time is established based on the master clock source.

[0078] The specific implementation method of this step is the same as that of step S101, and will not be repeated here.

[0079] Step S202: Measure the relative clock offset sequence and frequency deviation sequence between each sensor node and the master clock source according to a preset cycle.

[0080] During operation, each power smart terminal (such as smart meters, IEDs, and RTUs) communicates and compares with the master clock source of its region according to a preset time period (such as every 1 second, every 5 seconds, or other periods set according to accuracy requirements). Specifically, the terminal interacts with the master clock source according to the IEEE 1588 PTP or NTP protocol, recording the time difference between the transmission of the synchronization message from the master clock source to the node and the time difference between the transmission of the delay request message from the node to the master clock source. Half of the difference between the two is taken as the clock offset measurement value at the current moment. At the same time, the frequency deviation measurement value is calculated by comparing the change in clock offset between two adjacent measurements and dividing by the time interval. The above measurement values ​​are recorded in chronological order to form a clock offset sequence and a frequency deviation sequence.

[0081] The specific calculation method for the clock offset measurement is as follows: set up The moment when the master clock source sends a synchronization message (Sync message). The time at which the synchronization message is received from the clock node. The time when a delay request message (Delay_Req message) is sent from the clock node. The clock offset measurement is the time when the master clock source receives the delay request message. for: , in , , , All times are local times recorded by the local clocks of each device. The specific calculation method for the frequency deviation measurement is as follows: , in This is the current clock offset measurement. For the previous moment ( The clock offset measurement (before time). The time interval between the two measurements.

[0082] The preset synchronization period is adaptively configured based on the crystal oscillator type of the sensor node: for nodes using temperature-compensated crystal oscillators (TCXO), the synchronization period can be set to 5 to 10 seconds; for nodes using ordinary crystal oscillators, the synchronization period needs to be set to 1 to 2 seconds to ensure the accuracy of the clock offset measurement. The above synchronization period can be dynamically extended when the network load is high to reduce the network bandwidth occupied by synchronization messages.

[0083] Step S203: An adaptive filtering algorithm is used to smooth the clock offset sequence and frequency deviation sequence and predict their trends.

[0084] To extract the true terminal clock drift pattern from the clock deviation measurement data, which includes network jitter and instantaneous measurement errors, an adaptive filtering algorithm is used to process the clock offset sequence and frequency deviation sequence obtained in step S202.

[0085] This application employs a sliding weighted average filtering algorithm to smooth and predict the trend of the clock offset sequence and frequency deviation sequence obtained in step S202.

[0086] In the specific implementation, for each newly acquired clock offset observation, a fixed-length measurement window of 10 is maintained according to the first-in, first-out (FIFO) principle. After the window is filled, the ten data points within the window are assigned weights from most recent to furthest in time, with the most recent time having the highest weight and the furthest time having the lowest weight, and the sum of all weights is 1. The smoothed estimate of the current time is obtained by multiplying each measurement value within the window by its corresponding weight and summing the results. Through the above weighting process, the influence of recent measurement data on the smoothing result is greater than that of historical data, thus effectively filtering out short-term random noise while quickly tracking the true trend of clock drift. The frequency deviation sequence is processed independently using the same moving weighted average filtering algorithm. The clock offset estimate and frequency deviation estimate output after smoothing will be used as input data for generating the dynamic compensation in step S204.

[0087] The weight allocation in the moving weighted average filtering algorithm uses either a linearly decreasing or exponentially decreasing method: When using a linear decreasing method, within the window Weight of each data point for: , in This represents data from the most recent moment. This represents the data at the furthest point in time.

[0088] When using an exponentially decreasing method, within the window Weight of each data point for: , in The attenuation factor has a value ranging from 0.1 to 0.5, preferably... .

[0089] Among them, the sliding window length Dynamically adjust based on the crystal stability of the sensor node: when the crystal stability is higher than a preset threshold (e.g., daily stability better than 10), -9 When the crystal oscillator stability is below a preset threshold (e.g., daily stability is worse than 10), the window length can be set to 15 to 20. -8When using this feature, the window length should be set to 5 to 10. By dynamically adjusting the window length, an optimal balance can be achieved between smoothness and responsiveness.

[0090] Step S204: Generate a dynamic compensation amount based on the prediction results and apply the compensation amount to the local soft clock to maintain synchronization with the global reference time.

[0091] Based on the smoothed estimate and trend prediction results of the output, the adjustment amount required to offset the expected future drift is calculated, i.e., the dynamic compensation amount. This compensation amount is dynamically applied to the terminal's local software clock logic, enabling the terminal's local software clock to accurately track the global reference time, maintaining high-precision synchronization in the logic output even in the event of inherent drift in the hardware clock.

[0092] The dynamic compensation amount includes two components: frequency compensation value and phase compensation value. Frequency compensation value ,in The smoothed frequency deviation estimate output in step S203 is used to adjust the counting frequency of the local soft clock, thereby suppressing the accumulation of clock drift at its source. Phase compensation value ,in The smoothed clock offset estimate output in step S203 is used to gradually correct the current reading of the local soft clock and eliminate accumulated phase errors.

[0093] The frequency compensation value and phase compensation value are applied as follows: the frequency compensation value is accumulated in steps each time a timer interrupt occurs, and the phase compensation value is gradually applied at a preset correction rate (e.g., no more than 10 microseconds per second) to avoid clock jumps caused by a large one-time correction. The correction rate is configured according to the system's requirements for clock stability, and is preferably set to no more than 10% of the current offset per second.

[0094] Step S205: Collect sensor data from each sensor node.

[0095] The specific implementation method of this step is the same as that of step S103, and will not be repeated here.

[0096] Step S206: Use a local soft clock to add a local timestamp to the sensor data, synchronously record the estimated clock offset and data quality identifier, and form a data frame.

[0097] At the instant sensor data generation or acquisition is completed, the node uses its local soft clock to timestamp the data point, and simultaneously records the estimated clock offset and data quality identifier, encapsulating them together with the original data into a structured data frame. The estimated clock offset is calculated as follows: Each sensor node interacts with the master clock source via a precise time synchronization protocol according to a preset period. It records the time difference between the synchronization message sent from the master clock source to the node and the time difference between the delay request message sent from the node to the master clock source. Half of this difference is taken as the estimated clock offset for the current moment. A positive value indicates that the node's local soft clock is ahead of the global reference time, while a negative value indicates lag. Between two synchronization interactions, the node continuously tracks and predicts the clock offset using the frequency deviation estimate output in step S203, ensuring that each frame of data includes the latest clock offset estimate.

[0098] The clock skew estimate is continuously updated between two synchronization interactions in the following manner: , in This is the estimated clock offset value measured during the most recent synchronization interaction. The smoothed frequency deviation estimate output from step S203. For the current moment, This refers to the most recent synchronization interaction time. Through this continuous update method, we ensure that each data frame carries the most accurate clock offset estimate for the current moment, maintaining high accuracy even with long intervals between synchronization interactions.

[0099] The data quality identifier also includes a timestamp validity flag. When a node fails to successfully complete synchronization with the master clock source within the preset synchronization timeout period, the timestamp validity flag is set to "invalid", indicating that the timestamps of data frames generated during that period may have a large error. When the node successfully resumes synchronization, the timestamp validity flag is restored to "valid".

[0100] Step S207: Configure an independent buffer for each sensor data stream to temporarily store data frames with multidimensional time information.

[0101] Data frames from different smart power terminals or different sensors on the same terminal are first temporarily stored in their respective buffers, awaiting subsequent unified scheduling and processing, thus solving the problem of inconsistent data arrival times.

[0102] The capacity of each buffer is pre-set based on the sampling frequency and data frame size of the corresponding sensor, with the goal of storing at least all data frames within a preset time period (e.g., 1 second). Specifically, for voltage and current sensors with a sampling frequency of 256 points per cycle (i.e., 12,800 points per second), the buffer capacity is at least 12,800 frames; for temperature sensors with a sampling frequency of once per second, the buffer capacity is at least 1 frame. The buffers are implemented using a circular queue structure. When the buffer is full, new data overwrites the oldest data, ensuring that the buffer always stores the latest data frames.

[0103] Each buffer independently maintains its own read and write pointers. The write pointer is controlled by the data acquisition module, and the read pointer is controlled by the preprocessing dataset generation module. When the number of frames between the read and write pointers is less than a preset safety threshold (e.g., 10 frames), the preprocessing dataset generation module pauses reading and waits for new data to arrive to avoid data loss due to an empty buffer.

[0104] Step S208: Determine the target reference time point based on the preset fusion time granularity or event trigger signal.

[0105] One of two strategies is adopted to determine the target reference time point for the alignment operation: one is to perform fusion periodically based on a fixed time interval, i.e., the fusion time granularity; the other is to initiate fusion immediately when a specific event trigger signal is detected. This time point is a logical moment on a unified virtual timeline.

[0106] The preset fusion time granularity is any one of 1 millisecond, 2 milliseconds, 5 milliseconds, or 10 milliseconds. The specific value is set according to the time resolution requirements of the upper-layer application: for relay protection applications, the fusion time granularity is set to 1 millisecond or less; for condition monitoring applications, the fusion time granularity can be set to 5 milliseconds to 10 milliseconds. Event trigger signals include at least one of voltage sag detection signals, short-circuit fault detection signals, harmonic over-limit detection signals, and switch displacement detection signals.

[0107] Step S209: Based on the local timestamp and clock offset estimate, calculate the state estimate of each sensor data at the target reference time point and mapped onto the virtual time axis.

[0108] This involves constructing a correlation model for sensor data. This model describes the temporal evolution of the same sensor data and the spatial correlation between different sensor data. When a sensor node lacks data at its target reference time point, the correlation model is used to interpolate or predict the state at nearby time points to obtain a state estimate. When cross-sensor correlation is required, the correlation model and aligned sensor data are used to cross-validate and correct the current sensor's state estimate.

[0109] The correlation model employs a linear regression model, fitting the linear relationship between measured values ​​and time using historical data frames within the vicinity of the target reference time point. The target reference time point is then substituted into this relationship to calculate the state estimate. For scenarios involving multiple sensors measuring the same physical quantity, the correlation model also includes a spatial correlation matrix pre-calibrated using historical normal operation data. When data from a particular sensor is missing, the state estimate for that sensor is obtained by multiplying the aligned data from other correlated sensors at the same time by the corresponding scaling factor in the spatial correlation matrix. When the two methods yield estimates independently, a weighted average is taken as the final value.

[0110] In the specific implementation, for each sensor data stream in the cache, the local timestamp and clock offset estimate in its data frame are used as input. Through a time transformation model, the data is uniformly transformed from its original local time coordinates (with offset) to a virtual time axis based on a global reference time. For sensors that lack data at the target reference time point, interpolation or state prediction is performed using its historical data sequence or a spatial correlation model with other sensor data to obtain the state estimate for that time point.

[0111] Specifically, the linear regression model for the time dimension in the correlation model is as follows: Let the target reference time point be For a given sensor node, take its position at... Adjacent intervals All aligned data points within ,in For the first The time of each data point on the virtual timeline For the first The measured values ​​of each data point , This represents the number of data points within a neighboring interval. A linear relationship is fitted using the least squares method. , where the slope and intercept The calculation formula is: , , Will Substituting the fitted line, we obtain the state estimate. .

[0112] Adjacent intervals The size is adaptively set according to the sensor sampling frequency: for high sampling frequency sensors (such as 256 points per cycle). Set to 1 to 2 sampling intervals; for low sampling frequency sensors (e.g., once per second). Set it to 1 to 2 times the sampling interval.

[0113] The spatial correlation matrix is ​​constructed as follows: For Each associated sensor collects synchronous data for a period of time under normal system operation, and the correlation coefficient or proportional coefficient between each sensor is calculated to form... Spatial correlation matrix ,in Indicates when the first When data from the first sensor is available, for the second... The estimated scaling factor for the data from the first sensor. When the first... Each sensor at the target reference time point When missing data is encountered, the aligned data from other available sensors at the same time is multiplied by the corresponding scaling factor to obtain the first data. Spatial estimates from individual sensors: ,in The indexes of all available sensors are traversed. The spatial correlation matrix is ​​continuously updated during system operation to track changes in the correlation relationships between sensors.

[0114] When the time and space dimensions are used to obtain estimates independently, the final weighted average is: , in This is a linear regression estimate over the time dimension. This is an estimate of the spatial dimension correlation matrix. and For the corresponding weights, The weights are assigned based on the confidence levels of two estimates: when the number of historical data points... More (e.g.) And goodness of fit Higher (e.g.) )hour, Take a larger value (e.g., 0.7); when the calibration data of the spatial correlation matrix is ​​sufficient and there is a large amount of available correlation sensor data, Take a larger value (e.g., 0.6). By using the weighted average above, information from both time and space dimensions is integrated, improving the accuracy and robustness of state estimation.

[0115] Step S210: Align and package the state estimates of all data streams at the target reference time point to generate a preprocessed dataset.

[0116] When the state estimates of all sensor data streams are extrapolated to the same target reference time point, these state estimates are arranged and combined in a time-aligned manner. This generates a preprocessed dataset with high temporal consistency on a virtual time axis, eliminating the time difference problem caused by the asynchronous original sampling times.

[0117] The preprocessed dataset is based on the target reference time point. The index contains the state estimates of all sensors at that moment and their corresponding confidence information. The data structure of the preprocessed dataset is as follows: Preprocessed dataset = { "target_time": T_target, "sensor_data": [ {"sensor_id": 1, "value": 1, "confidence": c1, "source": "time_space_fusion"}, {"sensor_id": 2, "value": 2, "confidence": c2, "source": "time_interpolation"}, ... ], "data_quality": "high" / "medium" / "low" } Where confidence is the confidence level of each estimate, source indicates the source of the estimate (temporal interpolation, spatial correlation, or spatiotemporal fusion), and data_quality is the overall quality rating of the entire dataset.

[0118] Step S211: Identify and extract multi-sensor data fragments describing the same physical event or state from the preprocessed dataset.

[0119] The system uses a fixed-length sliding window to scan the data streams of each sensor. The window length is set to three to five power frequency cycles based on the sensor sampling frequency. Within each window, the amplitude change rate and root mean square deviation of the data are calculated. When the above characteristic quantities of any sensor data stream exceed a preset detection threshold, a potential event is determined to exist within that window. Subsequently, with that sensor as the center, other sensor data streams are searched along the time axis before and after. Data segments that show synchronous characteristic changes within the same time window are extracted and aggregated to form a multi-sensor data segment set describing the same event.

[0120] The formula for calculating the rate of change of amplitude is as follows: , in For the first in the window The amplitude of each sampling point This represents the total number of sampling points within the window.

[0121] The formula for calculating the root mean square deviation is: , in It is the average value of the amplitude of each sampling point within the window, i.e. .

[0122] The detection threshold is determined based on historical data statistics from each sensor under normal operating conditions, specifically 3 to 5 times the standard deviation of the mean of the characteristic quantity under normal operating conditions. When the amplitude change rate or root mean square deviation of any sensor's data stream exceeds the corresponding detection threshold, a potential event is determined to exist within that window. This dual-indicator joint detection effectively avoids the problem of misjudgment by a single indicator, improving the reliability of event detection.

[0123] The search range for other sensor data streams along the time axis is centered on the event window, extending forward and backward by 2 to 5 window lengths. When other sensor data streams within the search range exhibit amplitude change rates or root mean square deviations exceeding their respective detection thresholds within the same time window, these data segments are included in the data segment set for the same event.

[0124] Step S212: Compare the alignment consistency and physical logic of the multi-sensor data segments on the virtual time axis, and calculate the residual synchronization error of each sensor node.

[0125] The extracted multi-sensor data fragment set is quantitatively evaluated from the following two dimensions: In terms of alignment consistency, the virtual time axis times of the signal mutation start points or peak points in each data fragment are extracted. Using one sensor as a reference, the differences between the characteristic point times of other sensors and the reference time are calculated. The standard deviation of this set of time differences is used as the quantitative index of alignment consistency; the smaller the standard deviation, the higher the alignment consistency. In terms of physical logic rationality, for sensors of the same type, the maximum and average deviations between their measurements of the same physical quantity at the same time are calculated and compared with a preset deviation tolerance threshold. Based on the above evaluation results, the residual synchronization error is calculated using the following formula: the time difference between the characteristic point times of each sensor and the reference time is subtracted from the theoretical time difference caused by differences in event propagation paths; the result is the residual synchronization error of the corresponding sensor node.

[0126] The quantitative indicator for alignment consistency is calculated as follows: Assume there is a total Each sensor produces a characteristic change in the same event, with the first sensor being the one that produces the characteristic change. One sensor serves as a reference ( for (The sensor corresponding to the median time of feature points in each sensor), and the time difference between the feature point times of other sensors and the reference time is... and Then the standard deviation of the time difference values ​​in this group for: , in It is the average of the time differences, i.e. .

[0127] when If the time interval is less than the preset alignment consistency threshold (e.g., 10 microseconds), the alignment consistency is deemed to meet the requirements; otherwise, the alignment consistency is deemed not to meet the requirements, indicating a synchronization error.

[0128] Specifically, for voltage sensor data and current sensor data of three-phase AC power in a power system, the physical logic rationality verification includes the following methods: Method 1: Verification of the three-phase current vector sum based on Kirchhoff's Current Law. This involves checking the three-phase current measurements at the same electrical node. , , The theoretical value of their vector sum is zero (ignoring the tiny leakage current from the line capacitance). Calculate the effective value of the vector sum. ,when Exceeding the preset first deviation threshold If the value is 1% of the rated current, the data segment is deemed to have a physical and logical inconsistency. and The difference is converted into additional residual synchronization error.

[0129] Method 2: Voltage amplitude consistency verification based on three-phase voltage balance constraints. This involves verifying the consistency of three-phase voltage measurements at the same electrical node. , , Under normal operating conditions, the three-phase voltage amplitudes should be basically balanced. Calculate the maximum deviation of the three-phase voltage amplitudes. ,in .when Exceeding the preset second deviation threshold If the voltage is 2% of the rated voltage, the data segment is deemed to have a physical and logical inconsistency. and The difference is converted into additional residual synchronization error.

[0130] Method 3: For the voltage and current data of the same electrical node, check whether its power factor is within a reasonable range (e.g., 0.8 to 1.0). When the power factor exceeds the reasonable range, it is determined that there is a physical logic inconsistency.

[0131] Among them, residual synchronization error The specific calculation method is as follows: , in The theoretical time difference caused by differences in the event propagation path. The time equivalent of the unreasonable deviation in physical logic. The calculation method is as follows: based on the propagation speed of electromagnetic waves on transmission lines in the power system. (Approximately 95% to 99% of the speed of light, i.e., 2.85 × 10⁻⁶) 8 m / s to 2.97×10 8 (m / s) and the distance from the event source to each sensor. ,calculate ,in This serves as a reference point for the distance between the sensor and the event source. When the exact location of the event source cannot be determined... We can approximate it to zero, and at this point the residual synchronization error is the direct difference between the times of the characteristic points.

[0132] Among them, when the physical logic rationality check reveals a severe imbalance in the three-phase current vector sum ( When the current exceeds 5% of the rated current, the system prioritizes determining that the measurement misalignment is caused by synchronization error rather than a three-phase imbalance fault in the power grid itself, in order to avoid misjudging the data synchronization problem as a power system fault.

[0133] Step S213: Based on the long-term sequence of residual synchronization error, identify the corresponding drift mode and change trend to dynamically evaluate the clock drift trend and synchronization error, and obtain the evaluation result.

[0134] The system continuously collects the residual synchronization errors of each sensor node, forming a long-term sequence. Linear regression analysis is performed on this sequence to calculate its slope, which is used as a quantitative indicator of the sensor's clock drift trend. Simultaneously, the standard deviation of the sequence is calculated as a comprehensive evaluation indicator of the synchronization error. These drift trend indicators and synchronization error indicators together constitute the evaluation result, which is output to the subsequent feedback command generation step.

[0135] The specific calculation method for linear regression analysis is as follows: Suppose a sensor node is in continuous The residual synchronization error sequence measured in this event is ,in For the first The moment this event occurred, For the first The residual synchronization error measured in this event , The sequence length. Fitting a straight line. , where the slope The calculation formula is: , slope This is the drift trend indicator, when The time indicates that the clock is continuously ahead. This indicates that the clock is continuously lagging behind. The larger the absolute value, the more severe the drift.

[0136] The synchronization error index is the standard deviation of the residual synchronization error sequence. : , in This represents the average value of the residual synchronization error. The larger the value, the greater the fluctuation in the synchronization error of the node and the worse its stability.

[0137] Among them, when the length of the residual synchronization error sequence When the number of events is less than the preset minimum statistical length (e.g., 10 times), the drift trend indicator will not be output temporarily; only the synchronization error indicator will be output to avoid statistical bias caused by small samples. Once N reaches the minimum statistical length, the evaluation result will be updated every time a new event data point is added.

[0138] Step S214: Generate feedback instructions based on the evaluation results to dynamically adjust the compensation logic and data fusion strategy of the local soft clock.

[0139] The evaluation results consist of drift trend indicators and synchronization error indicators. When a node's drift trend indicator shows a systematic and continuous shift and the standard deviation of its synchronization error exceeds a preset threshold, the system refits the optimal drift parameters based on the node's residual synchronization error long-term sequence, encapsulates the updated parameters as a clock correction command, and sends it to the corresponding node. Upon receiving the command, the node replaces the corresponding parameters in its local compensation logic. When the synchronization error indicators of each node do not exceed the clock correction threshold, but the overall fusion quality does not meet the preset requirements, the system generates a data realignment strategy command and sends it to the data preprocessing module. This command includes at least one of the following adjustments: reducing the fusion time granularity to half of the original value, replacing linear interpolation with cubic spline interpolation, or adjusting the confidence weight allocation strategy of the cross-validation stage. The preprocessing module completes the parameter switching before the next fusion cycle and re-executes alignment and fusion according to the new strategy. The above two types of feedback can be executed in parallel to ensure that synchronization accuracy and fusion quality are continuously maintained within the preset range.

[0140] The triggering conditions and specific details of the clock correction command are as follows: When the drift trend indicator The absolute value exceeds the preset drift threshold. When the error rate is 0.5 microseconds / second, a systematic drift is detected at the node, triggering a clock correction command. The system then refits the straight line using the least squares method based on the long-term residual synchronization error sequence of the node. To obtain new slope parameters and intercept parameter ,Will and The command is encapsulated as a clock correction instruction and sent to the corresponding node. Upon receiving the instruction, the node will convert the frequency compensation value from the local soft clock compensation logic in step S204. Updated to Phase compensation value Corresponding adjustments were made. Through the above updates, the compensation logic can more accurately predict and offset the clock drift of the node, suppressing the accumulation of synchronization errors at the source.

[0141] The triggering conditions and specific details of the data realignment strategy instruction are as follows: When all nodes drift trend indicators The absolute values ​​of none of them exceeded However, there is a synchronization error index for at least one node. Exceeding the preset quality threshold If, after 20 microseconds, the fusion quality of the preprocessed dataset is deemed to have failed to meet preset requirements, a data realignment policy instruction is triggered. This instruction includes at least one of the following adjustments: (1) Reduce the fusion time granularity to half of the original value (e.g., from 2 milliseconds to 1 millisecond) to increase the data density on the time axis; (2) Replace linear interpolation with cubic spline interpolation to fit the data change curve more smoothly; (3) Adjust the confidence weight allocation strategy of the cross-validation stage, reduce the weight of sensors with large drift trend indicators (e.g., reduce the weight from 1.0 to 0.6), and increase the weight of sensors with small synchronization error indicators (e.g., increase from 1.0 to 1.2).

[0142] After receiving the instruction, the preprocessing module completes the parameter switching before the next fusion cycle and re-executes the alignment and fusion steps S207 to S210 according to the new strategy.

[0143] The generation and issuance of the two types of feedback commands employ differentiated cycles: clock correction commands have a longer evaluation cycle (e.g., once per hour) to avoid oscillations caused by frequent adjustments; data realignment strategy commands have a shorter evaluation cycle (e.g., once every 10 fusion cycles) to quickly respond to changes in fusion quality. Through these differentiated cycle settings, the system can achieve a balance between long-term stability and short-term response speed.

[0144] The system also has an upper limit on the frequency of feedback commands. When the same node receives more than the preset number of clock correction commands (e.g., 3 times) within the same time period (e.g., within 1 hour), the system triggers an alarm, prompting maintenance personnel to check the hardware clock status of the node to avoid repeated invalid corrections due to hardware failure.

[0145] Please see Figure 3 A multi-sensor data acquisition and synchronization system includes: The reference establishment module 301 is used to establish a global reference time based on the master clock source when a data synchronization command is received. The soft clock configuration module 302 is used to configure a local soft clock for each sensor node based on the global reference time. The data acquisition module 303 is used to acquire sensor data from each sensor node; The data frame forming module 304 is used to add a local timestamp to the sensor data using a local soft clock, synchronously record the clock offset estimate and data quality identifier, and form a data frame; the data frame carries multi-dimensional time information. The preprocessing dataset generation module 305 is used to map different sensor data to a unified virtual time axis based on multidimensional time information, and generate a time-consistent preprocessing dataset. The correlation analysis module 306 is used to perform correlation analysis on the preprocessed dataset, dynamically evaluate the clock drift trend and synchronization error of each sensor node, and obtain the evaluation results. Feedback instruction generation module 307 is used to generate feedback instructions based on the evaluation results in order to dynamically adjust the compensation logic and data fusion strategy of the local soft clock.

[0146] The soft clock configuration module 302 includes: The sequence measurement submodule is used to measure the relative clock offset sequence and frequency deviation sequence between each sensor node and the master clock source according to a preset period. The prediction submodule is used to smooth and predict the trend of the clock offset sequence and frequency deviation sequence using an adaptive filtering algorithm. The configuration submodule is used to generate dynamic compensation amounts based on the prediction results and apply the compensation amounts to the local soft clock to maintain synchronization with the global reference time.

[0147] The preprocessing dataset generation module 305 includes: The temporary storage submodule is used to configure an independent buffer for each sensor data stream to temporarily store data frames with multi-dimensional time information; The target time point determination submodule is used to determine the target reference time point based on the preset fusion time granularity or event trigger signal; The estimation submodule is used to estimate the state of each sensor data at the target reference time point and mapped onto the virtual time axis based on the local timestamp and clock offset estimate. The preprocessing dataset generation submodule is used to align and package the state estimates of all data streams at the target reference time point to generate a preprocessing dataset.

[0148] The calculation submodule includes: The model building unit is used to build a correlation model of sensor data; the correlation model is used to describe the evolution of the same sensor data over time, as well as the spatial correlation of different sensor data. The estimation unit is used to obtain a state estimate by interpolating or predicting the state at nearby time points when the current sensor node is missing data at the corresponding target reference time point using an association model. The verification unit is used to cross-validate and correct the current sensor state estimate using the association model and aligned sensor data when cross-sensor association is required.

[0149] The correlation models constructed by the model building unit include a linear regression model in the time dimension and a correlation matrix model in the spatial dimension. The linear regression model is used to fit the linear relationship between the measured values ​​and time using historical data frames in the vicinity of the target reference time point. The spatial correlation matrix model is used to pre-calibrate the electrical quantity transfer coefficients between different electrical nodes or between different electrical quantities of the same node based on the network topology of the power system and the impedance parameters of each line. When the data of a certain sensor is missing, spatial estimation is performed using the aligned data of other correlated sensors at the same time.

[0150] When the estimation unit independently derives estimates from the linear regression model in the time dimension and the correlation matrix model in the spatial dimension, it takes the weighted average as the final state estimate. The estimation unit dynamically allocates weights based on the confidence levels of the two estimates. When the number of historical data points for the linear regression model is large and the goodness of fit is high, the weight of the time dimension estimate is increased. When the spatial correlation matrix calibration data is sufficient and there is a large number of available correlation sensor data, the weight of the spatial dimension estimate is increased.

[0151] The correlation analysis module 306 includes: The fragment recognition submodule is used to identify and extract multi-sensor data fragments describing the same physical event or state from the preprocessed dataset; The error calculation submodule is used to compare the alignment consistency and physical logic of multi-sensor data segments on the virtual time axis and calculate the residual synchronization error of each sensor node. The dynamic evaluation submodule is used to identify the corresponding drift patterns and trends based on the long-term sequence of residual synchronization errors, so as to dynamically evaluate the clock drift trend and synchronization error and obtain the evaluation results.

[0152] The error calculation submodule includes a physical logic rationality verification unit, which is used to: perform vector sum verification on the three-phase current measurement values ​​of the same electrical node based on Kirchhoff's current law, calculate the effective value of the vector sum and compare it with a preset first deviation threshold; and perform amplitude consistency verification on the three-phase voltage measurement values ​​of the same electrical node based on three-phase voltage balance constraints, calculate the maximum deviation of the three-phase voltage amplitude and compare it with a preset second deviation threshold; when the deviation of any verification exceeds the corresponding threshold, the deviation is converted into the additional residual synchronization error of the corresponding sensor node.

[0153] Feedback instruction generation module 307 includes: The first feedback submodule is used to generate a clock correction instruction containing the parameters of the deviation prediction model to update the compensation logic if the evaluation result indicates that there is a systematic drift in the local soft clock. The second feedback submodule is used to generate a data realignment strategy if the evaluation result indicates that the fusion quality of the preprocessed dataset does not meet the preset requirements. The strategy is used to indicate whether to change the interpolation algorithm or shorten the fusion time granularity when generating the preprocessed dataset.

[0154] The trigger condition for the first feedback submodule is: the absolute value of the drift trend indicator exceeds the preset drift threshold; the trigger condition for the second feedback submodule is: the drift trend indicators of all nodes do not exceed the drift threshold, but the synchronization error indicator of at least one node exceeds the preset quality threshold.

[0155] Based on the same inventive concept, this application also provides an electronic device. For example... Figure 4 As shown, it includes: one or more processors 101, a memory 102, and one or more I / O interfaces 103. The memory 102 stores one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement a data acquisition synchronization method for any of the multi-channel sensors described in the above embodiments; the one or more I / O interfaces 103 are connected between the processor and the memory and configured to enable information interaction between the processor and the memory.

[0156] The processor 101 is a device with data processing capabilities, including but not limited to a central processing unit (CPU); the memory 102 is a device with data storage capabilities, including but not limited to random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), and flash memory (FLASH); the I / O interface (read / write interface) 103 is connected between the processor 101 and the memory 102, and can realize information interaction between the processor 101 and the memory 102, including but not limited to a data bus (Bus).

[0157] The processor 101, memory 102 and I / O interface 103 are interconnected via bus 104, and thus connected to other components of the computing device.

[0158] The one or more processors 101 include field-programmable gate arrays.

[0159] This application also provides a computer storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the steps of a method for synchronizing data acquisition from multiple sensors.

[0160] This application also provides a computer program product on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of a method for synchronizing data acquisition from multiple sensors.

[0161] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. A method for synchronizing data acquisition from multiple sensors, characterized in that, Includes the following steps: When a data synchronization command is received, a global reference time is established based on the master clock source; Configure a local soft clock for each sensor node based on the global reference time; Collect sensor data from each sensor node; A local timestamp is added to the sensor data using a local soft clock, and the estimated clock offset and data quality identifier are recorded synchronously to form a data frame. Data frames carry multi-dimensional time information; Based on multidimensional time information, different sensor data are mapped to a unified virtual time axis to generate a time-consistent preprocessed dataset; Correlation analysis is performed on the preprocessed dataset to dynamically evaluate the clock drift trend and synchronization error of each sensor node and obtain the evaluation results; Feedback instructions are generated based on the evaluation results to dynamically adjust the compensation logic of the local soft clock and the data fusion strategy.

2. The data acquisition and synchronization method for multiple sensors according to claim 1, characterized in that, The association analysis of the preprocessed dataset includes: Identify and extract multi-sensor data fragments describing the same physical event from a preprocessed dataset; Compare the alignment consistency of the data segments from various multi-sensor systems on the virtual time axis, and verify the rationality of their physical logic based on the physical laws of the power system; Physical logic validity checks include at least one of the following: Based on Kirchhoff's current law, the vector sum of the three-phase current measurements at the same electrical node is verified: the effective value of the vector sum is calculated. ,when Exceeding the preset first deviation threshold When this occurs, it is determined that there is a physical logic inconsistency, and the deviation is converted into an additional residual synchronization error; Based on three-phase voltage balance constraints, amplitude consistency verification is performed on the three-phase voltage measurements at the same electrical node: the maximum deviation of the three-phase voltage amplitude is calculated. ,in ,when Exceeding the preset second deviation threshold If the physical logic is unreasonable, the deviation is converted into an additional residual synchronization error.

3. The data acquisition and synchronization method for multiple sensors according to claim 1, characterized in that, The process of mapping different sensor data to a unified virtual timeline based on multidimensional time information to generate a time-consistent preprocessed dataset includes: Configure an independent buffer for each sensor data stream to temporarily store data frames carrying multi-dimensional time information; Determine the target reference time point based on the preset fusion time granularity or event trigger signal; Based on the local timestamp and clock offset estimate, the state estimate of each sensor data point at the target reference time point is calculated and mapped onto the virtual time axis. Align and package the state estimates of all data streams at the target reference time point to generate a preprocessed dataset.

4. The data acquisition and synchronization method for multiple sensors according to claim 3, characterized in that, The calculation of the state estimate of each sensor data point at the target reference time point includes: Construct a correlation model for sensor data; the correlation model includes an evolution model in the time dimension and a correlation model in the spatial dimension; The evolutionary model in the time dimension is: taking the target reference time point. Adjacent intervals Aligned data points within The least squares method is used to fit the linear relationship. ,in: ， ， Will Substituting the linear relationship, we obtain the time dimension estimate. ; The spatial correlation model is based on a pre-calibrated spatial correlation matrix derived from the power system's network topology and line impedance parameters. When the first sensor is at the target reference time point When missing data is encountered, the spatial dimension estimate is obtained by multiplying the aligned data from other available sensors at the same time by the corresponding electrical quantity transfer coefficient. ; When the time dimension estimates and spatial dimension estimates are obtained independently, the weighted average is taken as the final state estimate. ,in This is a linear regression estimate over the time dimension. This is an estimate of the spatial dimension correlation matrix. and For the corresponding weights, .

5. The data acquisition and synchronization method for multiple sensors according to claim 4, characterized in that, Weight and Dynamically allocate based on the confidence levels of the two estimates: When the number of historical data points And goodness of fit hour, The value ranges from 0.6 to 0.8; When the amount of calibration data for the spatial correlation matrix is ​​sufficient and the number of available sensors participating in spatial estimation is ≥3, The value ranges from 0.5 to 0.

7.

6. The data acquisition and synchronization method for multiple sensors according to claim 2, characterized in that, The identification and extraction of multi-sensor data fragments describing the same physical event includes: A sliding window is used to scan the preprocessed dataset, with a window length of three to five power frequency cycles; Calculate the rate of change of the magnitude of the data within the window. and root mean square deviation ; When the amplitude change rate or root mean square deviation of any sensor data stream exceeds the preset detection threshold, it is determined that there is a potential event within the window. Then, with the sensor as the center, the search for other data streams is expanded by two to five window lengths before and after along the time axis. Data segments that show synchronous characteristic changes within the same time window are extracted and collected to form a multi-sensor data segment set describing the same event.

7. The data acquisition and synchronization method for multiple sensors according to claim 2, characterized in that, The comparison of the alignment consistency of each multi-sensor data segment on the virtual time axis includes: Extract the moments of signal abrupt change start points or peak points on the virtual time axis from each data segment; Using the sensor corresponding to the median time as a reference, calculate the time difference between the feature point times of other sensors and the reference time. ; Calculate the standard deviation of the time difference As a quantitative indicator of alignment consistency; The residual synchronization error is calculated using the following formula: ,in The theoretical time difference caused by differences in the event propagation path. The time equivalent of the unreasonable deviation in physical logic.

8. The data acquisition and synchronization method for multiple sensors according to claim 1, characterized in that, The step of generating feedback instructions based on evaluation results to dynamically adjust the compensation logic and data fusion strategy of the local soft clock includes: The evaluation results include a drift trend index and a synchronization error index; the drift trend index is the slope of the linear regression of the long-term residual synchronization error sequence. The synchronization error index is the standard deviation of the long-term residual synchronization error sequence. ; when Exceeding the preset drift threshold When a systemic drift is detected in the local soft clock, a clock correction command containing updated compensation parameters is generated. After receiving the command, the node replaces the corresponding parameters in the local compensation logic. When all nodes None exceeded But there is at least one node Exceeding the preset quality threshold If the fusion quality fails to meet the preset requirements, a data realignment strategy instruction is generated, which includes at least one of the following: reducing the fusion time granularity to half of the original value, replacing linear interpolation with cubic spline interpolation, or adjusting the confidence weight allocation strategy of the cross-validation step.

9. A data acquisition and synchronization system for multiple sensors, characterized in that, The method for synchronizing data acquisition from multiple sensors as described in any one of claims 1-8 includes: The reference establishment module is used to establish a global reference time based on the master clock source when a data synchronization command is received. The soft clock configuration module is used to configure a local soft clock for each sensor node based on the global reference time. The data acquisition module is used to collect sensor data from each sensor node; The data frame forming module is used to add a local timestamp to the sensor data using a local soft clock, synchronously record the estimated clock offset and data quality identifier, and form a data frame; the data frame carries multi-dimensional time information. The preprocessing dataset generation module is used to map different sensor data to a unified virtual time axis based on multidimensional time information, and generate a time-consistent preprocessing dataset. The correlation analysis module is used to perform correlation analysis on the preprocessed dataset, dynamically evaluate the clock drift trend and synchronization error of each sensor node, and obtain the evaluation results. The feedback instruction generation module is used to generate feedback instructions based on the evaluation results, so as to dynamically adjust the compensation logic and data fusion strategy of the local soft clock.

10. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions, which, when executed by the processor, run a data acquisition synchronization method for multiple sensors as described in any one of claims 1-8.

Images