Power distribution area multi-parameter acquisition method and system based on intelligent fusion terminal

Abstract

The application discloses a power distribution area multi-parameter acquisition method and system based on an intelligent fusion terminal, relates to the technical field of electric power information processing, and comprises the following steps: collecting multi-source operation data of a power distribution area, performing cleaning and abnormal marking, generating a multi-parameter acquisition data set, analyzing parameter fluctuation degree, equipment state and business importance, generating a dynamic acquisition data sequence, constructing a unified reference time axis, identifying clock deviation and communication time delay, generating a unified time sequence feature sequence, inputting the unified time sequence feature sequence into an improved ModernTCN model, generating a synchronous compensation time sequence, calculating equipment synchronization consistency, constructing an equipment synchronization relationship matrix, generating a trusted synchronization domain, performing abnormal isolation and trusted data replacement according to the trusted synchronization domain, and feeding back updated acquisition parameters. Through dynamic collaborative acquisition and trusted data fusion of the intelligent fusion terminal, the application realizes accurate acquisition and reliable transmission of multi-parameter data of the power distribution area.

Description

Technical Field

[0001] This invention relates to the field of power information processing technology, and in particular to a method and system for acquiring multiple parameters of a distribution substation based on an intelligent fusion terminal. Background Technology

[0002] With the continuous advancement of new power systems and smart distribution networks, distribution substations have gradually deployed smart converged terminals, smart meters, distribution automation terminals, distributed energy monitoring devices, and various status sensing equipment. This enables real-time collection and centralized management of multi-parameter data, including voltage, current, power, power quality, equipment status, and environmental information. As a crucial data aggregation node at the edge of the distribution substation, the smart converged terminal not only undertakes the tasks of collecting, converting, processing, and uploading multi-source operational data, but also facilitates information exchange between the master station and field equipment. Its operational efficiency directly impacts the operation monitoring, fault analysis, load management, and equipment maintenance of the distribution substation.

[0003] Existing intelligent fusion terminals typically collect and transmit data from various acquisition devices according to fixed sampling and upload cycles. Differences in data acquisition time, communication latency, and device clocks between different devices make it difficult to maintain a consistent time correspondence between data from multiple acquisition devices. When communication links fluctuate, device load changes, or some devices experience sampling delays, data from different sources is prone to time misalignment, data loss, and asynchronous sampling. Most existing technologies use simple time alignment or interpolation compensation methods to handle asynchronous data, lacking effective differentiation between device clock deviations and communication latency, making it difficult to accurately recover the true timing correlation. Furthermore, the lack of comprehensive utilization of cross-device timing correlation information affects the consistency and reliability of the fused data.

[0004] Existing intelligent converged terminals mostly employ a unified upload strategy, lacking a mechanism for categorizing uploads and dynamically controlling caching based on data reliability, business importance, and device operating status. When network congestion, communication anomalies, or changes in feedback from the master station occur, terminals typically cannot synchronously adjust their sampling strategies, upload strategies, and time synchronization parameters. This leads to continuous backlog of cached data, duplicate uploads, or reduced timeliness of critical data uploads, making it difficult to form a closed-loop processing flow of collection, synchronization, fusion, upload, and feedback updates.

[0005] Therefore, how to provide a method and system for multi-parameter acquisition in distribution radio areas based on intelligent fusion terminals is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a multi-parameter acquisition method and system for distribution radio stations based on intelligent fusion terminals. This invention fully utilizes intelligent fusion terminals, multi-source data fusion, time synchronization processing, and improved ModernTCN timing analysis technology to perform dynamic acquisition scheduling, unified time synchronization, cross-device timing compensation, reliable data fusion, and classified upload control on multi-source operating data of distribution radio stations. This forms a closed-loop processing flow of acquisition, synchronization, fusion, upload, and feedback updates, and has the advantages of high data timing consistency, strong terminal fusion reliability, high acquisition efficiency, and high upload reliability.

[0007] The multi-parameter acquisition method for distribution radio areas based on intelligent fusion terminals according to embodiments of the present invention includes:

[0008] Collect multi-source operating data from various acquisition devices within the distribution radio area, perform preprocessing on the multi-source operating data, and generate a multi-parameter acquisition data set;

[0009] Based on the statistical analysis of the parameter fluctuation characteristics, equipment operating status and business importance of each acquisition device through multi-parameter acquisition data set, the sampling period and upload period of each acquisition device are dynamically adjusted to generate dynamic acquisition data sequence;

[0010] A unified reference timeline is constructed based on the sampling timestamps, data arrival times, and sampling periods of each acquisition device in the dynamic data sequence. The device clock deviation and communication delay are separated through a time consistency constraint mechanism. Asynchronous sampling segments are identified and time resampling and missing mapping are performed to generate a unified time series feature sequence.

[0011] The unified temporal feature sequence is input into the improved ModernTCN model. A continuous temporal segment is constructed through the temporal misalignment awareness and recombination module. The reliable time propagation module is used to limit the cross-layer diffusion of misaligned features. The temporal correlation between different acquisition devices is established through the cross-device collaborative convolution module, and a synchronously compensated temporal sequence is output.

[0012] Based on the synchronization compensation time sequence, the synchronization consistency between each acquisition device is calculated, a trusted synchronization domain is constructed, and cross-device data reconstruction, anomaly isolation and trusted data replacement processing are performed according to the trusted synchronization domain to generate a terminal fusion data set.

[0013] Based on the terminal fusion data set, perform classified uploading and local caching control, write the upload results and main station feedback information back to the intelligent fusion terminal, update the dynamic acquisition strategy and time synchronization parameters, and generate the updated multi-parameter acquisition configuration.

[0014] Optionally, the multi-source operating data includes voltage data, current data, active power data, reactive power data, power factor data, frequency data, electrical energy data, harmonic data, equipment status data, communication status data, sampling timestamps, data arrival times, and sampling period data.

[0015] Optionally, the preprocessing of multi-source operational data to generate a multi-parameter acquisition data set includes performing data format unification, outlier removal, and missing value marking on the multi-source operational data to generate a multi-parameter acquisition data set.

[0016] Optionally, generating the dynamically acquired data sequence includes:

[0017] The multi-parameter acquisition data set is grouped according to the device identifier and parameter type, and the parameter values ​​of each group are arranged according to the sampling timestamp to generate the parameter time sequence segment corresponding to each acquisition device.

[0018] For each parameter time segment, the change amplitude, continuous change direction, fluctuation duration and fluctuation number per unit time of adjacent parameter values ​​are calculated sequentially to generate parameter fluctuation characteristics corresponding to each acquisition device.

[0019] Read the device status data and communication status data corresponding to each acquisition device, identify the device operation, device alarm, communication interruption and data backlog status, and generate the device operation status corresponding to each acquisition device.

[0020] The importance of the business is determined based on the monitoring purpose of each parameter type, the correlation of abnormal alarms, and the frequency of main station calls. The data collection and adjustment basis for each data collection device is generated by combining the parameter fluctuation characteristics and the equipment operating status.

[0021] Based on the data collection and adjustment criteria, the sampling period and upload period of each data acquisition device are adjusted within the preset sampling period and upload period range. Parameter data are collected according to the adjusted sampling period and arranged according to the adjusted upload period to generate a dynamic data acquisition sequence.

[0022] Optionally, generating a unified temporal feature sequence includes:

[0023] Read the device identifier, parameter type, sampling timestamp, data arrival time and sampling period corresponding to each parameter data in the dynamically acquired data sequence, and construct the parameter time sequence according to the device identifier and parameter type respectively;

[0024] A unified reference time axis is established based on the sampling period corresponding to the time sequence of each parameter, and the data of each parameter is mapped to the unified reference time axis according to the sampling timestamp to obtain the reference time position corresponding to each parameter data.

[0025] A time consistency constraint mechanism is established, which includes continuous sampling constraint, periodic consistency constraint and arrival delay constraint. The continuous sampling constraint is used to detect whether adjacent sampling timestamps maintain a continuous sampling relationship, the periodic consistency constraint is used to detect whether adjacent sampling time intervals are consistent with the corresponding sampling period, and the arrival delay constraint is used to detect whether the time difference between the data arrival time and the sampling timestamp remains continuous and stable.

[0026] Based on continuous sampling constraints, continuously offset sampling time segments are identified; based on period consistency constraints, sampling period drift segments are identified; based on arrival delay constraints, communication delay fluctuation segments are identified. Continuously offset sampling time segments are determined as device clock deviations, and communication delay fluctuation segments are determined as communication delays. The sampling timestamp is then corrected based on the device clock deviation.

[0027] The corrected sampling timestamps are remapped to a unified reference time axis, asynchronous sampling segments that exceed the corresponding sampling period between adjacent parameter data are identified, time resampling is performed on the asynchronous sampling segments, and missing markers are set at the unmapped reference time positions to generate a unified time series feature sequence.

[0028] Optionally, the output synchronization compensation timing sequence includes:

[0029] The improved ModernTCN model includes an input encoder, a time misalignment-aware reconstruction module, a reliable time propagation module, a ModernTCN time series extraction module, a cross-device collaborative convolution module, and a synchronous compensation output layer.

[0030] The input encoder receives and encodes a unified temporal feature sequence. The input encoder includes a parameter feature encoding branch and a time state encoding branch, which respectively extract parameter value features, missing position markers, sampling time offsets, and communication delay features, and perform feature concatenation to generate initial temporal features.

[0031] The time misalignment-aware reconstruction module includes a misalignment detection branch and a segment reconstruction branch. The misalignment detection branch performs one-dimensional convolution on the sampling time offset, communication delay, and missing position marker of adjacent reference times to output the time misalignment position. The segment reconstruction branch divides continuous time segments with the time misalignment position as the boundary and performs overlapping splicing on adjacent continuous time segments to generate time reconstruction features.

[0032] The ModernTCN temporal extraction module includes multiple sequentially connected ModernTCN blocks. Each ModernTCN block includes a deep convolutional layer with a large kernel, a channel expansion layer, an activation layer, a channel compression layer, and a residual connection layer. It performs multi-scale temporal convolution on the temporal reconstruction features to generate multi-level temporal features.

[0033] The trusted time propagation module is set between adjacent ModernTCN blocks and includes a trustworthiness extraction branch, a propagation mask generation branch, and a gated fusion branch. The trustworthiness extraction branch performs convolutional coding on the missing position marker, sampling time offset, and communication delay. The propagation mask generation branch outputs the propagation weights for each reference time. The gated fusion branch performs weighted fusion on the temporal features of adjacent levels according to the propagation weights to generate trusted propagation temporal features.

[0034] The cross-device collaborative convolution module includes a time-alignment branch, a device association branch, and a collaborative convolution branch. The time-alignment branch extracts reliable propagation time-series features of each acquisition device at the same reference time. The device association branch generates a device association matrix based on the direction and magnitude of parameter changes. The collaborative convolution branch performs cross-device feature convolution based on the device association matrix to generate device collaborative features.

[0035] The synchronous compensation output layer performs residual fusion of device collaboration features and trusted propagation timing features, and restores the timing arrangement according to device identifier, parameter type and unified reference time axis to output synchronous compensation timing sequence.

[0036] Optionally, the generation of the terminal fusion data set includes:

[0037] Extract the parameter values ​​of each acquisition device in the synchronization compensation time sequence according to the parameter type and unified reference time axis, calculate the consistency of parameter change direction, the degree of similarity of change magnitude and the proportion of effective data overlap between different acquisition devices at the same reference time, and generate the device synchronization consistency degree.

[0038] A device synchronization relationship matrix is ​​constructed based on the degree of device synchronization consistency. Each element in the device synchronization relationship matrix represents the synchronization association state of two corresponding acquisition devices at the corresponding reference time. The device synchronization relationship matrix is ​​continuously searched according to a unified reference time axis. Matrix regions that meet the preset synchronization consistency threshold at multiple consecutive reference times and whose corresponding devices maintain a synchronized association state are connected and merged to generate a reliable synchronization domain.

[0039] Read the synchronization compensation time sequence of the corresponding device in the trusted synchronization domain. Based on the device association and time association recorded in the trusted synchronization domain, determine the candidate device set corresponding to each missing position. Sort the parameter values ​​of the candidate device set corresponding to the reference time in descending order of device synchronization consistency. Select the parameter value with the highest synchronization consistency to perform cross-device data reconstruction and generate reconstruction parameter data.

[0040] The reconstructed parameter data is compared with the corresponding parameter values ​​of other acquisition devices at the same reference time within the trusted synchronization domain. When the reconstructed parameter data continuously deviates from the change trend of devices within the trusted synchronization domain, the corresponding parameter data is marked as abnormal data, and an abnormal isolation mark is generated.

[0041] The device synchronization consistency is recalculated based on the data in the trusted synchronization domain that does not have an abnormal isolation flag. The parameter data with the highest synchronization consistency is taken as trusted data. Trusted data replacement is performed on missing positions, time misalignment positions and abnormal isolation positions. The device identifier, parameter type, reference time, parameter value, synchronization flag, reconstruction flag and abnormal isolation flag are associated and encapsulated to generate a terminal fusion data set.

[0042] Optionally, generating the updated multi-parameter acquisition configuration includes:

[0043] Read the device identifier, parameter type, reference time, parameter value, reconstruction flag and abnormal isolation flag corresponding to each parameter data in the terminal fusion data set, and divide the real-time uploaded data, periodic uploaded data and local cached data according to parameter type, business importance, reconstruction flag and abnormal isolation flag, and generate classified upload queue and local cache queue;

[0044] Real-time uploads and periodic uploads are performed according to the classified upload queues, and the corresponding parameter data is written to the smart fusion terminal cache according to the local cache queue. The upload status, cache status, upload time and cache location of each parameter data are recorded, and upload result information is generated.

[0045] The system receives data reception status, data integrity verification results, data retransmission requests, and time synchronization information returned by the master station. It then associates and matches the uploaded result information with the feedback information from the master station to generate terminal feedback results. Based on the terminal feedback results, it updates the sampling period, upload period, and cache retention strategy corresponding to each acquisition device to generate a dynamic acquisition strategy.

[0046] Based on the time synchronization information in the feedback from the main station, the local reference time axis of the intelligent fusion terminal and the time synchronization parameters corresponding to each acquisition device are corrected. The updated dynamic acquisition strategy, time synchronization parameters and cache control strategy are then associated and encapsulated to generate the updated multi-parameter acquisition configuration.

[0047] According to an embodiment of the present invention, a distribution radio area multi-parameter acquisition system based on an intelligent fusion terminal includes:

[0048] The data acquisition and preprocessing module is used to collect multi-source operating data from various acquisition devices in the distribution area and perform preprocessing to generate a multi-parameter acquisition data set.

[0049] The dynamic acquisition and scheduling module is used to analyze parameter fluctuation characteristics, equipment operating status and business importance based on multi-parameter acquisition data sets, dynamically adjust the sampling period and upload period, and generate dynamic acquisition data sequences.

[0050] The time synchronization processing module is used to construct a unified reference time axis based on dynamically acquired data sequences, separate device clock deviation from communication delay, perform time resampling and missing mapping processing, and generate a unified time series feature sequence.

[0051] The temporal synchronization compensation module is used to input a unified temporal feature sequence into the improved ModernTCN model, perform temporal misalignment-aware recombination, reliable time propagation and cross-device collaborative convolution processing, and generate a synchronization-compensated temporal sequence.

[0052] The terminal fusion processing module is used to construct a trusted synchronization domain based on the synchronization compensation time sequence, perform cross-device data reconstruction, anomaly isolation and trusted data replacement processing, and generate a terminal fusion data set.

[0053] The upload configuration update module is used to perform classified upload and local cache control based on the terminal fusion data set, and generate updated multi-parameter acquisition configuration by combining the dynamic acquisition strategy and time synchronization parameters updated by the main station feedback.

[0054] The beneficial effects of this invention are:

[0055] This invention performs dynamic acquisition scheduling on multi-source operating data from various acquisition devices within a distribution radio area. It adaptively adjusts the sampling and uploading cycles based on parameter fluctuation characteristics, device operating status, and business importance. Furthermore, it combines a unified reference time axis to perform time synchronization processing on asynchronous sampled data, enabling data generated by different acquisition devices to be correlated under a unified time reference. This provides a consistent data foundation for subsequent time series analysis and data fusion.

[0056] This invention improves the ModernTCN model by performing time misalignment-aware reconstruction, reliable time propagation, and cross-device collaborative convolution processing on unified temporal feature sequences. It constructs a reliable synchronization domain, reconstructs cross-device data, isolates anomalies, and replaces reliable data. This effectively maintains the temporal correlation between different acquisition devices, improves the consistency, integrity, and reliability of terminal fusion data, and reduces data fusion errors caused by device clock deviations, communication delays, and data loss.

[0057] This invention performs classified uploading and local caching control based on the terminal fusion data set, and combines the main station feedback to dynamically update the acquisition strategy and time synchronization parameters, forming a closed-loop processing mechanism of acquisition, synchronization, fusion, uploading and feedback updates. This improves the intelligent fusion terminal's ability to collaboratively process multi-parameter data of distribution radio stations, and enhances the reliability of data uploading, the stability of terminal operation, and the real-time management capability of distribution radio station operation data in complex operating environments. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is a flowchart of the multi-parameter acquisition method for distribution radio areas based on intelligent fusion terminals proposed in this invention;

[0060] Figure 2 This is a schematic diagram of the structure of the distribution radio area multi-parameter acquisition system based on an intelligent fusion terminal proposed in this invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0062] refer to Figure 1 A method for acquiring multiple parameters of a distribution radio area based on an intelligent fusion terminal includes:

[0063] Collect multi-source operating data from various acquisition devices within the distribution radio area, perform preprocessing on the multi-source operating data, and generate a multi-parameter acquisition data set;

[0064] Based on the statistical analysis of the parameter fluctuation characteristics, equipment operating status and business importance of each acquisition device through multi-parameter acquisition data set, the sampling period and upload period of each acquisition device are dynamically adjusted to generate dynamic acquisition data sequence;

[0065] A unified reference timeline is constructed based on the sampling timestamps, data arrival times, and sampling periods of each acquisition device in the dynamic data sequence. The device clock deviation and communication delay are separated through a time consistency constraint mechanism. Asynchronous sampling segments are identified and time resampling and missing mapping are performed to generate a unified time series feature sequence.

[0066] The unified temporal feature sequence is input into the improved ModernTCN model. A continuous temporal segment is constructed through the temporal misalignment awareness and recombination module. The reliable time propagation module is used to limit the cross-layer diffusion of misaligned features. The temporal correlation between different acquisition devices is established through the cross-device collaborative convolution module, and a synchronously compensated temporal sequence is output.

[0067] Based on the synchronization compensation time sequence, the synchronization consistency between each acquisition device is calculated, a trusted synchronization domain is constructed, and cross-device data reconstruction, anomaly isolation and trusted data replacement processing are performed according to the trusted synchronization domain to generate a terminal fusion data set.

[0068] Based on the terminal fusion data set, perform classified uploading and local caching control, write the upload results and main station feedback information back to the intelligent fusion terminal, update the dynamic acquisition strategy and time synchronization parameters, and generate the updated multi-parameter acquisition configuration.

[0069] In this embodiment, the multi-source operating data includes voltage data, current data, active power data, reactive power data, power factor data, frequency data, electrical energy data, harmonic data, equipment status data, communication status data, sampling timestamp, data arrival time, and sampling period data.

[0070] In this embodiment, the preprocessing of multi-source operational data to generate a multi-parameter acquisition data set includes performing data format unification, outlier removal, and missing value marking on the multi-source operational data to generate a multi-parameter acquisition data set.

[0071] In this embodiment, generating the dynamically acquired data sequence includes:

[0072] The multi-parameter acquisition data set is grouped according to the device identifier and parameter type, and the parameter values ​​of each group are arranged according to the sampling timestamp to generate the parameter time sequence segment corresponding to each acquisition device.

[0073] For each parameter time series segment, the change amplitude, continuous change direction, fluctuation duration, and number of fluctuations per unit time of adjacent parameter values ​​are calculated sequentially to generate parameter fluctuation characteristics corresponding to each acquisition device. Specifically, the calculation of the change amplitude, continuous change direction, fluctuation duration, and number of fluctuations per unit time of adjacent parameter values ​​for each parameter time series segment is as follows:

[0074] Traverse the parameter time series segments in ascending order of sampling time, record the absolute value of the difference between the current sample value and the previous sample value as a change amplitude, and record the change amplitude sequence in sequence.

[0075] Compare the current sample value with the previous sample value. If the current value is greater than the previous value, it is recorded as a positive change. If it is less than the previous value, it is recorded as a negative change. If they are equal, the direction remains unchanged. Record the direction sequence in sequence.

[0076] Find continuous segments in the direction sequence that maintain the same direction, count the number of sampling points for each continuous segment and multiply by the sampling period to obtain the duration of continuous change for that segment;

[0077] Find continuous non-zero segments in the change amplitude sequence, treat each continuous non-zero segment as a fluctuation, count the duration of the segment, and accumulate the number of fluctuations within a fixed time window;

[0078] The average and variance of the change amplitude sequence are calculated, and the duration of continuous change, the duration of fluctuation, and the number of fluctuations per unit time are summarized. Combined with the device identifier and parameter type, the parameter fluctuation characteristics of the corresponding acquisition device are generated.

[0079] Read the device status data and communication status data corresponding to each acquisition device, identify the device operation, device alarm, communication interruption and data backlog status, and generate the device operation status corresponding to each acquisition device.

[0080] The importance of the business is determined based on the monitoring purpose of each parameter type, the correlation of abnormal alarms, and the frequency of main station calls. The data collection and adjustment basis for each data collection device is generated by combining the parameter fluctuation characteristics and the equipment operating status.

[0081] Based on the data collection and adjustment criteria, the sampling period and upload period of each data acquisition device are adjusted within the preset sampling period and upload period range. Parameter data are collected according to the adjusted sampling period and arranged according to the adjusted upload period to generate a dynamic data acquisition sequence.

[0082] In this embodiment, generating a unified temporal feature sequence includes:

[0083] Read the device identifier, parameter type, sampling timestamp, data arrival time and sampling period corresponding to each parameter data in the dynamically acquired data sequence, and construct the parameter time sequence according to the device identifier and parameter type respectively;

[0084] A unified reference time axis is established based on the sampling period corresponding to the time sequence of each parameter, and the data of each parameter is mapped to the unified reference time axis according to the sampling timestamp to obtain the reference time position corresponding to each parameter data.

[0085] A time consistency constraint mechanism is established, which includes continuous sampling constraint, periodic consistency constraint and arrival delay constraint. The continuous sampling constraint is used to detect whether adjacent sampling timestamps maintain a continuous sampling relationship, the periodic consistency constraint is used to detect whether adjacent sampling time intervals are consistent with the corresponding sampling period, and the arrival delay constraint is used to detect whether the time difference between the data arrival time and the sampling timestamp remains continuous and stable.

[0086] Based on continuous sampling constraints, continuously offset sampling time segments are identified; based on period consistency constraints, sampling period drift segments are identified; and based on arrival delay constraints, communication delay fluctuation segments are identified. Continuously offset sampling time segments are determined as device clock skew, and communication delay fluctuation segments are determined as communication delay. The sampling timestamp is then corrected based on the device clock skew.

[0087] Identifying continuous offset sampling time segments based on continuous sampling constraints, specifically:

[0088] The sampling timestamps of each sampling point are read sequentially according to the unified reference time axis, subtracted from the sampling timestamp of the previous sampling point and the corresponding sampling period, to obtain the sampling time offset of the sampling point.

[0089] Determine the sign of the sampling time offset. If the offset is always positive and the difference between the offsets of adjacent sampling points is less than 10% of the sampling period, it is marked as a positive continuous offset segment. If the offset is always negative and the difference between the offsets of adjacent sampling points is less than 10% of the sampling period, it is marked as a negative continuous offset segment.

[0090] The positive and negative continuous offset segments are merged into a continuous offset sampling time segment, and the start time, end time and number of continuous sampling points of the segment in the unified reference time axis are recorded as the basis for determining the device clock deviation.

[0091] Identifying sampling period drift segments based on period consistency constraints, specifically:

[0092] The difference in sampling timestamps between adjacent sampling points is calculated sequentially according to a unified reference time axis, and the sampling interval deviation is obtained by comparing it with the corresponding sampling period; a positive threshold is set to 5% of the corresponding sampling period, and a negative threshold is set to -5% of the corresponding sampling period.

[0093] Continuous detection of sampling interval deviation: When the interval deviation of three or more consecutive sampling points is greater than the positive threshold or less than the negative threshold, the continuous segment is marked as a candidate segment for sampling period drift.

[0094] All candidate segments are merged. If the interval between adjacent candidate segments is less than 20% of the corresponding sampling period, they are merged into drift segments of the same sampling period. The start time, end time and number of continuous sampling points of the segment in the unified reference time axis are recorded as the result of the sampling period drift determination.

[0095] Identifying communication delay fluctuation segments based on arrival delay constraints, specifically:

[0096] Calculate the arrival delay for each sampling point to obtain the arrival delay sequence; use the average arrival delay of the most recent 30 sampling points as the baseline delay, and set the forward delay threshold to 115% of the baseline delay and the reverse delay threshold to 85% of the baseline delay;

[0097] The arrival delay sequence is continuously detected. When the arrival delay of three or more consecutive sampling points is higher than the forward delay threshold or lower than the reverse delay threshold, the continuous segment is marked as a candidate segment for communication delay fluctuation.

[0098] All candidate segments of communication delay fluctuation are merged. If the interval between adjacent candidate segments is less than 20% of the corresponding sampling period, they are merged into the same communication delay fluctuation segment. The start time, end time and number of continuous sampling points of the segment in the unified reference time axis are recorded as the result of communication delay determination.

[0099] The corrected sampling timestamps are remapped to a unified reference time axis, asynchronous sampling segments that exceed the corresponding sampling period between adjacent parameter data are identified, time resampling is performed on the asynchronous sampling segments, and missing markers are set at the unmapped reference time positions to generate a unified time series feature sequence.

[0100] In this embodiment, the output synchronization compensation timing sequence includes:

[0101] The improved ModernTCN model includes an input encoder, a time misalignment-aware reconstruction module, a reliable time propagation module, a ModernTCN time series extraction module, a cross-device collaborative convolution module, and a synchronous compensation output layer.

[0102] The improved ModernTCN model is constructed as follows:

[0103] The improved ModernTCN model is built upon the traditional ModernTCN model. It retains the core network structure used for temporal feature extraction, replaces the original input layer with an input encoder, and performs unified dimensional mapping and feature encoding on unified temporal feature sequences from multiple devices. A time misalignment-aware reconstruction module is added between the input encoder and the ModernTCN temporal extraction module to reconstruct misaligned time sequences based on device clock deviations and communication delays. A reliable time propagation module is then added in series to propagate and modify the reconstructed temporal features according to temporal continuity and reliability. The corrected temporal features are then input into the ModernTCN temporal extraction module. A cross-device collaborative convolution module is added to the output of the ModernTCN temporal extraction module to perform cross-device associated convolution and fusion calculations on the temporal features output by different acquisition devices to enhance the temporal correlation information between devices. Finally, the fused features are input into the synchronization compensation output layer to output the synchronization compensation temporal sequence, forming an improved ModernTCN model consisting of an input encoder, a temporal misalignment-aware reconstruction module, a reliable time propagation module, a ModernTCN temporal extraction module, a cross-device collaborative convolution module, and a synchronization compensation output layer.

[0104] The input encoder receives and encodes a unified temporal feature sequence. The input encoder includes a parameter feature encoding branch and a time state encoding branch, which respectively extract parameter value features, missing position markers, sampling time offsets, and communication delay features, and perform feature concatenation to generate initial temporal features.

[0105] The temporal misalignment-aware reconstruction module includes a misalignment detection branch and a segment reconstruction branch. The misalignment detection branch performs a one-dimensional convolution on the sampling time offset, communication delay, and missing position markers of adjacent reference times to output the temporal misalignment position. The segment reconstruction branch divides continuous temporal segments using the temporal misalignment position as the boundary and performs overlapping splicing on adjacent continuous temporal segments to generate temporal reconstruction features, wherein:

[0106] Output the time misalignment position, specifically:

[0107] The sampling time offset, communication delay, and missing position marker are arranged in chronological order to form a multi-channel input sequence. The input misalignment detection branch is subjected to one-dimensional convolution processing to extract local time change features.

[0108] The local time-varying features are further processed by multi-layer one-dimensional convolution and continuous feature fusion to obtain the time misalignment response value corresponding to each sampling position;

[0109] The time misalignment response value is compared with the preset misalignment judgment threshold. When the time misalignment response value is greater than 0.65, the corresponding sampling position is marked as the time misalignment position. The continuously marked sampling positions are merged to form a time misalignment segment, and the start and end positions of each time misalignment segment are output as the time misalignment position.

[0110] The generation time recombination features are as follows:

[0111] Using the time misalignment position as the boundary, the unified time series feature sequence is divided into multiple continuous time series segments, and the original time order of each continuous time series segment is preserved.

[0112] Adjacent consecutive time segments are overlapped and spliced ​​according to their time sequence. The last 5 consecutive sampling points of the previous consecutive time segment and the first 5 consecutive sampling points of the next consecutive time segment are used as the overlapping area. The feature values ​​of the corresponding positions in the overlapping area are weighted and fused to generate consecutive spliced ​​segments.

[0113] All consecutive spliced ​​segments are rearranged according to a unified reference timeline to obtain temporally continuous and feature-complete time reconstruction features.

[0114] The ModernTCN temporal extraction module includes multiple sequentially connected ModernTCN blocks. Each ModernTCN block includes a deep convolutional layer with a large kernel, a channel expansion layer, an activation layer, a channel compression layer, and a residual connection layer. It performs multi-scale temporal convolution on the temporal reconstruction features to generate multi-level temporal features.

[0115] The trusted time propagation module is located between adjacent ModernTCN blocks and includes a trustworthiness extraction branch, a propagation mask generation branch, and a gated fusion branch. The trustworthiness extraction branch performs convolutional coding on the missing position markers, sampling time offsets, and communication delays. The propagation mask generation branch outputs the propagation weights for each reference time. The gated fusion branch performs weighted fusion of the temporal features of adjacent levels based on the propagation weights to generate trusted propagation temporal features, wherein:

[0116] The propagation mask generation branch outputs the propagation weights at each reference time, specifically:

[0117] Read the credibility features output by the credibility extraction branch, perform one-dimensional convolution and local feature aggregation on the credibility features, and obtain the credibility response value corresponding to each reference time.

[0118] The credibility response values ​​at each reference time are normalized, and the normalized credibility response values ​​are mapped to a range of 0 to 1 to generate the initial propagation weights corresponding to each reference time. Specifically, when the credibility response value is greater than 0.8, the corresponding propagation weight is set to 1; when the credibility response value is less than 0.2, the corresponding propagation weight is set to 0; and the remaining credibility response values ​​are mapped to propagation weights between 0 and 1 according to their corresponding magnitudes.

[0119] The initial propagation weights of adjacent reference times are subjected to continuous smoothing. When the difference between the propagation weights of two adjacent reference times is greater than 0.3, the current propagation weight is replaced by the average of the propagation weights of five adjacent reference times to generate a continuous propagation weight sequence.

[0120] The gated fusion branch performs weighted fusion of temporal features from adjacent levels based on propagation weights, specifically as follows:

[0121] Read the preceding time sequence features output by the previous ModernTCN block, the current level time sequence features output by the next ModernTCN block, and the propagation weights corresponding to each reference time, and complete the position correspondence according to the unified reference time axis;

[0122] The propagation weights at each reference time are multiplied by the feature values ​​at the corresponding positions of the previous time series features to obtain the reliable propagation features; the difference between 1 and the propagation weights is used as the current level retention weights, and multiplied by the feature values ​​at the corresponding positions of the current level time series features to obtain the level retention features;

[0123] The trusted propagation features and hierarchical retention features corresponding to the same reference time are added position by position to obtain the initial fused temporal features; where the larger the propagation weight, the higher the retention ratio of the previous level temporal features, and the smaller the propagation weight, the higher the retention ratio of the current level temporal features.

[0124] One-dimensional convolution, feature normalization, and non-linear activation are performed on the initial fused temporal features, and residual superposition is performed with the current level temporal features to generate reliable propagation temporal features;

[0125] The cross-device collaborative convolution module includes a time-alignment branch, a device association branch, and a collaborative convolution branch. The time-alignment branch extracts reliable propagation time-series features of each acquisition device at the same reference time. The device association branch generates a device association matrix based on the direction and magnitude of parameter changes. The collaborative convolution branch performs cross-device feature convolution based on the device association matrix to generate device collaborative features, wherein:

[0126] The equipment association branch generates an equipment association matrix based on the direction and magnitude of parameter changes, specifically:

[0127] According to the unified reference time axis, read the reliable propagation time sequence characteristics of each acquisition device at the current reference time and the previous reference time, calculate the change of each parameter between adjacent reference times, mark the direction of parameter change according to the positive or negative value of the change, and record the parameter change amplitude according to the absolute value of the change.

[0128] Compare the same type of parameters of any two acquisition devices one by one. When the parameters change in the same direction, record the consistent direction indicator. When the parameters change in opposite directions, record the opposite direction indicator. Calculate the difference between the magnitude of the corresponding parameter changes of the two acquisition devices.

[0129] Divide the difference in parameter variation by the larger of the two variation values ​​to obtain the degree of variation; when both variation values ​​are 0, the degree of variation is set to 0; the difference between 1 and the degree of variation is taken as the degree of similarity.

[0130] When the parameters of two acquisition devices change in the same direction and the similarity of their amplitudes is not less than 0.7, the corresponding parameters are determined to be strongly correlated parameters; when the parameters change in the same direction and the similarity of their amplitudes is less than 0.7 but not less than 0.4, the corresponding parameters are determined to be generally correlated parameters; the remaining corresponding parameters are determined to be weakly correlated parameters.

[0131] The number of strongly correlated parameters, generally correlated parameters, and weakly correlated parameters between any two acquisition devices are counted. The results are then weighted and summed according to 1, 0.6, and 0.2, respectively. The weighted result is divided by the total number of parameters involved in the comparison to obtain the initial correlation strength between the two acquisition devices.

[0132] The initial correlation strength is mapped to between 0 and 1, and the correlation strength between each acquisition device and itself is set to 1. The correlation strength between all devices is written into the corresponding positions of the matrix according to the fixed arrangement order of the acquisition devices to generate the device correlation matrix.

[0133] The collaborative convolution branch performs cross-device feature convolution based on the device association matrix, specifically as follows:

[0134] According to the fixed arrangement order of the acquisition devices, the reliable propagation time sequence features of each acquisition device outputting the same-time aligned branch are arranged into a device feature set, and the correlation strength between corresponding devices in the device correlation matrix is ​​read.

[0135] For any target acquisition device, select associated acquisition devices with an association strength of not less than 0.40, and determine the reliable propagation time series features corresponding to the target acquisition device and associated acquisition devices as cross-device convolution inputs;

[0136] The trusted propagation time-series features corresponding to each associated acquisition device are multiplied by their association strength to obtain the association weighted features. All association weighted features are accumulated position by position and divided by the sum of the association strengths involved in the accumulation to generate the association aggregate features of the target acquisition device.

[0137] The reliable propagation time-series features of the target acquisition device itself and the associated aggregation features are concatenated according to the feature channels. Cross-device one-dimensional convolution, feature normalization and non-linear activation processing are performed on the concatenation result to extract the collaborative change features between the target acquisition device and the associated acquisition devices.

[0138] The cooperative change characteristics are superimposed with the reliable propagation time sequence characteristics of the target acquisition device itself to generate the initial device cooperative characteristics corresponding to the target acquisition device.

[0139] All acquisition devices are sequentially processed with association filtering, feature weighting, cross-device convolution and residual superposition. The initial device collaborative features of each acquisition device are then summarized according to the fixed device order to generate device collaborative features.

[0140] The synchronous compensation output layer performs residual fusion of device collaboration features and trusted propagation timing features, and restores the timing arrangement according to device identifier, parameter type and unified reference time axis to output synchronous compensation timing sequence.

[0141] To address the timing misalignment issue caused by clock skew, communication delay, and data loss in multiple data acquisition devices, this invention introduces a timing misalignment-aware reconstruction module at the input of the traditional ModernTCN model. This module identifies timing misalignment locations through a misalignment detection branch and uses a segment reconstruction branch to divide and overlap continuous time segments based on these misalignment locations. This restores the temporal continuity of the misaligned data before it is input into the ModernTCN network for feature extraction. This reduces the interference of timing misalignment on the correlation of temporal features and improves the ModernTCN model's ability to model the temporal sequence of asynchronously sampled data and its feature extraction accuracy.

[0142] To address the issue that traditional ModernTCN does not consider data reliability during feature propagation between layers, this invention adds a reliable time propagation module between adjacent ModernTCN blocks. By extracting the missing location markers of the branch fusion based on reliability, sampling time offset, and communication delay information, propagation weights are generated. Then, the gated fusion branch controls the propagation ratio of temporal features of adjacent layers based on the propagation weights, ensuring that reliable data is continuously propagated while low-reliability data is suppressed. This effectively reduces the accumulation of abnormal sampling and communication jitter layer by layer in the network, improving the consistency and stability of deep temporal features.

[0143] To address the issue that traditional ModernTCN models only single-device time series and struggles to leverage inter-device operational relationships, this invention adds a cross-device collaborative convolution module to the ModernTCN output. This module constructs a device association matrix based on the direction and magnitude of parameter changes through device association branches, and then uses the collaborative convolution branches to perform cross-device convolutional fusion on the temporal features of associated devices. This allows the model to simultaneously learn the temporal variation patterns of each device and the collaborative variation relationships between devices, enhancing the ability to express joint features across multiple devices and improving the temporal consistency, association integrity, and synchronization compensation accuracy of the fused terminal data.

[0144] In this embodiment, the generation of the terminal fusion data set includes:

[0145] Parameter values ​​of each acquisition device in the synchronization compensation time series are extracted according to parameter type and unified reference time axis. The consistency of parameter change direction, the similarity of change magnitude, and the proportion of effective data overlap among different acquisition devices at the same reference time are calculated to generate the device synchronization consistency level. Specifically, the calculation of the consistency of parameter change direction, the similarity of change magnitude, and the proportion of effective data overlap among different acquisition devices at the same reference time is as follows:

[0146] Select any two acquisition devices according to parameter type, match the parameter values ​​of the two acquisition devices at the same reference time in a unified reference time axis, and read the valid position mark corresponding to each parameter value;

[0147] For reference times where both acquisition devices have valid parameter values, calculate the difference between the parameter value at the current reference time and the parameter value at the previous valid reference time. Based on whether the difference is positive, negative, or zero, record the direction of change of the corresponding parameter, whether it increases, decreases, or remains unchanged.

[0148] Compare the parameter change direction of the two acquisition devices one by one at the same reference time. When both are increasing, both are decreasing or remain unchanged, record a consistent direction. When the two change directions are different, record a inconsistent direction. Divide the number of consistent directions by the total number of valid reference times participating in the direction comparison to obtain the consistency of parameter change direction.

[0149] The parameter change amplitudes of the two acquisition devices at the same reference time are taken respectively. The absolute value of the difference between the two change amplitudes is divided by the larger value of the two change amplitudes to obtain the amplitude difference degree. The difference between 1 and the amplitude difference degree is taken as the amplitude closeness at the reference time. When both change amplitudes are 0, the amplitude closeness is set to 1.

[0150] The amplitude similarity of all valid reference times involved in the comparison is averaged to obtain the amplitude similarity between the two acquisition devices. When the amplitude similarity is not less than 0.8, it is marked as highly similar. When the amplitude similarity is less than 0.8 but not less than 0.5, it is marked as generally similar. The rest are marked as amplitude deviation.

[0151] The number of reference moments in the unified reference time axis where both acquisition devices have valid parameter values ​​is counted, and this number is divided by the smaller of the number of reference moments with valid parameter values ​​from the two acquisition devices to obtain the effective data overlap ratio.

[0152] The consistency of parameter change direction, the similarity of change magnitude, and the overlap ratio of effective data were weighted and summarized according to weights of 0.4, 0.35, and 0.25 respectively, to obtain the degree of device synchronization consistency between the two acquisition devices under the corresponding parameter type.

[0153] A device synchronization relationship matrix is ​​constructed based on the degree of device synchronization consistency. Each element in the device synchronization relationship matrix represents the synchronization association state of two corresponding acquisition devices at a corresponding reference time. The device synchronization relationship matrix is ​​continuously searched according to a unified reference time axis. Matrix regions where multiple consecutive reference times meet a preset synchronization consistency threshold and the corresponding devices maintain a synchronized association state are connected and merged to generate a trusted synchronization domain. Specifically, the generation of the trusted synchronization domain involves:

[0154] According to the fixed arrangement order of the acquisition devices, the synchronization consistency of the devices at each reference time is written into the device synchronization relationship matrix. Matrix elements with a synchronization consistency of not less than 0.75 are marked as synchronized associations, and matrix elements with a synchronization consistency of less than 0.75 are marked as asynchronous associations.

[0155] Search the device synchronization relationship matrix sequentially along the unified reference time axis. For any two acquisition devices, count the number of reference times when the synchronization association state is continuously maintained. When the synchronization association state is continuously maintained for 5 or more reference times, the corresponding matrix area is determined as a reliable synchronization candidate area.

[0156] Record the acquisition device, parameter type, start reference time, end reference time and continuous holding time corresponding to each trusted synchronization candidate region, and calculate the average value of the synchronization consistency of devices in the candidate region to generate the candidate region confidence value;

[0157] Connectivity determination is performed on adjacent trusted synchronization candidate regions. When two candidate regions contain the same acquisition device and the time interval does not exceed 2 reference times, and the difference in the confidence value of the candidate regions is not greater than 0.15, the two candidate regions are merged into the same connected region.

[0158] Repeat the connectivity check and region merging until there are no adjacent candidate regions that meet the merging conditions, and remove isolated matrix regions with fewer than 5 consecutive reference times or a region confidence value of less than 0.75.

[0159] Based on a unified reference timeline, the device range, parameter type, time range, area confidence value, and synchronization association status of each connected area are summarized to generate a trusted synchronization domain.

[0160] Read the synchronization compensation time sequence of the corresponding device in the trusted synchronization domain. Based on the device association and time association recorded in the trusted synchronization domain, determine the candidate device set corresponding to each missing position. Sort the parameter values ​​of the candidate device set corresponding to the reference time in descending order of device synchronization consistency. Select the parameter value with the highest synchronization consistency to perform cross-device data reconstruction and generate reconstruction parameter data.

[0161] The reconstructed parameter data is compared with the corresponding parameter values ​​of other acquisition devices at the same reference time within the trusted synchronization domain. When the reconstructed parameter data continuously deviates from the change trend of devices within the trusted synchronization domain, the corresponding parameter data is marked as abnormal data, and an abnormal isolation mark is generated.

[0162] The device synchronization consistency is recalculated based on the data in the trusted synchronization domain that does not have an abnormal isolation flag. The parameter data with the highest synchronization consistency is taken as trusted data. Trusted data replacement is performed on missing positions, time misalignment positions and abnormal isolation positions. The device identifier, parameter type, reference time, parameter value, synchronization flag, reconstruction flag and abnormal isolation flag are associated and encapsulated to generate a terminal fusion data set.

[0163] In this embodiment, generating the updated multi-parameter acquisition configuration includes:

[0164] Read the device identifier, parameter type, reference time, parameter value, reconstruction flag and abnormal isolation flag corresponding to each parameter data in the terminal fusion data set, and divide the real-time uploaded data, periodic uploaded data and local cached data according to parameter type, business importance, reconstruction flag and abnormal isolation flag, and generate classified upload queue and local cache queue;

[0165] Real-time uploads and periodic uploads are performed according to the classified upload queues, and the corresponding parameter data is written to the smart fusion terminal cache according to the local cache queue. The upload status, cache status, upload time and cache location of each parameter data are recorded, and upload result information is generated.

[0166] The system receives data reception status, data integrity verification results, data retransmission requests, and time synchronization information returned by the master station. It then associates and matches the uploaded result information with the feedback information from the master station to generate terminal feedback results. Based on the terminal feedback results, it updates the sampling period, upload period, and cache retention strategy corresponding to each acquisition device to generate a dynamic acquisition strategy.

[0167] Based on the time synchronization information in the feedback from the main station, the local reference time axis of the intelligent fusion terminal and the time synchronization parameters corresponding to each acquisition device are corrected. The updated dynamic acquisition strategy, time synchronization parameters and cache control strategy are then associated and encapsulated to generate the updated multi-parameter acquisition configuration.

[0168] refer to Figure 2 A multi-parameter acquisition system for distribution radio areas based on intelligent fusion terminals includes:

[0169] The data acquisition and preprocessing module is used to collect multi-source operating data from various acquisition devices in the distribution area and perform preprocessing to generate a multi-parameter acquisition data set.

[0170] The dynamic acquisition and scheduling module is used to analyze parameter fluctuation characteristics, equipment operating status and business importance based on multi-parameter acquisition data sets, dynamically adjust the sampling period and upload period, and generate dynamic acquisition data sequences.

[0171] The time synchronization processing module is used to construct a unified reference time axis based on dynamically acquired data sequences, separate device clock deviation from communication delay, perform time resampling and missing mapping processing, and generate a unified time series feature sequence.

[0172] The temporal synchronization compensation module is used to input a unified temporal feature sequence into the improved ModernTCN model, perform temporal misalignment-aware recombination, reliable time propagation and cross-device collaborative convolution processing, and generate a synchronization-compensated temporal sequence.

[0173] The terminal fusion processing module is used to construct a trusted synchronization domain based on the synchronization compensation time sequence, perform cross-device data reconstruction, anomaly isolation and trusted data replacement processing, and generate a terminal fusion data set.

[0174] The upload configuration update module is used to perform classified upload and local cache control based on the terminal fusion data set, and generate updated multi-parameter acquisition configuration by combining the dynamic acquisition strategy and time synchronization parameters updated by the main station feedback.

[0175] Example 1: During a continuous data acquisition cycle in a distribution substation, the intelligent fusion terminal receives multi-parameter data from the substation's main meter, 6 branch metering devices, 42 user-side metering devices, and 2 distribution transformer status acquisition devices. The raw data contains a total of 518,400 records, including voltage, current, active power, reactive power, power factor, electrical energy, equipment status, sampling timestamps, and data arrival times. The initial sampling period for the main substation meter and branch metering devices is 5 seconds, the initial sampling period for user-side metering devices is 15 seconds, and the equipment status sampling period is 30 seconds. During the simulated operation, some user-side loads were started up simultaneously, causing the current on two branch lines to rise from 46.3A to 132.8A and the active power to rise from 10.7kW to 30.4kW within 180 seconds. Simultaneously, clock drift occurred on 3 acquisition devices, with a maximum sampling time offset of 2.86 seconds. The communication link jittered during the load surge phase, with a maximum arrival delay of 1850 milliseconds, and 7 consecutive missing sampling points occurred.

[0176] After the data enters the intelligent fusion terminal, the terminal first cleans the original records. The system removes 1264 data entries with a voltage of 0, duplicate timestamps, and missing device numbers. Abrupt data with voltages exceeding 260V or currents exceeding 1.4 times the rated range are marked as abnormal instead of being directly deleted. After processing, there are 517,136 valid data entries, increasing the valid data ratio from 99.42% to 99.76%. The terminal then rearranges the data according to device number, parameter type, and sampling timestamp. For example, branch metering device A should generate 60 sampling points in a 300-second segment, but actually receives 58 sampling points. The 21st and 22nd sampling points are missing, and the arrival time of the 35th sampling point is delayed by 1260 milliseconds compared to the normal link. This segment is marked as an asynchronous sampling segment.

[0177] During dynamic data acquisition and scheduling, the system reads the continuous change amplitude, direction of change, and number of fluctuations per unit time for each type of parameter. In the stable operation segment, the average voltage change amplitude on the user side is 0.34V, the average current change amplitude is 0.18A, and the number of fluctuations per unit time is 2 times / minute. The system adjusts the sampling period for user-side conventional power data from 15 seconds to 30 seconds, and the upload period from 30 seconds to 60 seconds. In the load surge segment, the branch current continuously rises for 12 sampling points, with a maximum single-point change amplitude of 8.7A and a number of fluctuations per unit time reaching 14 times / minute. The system adjusts the sampling period for this branch current, active power, and voltage from 5 seconds to 1 second, and the upload period from 30 seconds to 5 seconds. After the adjustments, the number of effective sampling points during the load surge process increases from 36 using the traditional method to 172, and the sampling interval near the current peak is shortened from 5 seconds to 1 second, enabling continuous recording of peak change processes.

[0178] During the time synchronization processing phase, the system constructs a unified reference time axis with a reference interval set to 1 second. When detecting the sampling segment of branch metering device B, the deviations of adjacent sampling intervals relative to the 5-second sampling period are 5.3%, 5.7%, 6.2%, 6.6%, and 6.8%, respectively, continuously exceeding the 5% positive threshold. The system identifies this segment as a sampling period drift segment. Simultaneously, the average arrival delay of the device's most recent 30 sampling points is 118 milliseconds. The arrival delays of the subsequent five consecutive sampling points are 862 milliseconds, 1146 milliseconds, 1398 milliseconds, 1284 milliseconds, and 946 milliseconds, all exceeding 115% of the reference arrival delay. The system identifies this segment as a communication delay fluctuation segment. After separation processing, the device clock deviation is corrected to 2.31 seconds, and the communication delay fluctuation is written separately into the delay marker, no longer mixed into the sampling timestamp correction process. Traditional methods locate the current peak value 2.8 seconds after the actual reference time within the same segment; the peak value positioning error corrected by this invention is 0.21 seconds.

[0179] After generating a unified temporal feature sequence, the system inputs the sampling time offset, communication delay, missing position markers, and parameter values ​​into the improved ModernTCN model. The input encoder maps the multi-parameter records of each reference time into a 64-dimensional feature vector, forming a 600×64 input feature from 600 consecutive seconds of segments. In the temporal misalignment-aware reconstruction module, the misalignment detection branch performs a one-dimensional convolution on the sampling time offset, communication delay, and missing position markers, outputting 9 temporal misalignment positions in the segment. Seven of these positions are consistent with manually labeled misalignment positions, and two are communication delay boundary positions. The segment reconstruction branch divides the segment into 10 consecutive temporal segments based on the misalignment positions, and then splices them together using the last 5 sampling points of the previous segment and the first 5 sampling points of the next segment as the overlapping area. Before splicing, the branch power curve showed a sudden jump from 11.6kW to 27.9kW around the 214th second; after splicing, the power values ​​in this region were 18.4kW, 20.1kW, 22.6kW, 25.3kW, and 27.4kW respectively, and the change process returned to a continuous increase.

[0180] In the reliable time propagation module, the system generates propagation weights based on missing location markers, sampling time offsets, and communication delays. The propagation weights for normal reference times are mainly distributed between 0.86 and 0.97, while those for locations with communication delays exceeding 1000 milliseconds and missing features decrease to 0.12 to 0.23. Taking the consecutive missing segments from 318 seconds to 324 seconds as an example, the traditional ModernTCN model also causes diffusion offsets to the normal features on both sides of the missing points, resulting in current reconstruction errors of 3.8A and 4.2A at 317 seconds and 325 seconds, respectively. This invention suppresses the propagation of low-reliability features through propagation weights, reducing the errors at the normal points on both sides to 0.9A and 1.1A, respectively. After four ModernTCN timing extraction blocks, the current recovery values ​​of the missing segments in the synchronous compensation timing sequence output by the model are 96.4A, 101.8A, 107.3A, 113.6A, 119.2A, 124.1A and 128.5A, with an average error of 1.34A compared with the simulation baseline value.

[0181] The cross-device collaborative convolution module further reads the reliable propagation timing characteristics of each acquisition device at the same reference time. The device association branch compares the direction and magnitude of parameter changes of different devices. In a 180-second load increase segment, the current change direction consistency between the main meter of the distribution area and branch metering device A is 0.96, the similarity of change magnitude is 0.88, and the effective data overlap ratio is 0.97, resulting in a calculated device synchronization consistency of 0.94. However, the synchronization consistency with branch metering device B, which exhibits clock drift, is only 0.48. Based on this, the system generates a device association matrix, marking device pairs with a synchronization consistency of not less than 0.75 as synchronously associated, and continuously searches the unified reference time axis. Ultimately, three reliable synchronization domains are formed. The first reliable synchronization domain includes the main meter of the distribution area, 3 branch metering devices, and 18 user-side metering devices, lasting for 1260 seconds, with a regional reliability value of 0.91. Devices exhibiting clock drift are excluded from the reliable synchronization domain during abnormal duration intervals.

[0182] In the trusted data replacement phase, the system reconstructs the missing parameters of the isolated equipment. Taking the missing active power segment of branch metering equipment B as an example, the traditional linear interpolation results are 19.2kW, 20.1kW, 21.0kW, 21.9kW, 22.8kW, 23.7kW, and 24.6kW, while the simulation baseline values ​​are 19.8kW, 21.6kW, 23.4kW, 25.8kW, 27.6kW, 29.1kW, and 30.0kW. The average error of the traditional method is 3.17kW. This invention utilizes the cooperative characteristics of associated equipment within the trusted synchronization domain for replacement, and the restored results are 20.0kW, 21.4kW, 23.1kW, 25.2kW, 27.1kW, 28.7kW, and 29.5kW, with the average error reduced to 0.43kW.

[0183] After completing the terminal fusion data set, the system performs categorized uploading according to business importance. Load surge fragments, abnormal equipment status, and time deviation correction records are prioritized for uploading, while stable operating power statistics are written to the local cache. In a 72-hour simulation, the traditional fixed upload method uploaded a total of 518,400 data entries, while the present invention uploaded only 371,280 data entries, a reduction of 28.4%. The traditional method experienced 11 buffer overflows due to link congestion, while the present invention only experienced one, and no load surge fragments were lost.

[0184] In the comparative experiment, the traditional method uses fixed sampling, original timestamp sorting, and linear interpolation, while the present invention uses the complete process described above. Both methods used the same 14,400 training samples, 2,400 validation samples, and 2,400 test samples. Test results show that the accuracy rate for time misalignment identification is 76.8% for the traditional method and 96.1% for the present invention; the average time alignment error is 486 milliseconds for the traditional method and 72 milliseconds for the present invention; the effective data integrity rate is 92.3% for the traditional method and 98.7% for the present invention; the average error for current loss recovery is 4.84A for the traditional method and 1.34A for the present invention; the average error for active power loss recovery is 3.17kW for the traditional method and 0.43kW for the present invention; the average cross-device synchronization consistency rate is 0.71 for the traditional method and 0.92 for the present invention; the accuracy rate for abnormal device isolation is 79.8% for the traditional method and 95.1% for the present invention; and the complete retention rate of important load change segments is 86.4% for the traditional method and 98.1% for the present invention.

[0185] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for multi-parameter acquisition in distribution radio areas based on intelligent fusion terminals, characterized in that, include: Collect multi-source operating data from various acquisition devices within the distribution radio area, perform preprocessing on the multi-source operating data, and generate a multi-parameter acquisition data set; Based on the statistical analysis of the parameter fluctuation characteristics, equipment operating status and business importance of each acquisition device through multi-parameter acquisition data set, the sampling period and upload period of each acquisition device are dynamically adjusted to generate dynamic acquisition data sequence; A unified reference timeline is constructed based on the sampling timestamps, data arrival times, and sampling periods of each acquisition device in the dynamic data sequence. The device clock deviation and communication delay are separated through a time consistency constraint mechanism. Asynchronous sampling segments are identified and time resampling and missing mapping are performed to generate a unified time series feature sequence. The unified temporal feature sequence is input into the improved ModernTCN model. A continuous temporal segment is constructed through the temporal misalignment awareness and recombination module. The reliable time propagation module is used to limit the cross-layer diffusion of misaligned features. The temporal correlation between different acquisition devices is established through the cross-device collaborative convolution module, and a synchronously compensated temporal sequence is output. Based on the synchronization compensation time sequence, the synchronization consistency between each acquisition device is calculated, a trusted synchronization domain is constructed, and cross-device data reconstruction, anomaly isolation and trusted data replacement processing are performed according to the trusted synchronization domain to generate a terminal fusion data set. Based on the terminal fusion data set, perform classified uploading and local caching control, write the upload results and main station feedback information back to the intelligent fusion terminal, update the dynamic acquisition strategy and time synchronization parameters, and generate the updated multi-parameter acquisition configuration.

2. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The multi-source operating data includes voltage data, current data, active power data, reactive power data, power factor data, frequency data, electrical energy data, harmonic data, equipment status data, communication status data, sampling timestamps, data arrival times, and sampling period data.

3. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The preprocessing of multi-source operational data to generate a multi-parameter acquisition data set includes performing data format unification, outlier removal, and missing value marking on the multi-source operational data to generate a multi-parameter acquisition data set.

4. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The generation of dynamically acquired data sequences includes: The multi-parameter acquisition data set is grouped according to the device identifier and parameter type, and the parameter values ​​of each group are arranged according to the sampling timestamp to generate the parameter time sequence segment corresponding to each acquisition device. For each parameter time segment, the change amplitude, continuous change direction, fluctuation duration and fluctuation number per unit time of adjacent parameter values ​​are calculated sequentially to generate parameter fluctuation characteristics corresponding to each acquisition device. Read the device status data and communication status data corresponding to each acquisition device, identify the device operation, device alarm, communication interruption and data backlog status, and generate the device operation status corresponding to each acquisition device. The importance of the business is determined based on the monitoring purpose of each parameter type, the correlation of abnormal alarms, and the frequency of main station calls. The data collection and adjustment basis for each data collection device is generated by combining the parameter fluctuation characteristics and the equipment operating status. Based on the data collection and adjustment criteria, the sampling period and upload period of each data acquisition device are adjusted within the preset sampling period and upload period range. Parameter data are collected according to the adjusted sampling period and arranged according to the adjusted upload period to generate a dynamic data acquisition sequence.

5. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The generation of a unified temporal feature sequence includes: Read the device identifier, parameter type, sampling timestamp, data arrival time and sampling period corresponding to each parameter data in the dynamically acquired data sequence, and construct the parameter time sequence according to the device identifier and parameter type respectively; A unified reference time axis is established based on the sampling period corresponding to the time sequence of each parameter, and the data of each parameter is mapped to the unified reference time axis according to the sampling timestamp to obtain the reference time position corresponding to each parameter data. A time consistency constraint mechanism is established, which includes continuous sampling constraint, periodic consistency constraint and arrival delay constraint. The continuous sampling constraint is used to detect whether adjacent sampling timestamps maintain a continuous sampling relationship, the periodic consistency constraint is used to detect whether adjacent sampling time intervals are consistent with the corresponding sampling period, and the arrival delay constraint is used to detect whether the time difference between the data arrival time and the sampling timestamp remains continuous and stable. Based on continuous sampling constraints, continuously offset sampling time segments are identified; based on period consistency constraints, sampling period drift segments are identified; based on arrival delay constraints, communication delay fluctuation segments are identified. Continuously offset sampling time segments are determined as device clock deviations, and communication delay fluctuation segments are determined as communication delays. The sampling timestamp is then corrected based on the device clock deviation. The corrected sampling timestamps are remapped to a unified reference time axis, asynchronous sampling segments that exceed the corresponding sampling period between adjacent parameter data are identified, time resampling is performed on the asynchronous sampling segments, and missing markers are set at the unmapped reference time positions to generate a unified time series feature sequence.

6. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The output synchronization compensation timing sequence includes: The improved ModernTCN model includes an input encoder, a time misalignment-aware reconstruction module, a reliable time propagation module, a ModernTCN time series extraction module, a cross-device collaborative convolution module, and a synchronous compensation output layer. The input encoder receives and encodes a unified temporal feature sequence. The input encoder includes a parameter feature encoding branch and a time state encoding branch, which respectively extract parameter value features, missing position markers, sampling time offsets, and communication delay features, and perform feature concatenation to generate initial temporal features. The time misalignment-aware reconstruction module includes a misalignment detection branch and a segment reconstruction branch. The misalignment detection branch performs one-dimensional convolution on the sampling time offset, communication delay, and missing position marker of adjacent reference times to output the time misalignment position. The segment reconstruction branch divides continuous time segments with the time misalignment position as the boundary and performs overlapping splicing on adjacent continuous time segments to generate time reconstruction features. The ModernTCN temporal extraction module includes multiple sequentially connected ModernTCN blocks. Each ModernTCN block includes a deep convolutional layer with a large kernel, a channel expansion layer, an activation layer, a channel compression layer, and a residual connection layer. It performs multi-scale temporal convolution on the temporal reconstruction features to generate multi-level temporal features. The trusted time propagation module is set between adjacent ModernTCN blocks and includes a trustworthiness extraction branch, a propagation mask generation branch, and a gated fusion branch. The trustworthiness extraction branch performs convolutional coding on the missing position marker, sampling time offset, and communication delay. The propagation mask generation branch outputs the propagation weights for each reference time. The gated fusion branch performs weighted fusion on the temporal features of adjacent levels according to the propagation weights to generate trusted propagation temporal features. The cross-device collaborative convolution module includes a time-alignment branch, a device association branch, and a collaborative convolution branch. The time-alignment branch extracts reliable propagation time-series features of each acquisition device at the same reference time. The device association branch generates a device association matrix based on the direction and magnitude of parameter changes. The collaborative convolution branch performs cross-device feature convolution based on the device association matrix to generate device collaborative features. The synchronous compensation output layer performs residual fusion of device collaboration features and trusted propagation timing features, and restores the timing arrangement according to device identifier, parameter type and unified reference time axis to output synchronous compensation timing sequence.

7. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The generated terminal fusion data set includes: Extract the parameter values ​​of each acquisition device in the synchronization compensation time sequence according to the parameter type and unified reference time axis, calculate the consistency of parameter change direction, the degree of similarity of change magnitude and the proportion of effective data overlap between different acquisition devices at the same reference time, and generate the device synchronization consistency degree. A device synchronization relationship matrix is ​​constructed based on the degree of device synchronization consistency. Each element in the device synchronization relationship matrix represents the synchronization association state of two corresponding acquisition devices at the corresponding reference time. The device synchronization relationship matrix is ​​continuously searched according to a unified reference time axis. Matrix regions that meet the preset synchronization consistency threshold at multiple consecutive reference times and whose corresponding devices maintain a synchronized association state are connected and merged to generate a reliable synchronization domain. Read the synchronization compensation time sequence of the corresponding device in the trusted synchronization domain. Based on the device association and time association recorded in the trusted synchronization domain, determine the candidate device set corresponding to each missing position. Sort the parameter values ​​of the candidate device set corresponding to the reference time in descending order of device synchronization consistency. Select the parameter value with the highest synchronization consistency to perform cross-device data reconstruction and generate reconstruction parameter data. The reconstructed parameter data is compared with the corresponding parameter values ​​of other acquisition devices at the same reference time within the trusted synchronization domain. When the reconstructed parameter data continuously deviates from the change trend of devices within the trusted synchronization domain, the corresponding parameter data is marked as abnormal data, and an abnormal isolation mark is generated. The device synchronization consistency is recalculated based on the data in the trusted synchronization domain that does not have an abnormal isolation flag. The parameter data with the highest synchronization consistency is taken as trusted data. Trusted data replacement is performed on missing positions, time misalignment positions and abnormal isolation positions. The device identifier, parameter type, reference time, parameter value, synchronization flag, reconstruction flag and abnormal isolation flag are associated and encapsulated to generate a terminal fusion data set.

8. The method for multi-parameter acquisition of distribution radio areas based on intelligent fusion terminals according to claim 1, characterized in that, The generation of the updated multi-parameter acquisition configuration includes: Read the device identifier, parameter type, reference time, parameter value, reconstruction flag and abnormal isolation flag corresponding to each parameter data in the terminal fusion data set, and divide the real-time uploaded data, periodic uploaded data and local cached data according to parameter type, business importance, reconstruction flag and abnormal isolation flag, and generate classified upload queue and local cache queue; Real-time uploads and periodic uploads are performed according to the classified upload queues, and the corresponding parameter data is written to the smart fusion terminal cache according to the local cache queue. The upload status, cache status, upload time and cache location of each parameter data are recorded, and upload result information is generated. The system receives data reception status, data integrity verification results, data retransmission requests, and time synchronization information returned by the master station. It then associates and matches the uploaded result information with the feedback information from the master station to generate terminal feedback results. Based on the terminal feedback results, it updates the sampling period, upload period, and cache retention strategy corresponding to each acquisition device to generate a dynamic acquisition strategy. Based on the time synchronization information in the feedback from the main station, the local reference time axis of the intelligent fusion terminal and the time synchronization parameters corresponding to each acquisition device are corrected. The updated dynamic acquisition strategy, time synchronization parameters and cache control strategy are then associated and encapsulated to generate the updated multi-parameter acquisition configuration.

9. A distribution radio area multi-parameter acquisition system based on an intelligent fusion terminal, comprising the distribution radio area multi-parameter acquisition method based on an intelligent fusion terminal as described in any one of claims 1 to 8, characterized in that, include: The data acquisition and preprocessing module is used to collect multi-source operating data from various acquisition devices in the distribution area and perform preprocessing to generate a multi-parameter acquisition data set. The dynamic acquisition and scheduling module is used to analyze parameter fluctuation characteristics, equipment operating status and business importance based on multi-parameter acquisition data sets, dynamically adjust the sampling period and upload period, and generate dynamic acquisition data sequences. The time synchronization processing module is used to construct a unified reference time axis based on dynamically acquired data sequences, separate device clock deviation from communication delay, perform time resampling and missing mapping processing, and generate a unified time series feature sequence. The temporal synchronization compensation module is used to input a unified temporal feature sequence into the improved ModernTCN model, perform temporal misalignment-aware recombination, reliable time propagation and cross-device collaborative convolution processing, and generate a synchronization-compensated temporal sequence. The terminal fusion processing module is used to construct a trusted synchronization domain based on the synchronization compensation time sequence, perform cross-device data reconstruction, anomaly isolation and trusted data replacement processing, and generate a terminal fusion data set. The upload configuration update module is used to perform classified upload and local cache control based on the terminal fusion data set, and generate updated multi-parameter acquisition configuration by combining the dynamic acquisition strategy and time synchronization parameters updated by the main station feedback.

Images