A multi-source heterogeneous power distribution network dispatching data adaptive collection and space-time fusion method

Abstract

The application provides a multi-source heterogeneous power distribution network dispatching data adaptive collection and space-time fusion method, relates to the power distribution network dispatching data processing technical field, obtains original data of heterogeneous data sources in a power distribution network, and generates a unified data object; calculates a collection control factor, adjusts a collection period, and obtains real-time data of the heterogeneous data sources; time stamp correction and alignment are performed on the real-time data to obtain time-aligned data; a spatial correlation relationship between the heterogeneous data sources is determined; low-frequency background data is screened out from the unified data object, and the low-frequency background data and high-frequency data in the time-aligned data are fused to obtain fusion samples; quality scores of fusion consistency of the fusion samples are calculated, and reliability scores of the heterogeneous data sources are adjusted according to the quality scores. The application can realize space-time consistent fusion of multi-source heterogeneous power distribution network dispatching data in a complex dispatching scene, thereby improving the credibility of the power distribution network dispatching data fusion result.

Description

Technical Field

[0001] This application relates to the field of power distribution network dispatch data processing technology, and more specifically, to a method for adaptive acquisition and spatiotemporal fusion of multi-source heterogeneous power distribution network dispatch data. Background Technology

[0002] Distribution network dispatch data processing technology is a crucial foundational technology supporting distribution network operation monitoring, fault assessment, risk warning, and dispatch decision-making. It is mainly used to access, transform, correlate, and apply data from sources such as dispatch automation systems, edge acquisition terminals, equipment ledger systems, maintenance defect systems, inspection log systems, and environmental monitoring systems. As the scale of distribution network equipment expands and operating scenarios become more complex, dispatch data exhibits characteristics such as diverse sources, inconsistent formats, significant differences in sampling frequencies, and complex spatial attributions. How to effectively process multi-source heterogeneous data directly affects the completeness of distribution network operation status perception and the reliability of dispatch analysis results.

[0003] In existing technologies, power distribution network data processing is mostly concentrated on a single data source or a single business system. Data exchange typically relies on manual interface development and specific protocol adaptation methods. Adding new equipment or systems requires repeated modifications, making system expansion difficult. Furthermore, existing multi-source data association methods often rely on original timestamps, unified time segments, or fixed time windows, making it difficult to adapt to inconsistencies in sampling periods, communication delays, and clock precision across different data sources. This easily leads to event mismatches, equipment mismatches, and distortion of state changes. Moreover, high-frequency real-time data and low-frequency non-real-time data are often stored separately or merged using simple field concatenation, lacking... Comprehensive consideration of the effective range, spatial correlation range, data quality, update time and source reliability of low-frequency data can easily lead to the inclusion of failed records, irrelevant equipment records or low-quality data in the fusion process, reducing the credibility of the fusion sample of scheduling data. Existing unified data pools also focus more on standardized storage and query calls, lacking continuous recording and feedback on fusion process parameters and fusion quality, making it difficult to support subsequent optimization of acquisition strategies, time windows and fusion weights. Therefore, how to achieve spatiotemporal consistent fusion of multi-source heterogeneous distribution network scheduling data in complex scheduling scenarios, thereby improving the credibility of distribution network scheduling data fusion results, has become a challenge for the industry. Summary of the Invention

[0004] This application provides an adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network scheduling data, which can realize spatiotemporal consistent fusion of multi-source heterogeneous distribution network scheduling data in complex scheduling scenarios, thereby improving the reliability of distribution network scheduling data fusion results.

[0005] In a first aspect, this application provides a method for adaptive acquisition and spatiotemporal fusion of multi-source heterogeneous distribution network dispatch data, comprising the following steps: Obtain raw data from heterogeneous data sources in the power distribution network, perform standardized transformation, and generate a unified data object; The acquisition control factor is calculated based on the state awareness characteristics of the heterogeneous data source, and the acquisition cycle is adjusted through the acquisition control factor to obtain real-time data from the heterogeneous data source. The real-time data is timestamped and aligned using a dynamic time window based on the heterogeneous data source to obtain time-aligned data. The spatial relationships between heterogeneous data sources are determined based on the topological neighborhood of device nodes in the distribution network equipment topology model; Based on the spatial correlation, low-frequency background data is filtered out from the unified data object, and the low-frequency background data and high-frequency data in the time-aligned data are fused to obtain a fused sample. The fused samples are scored for fusion consistency, and the reliability scores of the heterogeneous data sources are adjusted based on the quality scores.

[0006] In this embodiment, acquiring raw data from heterogeneous data sources in the distribution network and performing standardized transformation to generate a unified data object specifically includes: Obtain raw data from heterogeneous data sources in the power distribution network; Convert the field names, types, and units of the original data into unified standard fields, standard types, and standard units; Records with field conflicts, missing key data, inconsistent units, and out-of-bounds values ​​after conversion are cleaned, repaired, and quality tags are generated to obtain a unified data object.

[0007] In this embodiment, calculating the acquisition control factor based on the state-aware characteristics of heterogeneous data sources specifically includes: Obtain device type score, topology location score, load importance score, and historical defect score for devices corresponding to heterogeneous data sources; The degree of data variability is calculated based on observations from heterogeneous data sources and statistical characteristics within a historical sliding window; The service importance of the equipment is calculated based on the equipment type score, the topology location score, the load importance score, and the historical defect score. Determine the low-frequency data update status of the corresponding devices of the heterogeneous data source and the event triggering intensity of the heterogeneous data source; The data change rate, the device service importance, the low-frequency data update status, and the event trigger intensity are used as state awareness features of heterogeneous data sources. They are then weighted and summed according to preset weight coefficients to obtain the acquisition control factor.

[0008] In this embodiment, adjusting the acquisition period and acquiring real-time data from heterogeneous data sources through the acquisition control factor specifically includes: Obtain the basic collection cycle corresponding to heterogeneous data sources; The actual acquisition cycle of the heterogeneous data source is calculated based on the basic acquisition cycle and the acquisition control factor. Real-time data from heterogeneous data sources is acquired according to the stated actual acquisition cycle.

[0009] In this embodiment, the real-time data is timestamped and aligned according to the dynamic time window of the heterogeneous data source to obtain time-aligned data, specifically including: A linear correction relationship is established based on the historical timestamp sequence of heterogeneous data sources, and the original timestamps of the real-time data are fundamentally corrected based on the linear correction relationship. Determine the reliability score of heterogeneous data sources; A dynamic time window for heterogeneous data sources is generated based on the communication latency fluctuations and data types of the heterogeneous data sources, as well as the reliability score. The dynamic time window is used to perform timestamp matching and alignment on the real-time data after basic correction to obtain time-aligned data.

[0010] In this embodiment, determining the spatial association between heterogeneous data sources based on the topological neighborhood of device nodes in the distribution network equipment topology model specifically includes: Map the device identifiers corresponding to each data point in the time-aligned data to the device nodes in the distribution network topology model; The topological neighborhood of a device node is determined based on its feeder connection relationship, transformer substation affiliation relationship, and upstream and downstream adjacency relationship in the distribution network topology model. The spatial association between heterogeneous data sources is determined based on the overlapping and adjacency relationships between the topological neighborhoods of the corresponding device nodes.

[0011] In this embodiment, determining the topological neighborhood of a device node based on its feeder connection relationship, transformer substation affiliation relationship, and upstream / downstream adjacency relationship in the distribution network topology model specifically includes: Construct an electrical connection diagram between equipment nodes based on their feeder connection relationships, transformer substation affiliation relationships, and upstream and downstream adjacency relationships in the distribution network topology model; Select a device node as the selected device node. In the electrical connection diagram, extend a preset topological distance outward along the connection path with the selected device node as the center. All device nodes falling within the preset topological distance are determined as the topological neighborhood of the selected device node.

[0012] In this embodiment, filtering low-frequency background data from the unified data object based on the spatial correlation specifically includes: Extract high-frequency data from the time-aligned data; Determine the spatial candidate set of high-frequency data based on the spatial correlation; Low-frequency background data that fall into the spatial candidate set are extracted from the unified data object.

[0013] In this embodiment, fusing the low-frequency background data and the high-frequency data in the time-aligned data to obtain the fused sample specifically includes: Candidate fusion data with an effective time interval covering the current moment are selected from the low-frequency background data; The fusion weights are generated based on the data quality score, time freshness, topological distance, and source reliability of the candidate fusion data. The candidate fusion data and the high-frequency data in the time-aligned data are weighted and fused according to the fusion weight, and the samples are combined to obtain fused samples.

[0014] In this embodiment, the process of scoring the fusion consistency of the fused samples and adjusting the reliability score of the heterogeneous data source based on the quality score specifically includes: The quality score of the fused sample is calculated by weighting the scores of each of the four dimensions: data source quality, time alignment quality, spatial correlation quality, and fusion consistency. The quality score is compared with a preset quality threshold. If it is lower than the preset quality threshold, the feedback direction is determined based on the relative difference between each sub-score and its respective threshold. The feedback processing action is executed according to the feedback direction, and after the feedback processing action is executed, the reliability score of the heterogeneous data source is updated based on the average fusion quality score within the time window.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The process involves: acquiring raw data from heterogeneous data sources in the distribution network and performing standardized transformation to generate a unified data object; calculating acquisition control factors based on the state-aware characteristics of the heterogeneous data sources, adjusting the acquisition cycle using these control factors, and acquiring real-time data from the heterogeneous data sources; performing timestamp correction and alignment on the real-time data based on the dynamic time windows of the heterogeneous data sources to obtain time-aligned data; determining the spatial relationships between heterogeneous data sources based on the topological neighborhoods of device nodes in the distribution network equipment topology model; filtering low-frequency background data from the unified data object based on these spatial relationships, fusing the low-frequency background data and high-frequency data from the time-aligned data to obtain a fused sample; performing a quality score on the fused sample based on its fusion consistency, and adjusting the reliability score of the heterogeneous data sources based on the quality score.

[0016] Therefore, this application can perform a quality score on the fusion consistency of the fused samples and adjust the reliability score of heterogeneous data sources based on the quality score. Firstly, by standardizing and converting the original data from heterogeneous data sources in the distribution network to generate unified data objects, differences in data format, field names, device identifiers, and data types between different business systems can be eliminated, avoiding difficulties in correlation caused by inconsistent expression of multi-source data. Secondly, by calculating the acquisition control factor based on the state awareness characteristics of heterogeneous data sources and adjusting the acquisition cycle through the acquisition control factor, the data acquisition process can adapt to the operating state of equipment and data changes, reducing redundant acquisition during stable operation and increasing the density of real-time data acquisition during abnormal fluctuations or critical states, thus enhancing the ability of scheduling data to capture changes in equipment state from the data source. Furthermore, by performing timestamp correction and alignment of real-time data through dynamic time windows and combining the topological neighborhood of device nodes in the distribution network equipment topology model to determine the spatial correlation between heterogeneous data sources, the time misalignment problem caused by sampling cycle differences, communication delays, and clock deviations can be alleviated. This approach reduces erroneous device associations caused by relying solely on text field matching, enabling multi-source data under the same operational event to participate in subsequent processing within a more accurate time reference and device topology scope. Furthermore, by filtering low-frequency background data from a unified data object based on spatial correlations and fusing this low-frequency background data with high-frequency data in time-aligned data, low-frequency background information such as equipment ledgers, defect records, maintenance records, and inspection logs can participate in the construction of high-frequency operational status data within an effective correlation range. This avoids simply splicing failure records, irrelevant equipment records, or low-quality background data into the fused sample, thereby enhancing the fused sample's ability to reflect the actual operational status of the distribution network. Finally, by performing a quality score on the consistency of the fused sample and adjusting the reliability scores of heterogeneous data sources based on the quality score, consistency issues in the fused sample can be fed back to the data source evaluation stage. This provides a basis for subsequent acquisition cycle correction, time alignment range adjustment, and data source weight reduction, enabling the distribution network dispatch data processing process to have continuous correction capabilities and improving the credibility of the fused sample when entering the fault assessment, risk warning, and dispatch decision-making stages.

[0017] In summary, the technical solution adopted in this application can achieve spatiotemporal consistency fusion of multi-source heterogeneous distribution network dispatch data in complex dispatch scenarios, thereby improving the reliability of distribution network dispatch data fusion results. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is an exemplary flowchart of a method for adaptive acquisition and spatiotemporal fusion of multi-source heterogeneous distribution network scheduling data provided in this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] This application provides an adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network scheduling data. The core of this method involves acquiring raw data from heterogeneous data sources in the distribution network, performing standardized transformation to generate a unified data object; calculating acquisition control factors based on the state-aware characteristics of the heterogeneous data sources, adjusting the acquisition cycle using these control factors, and acquiring real-time data from the heterogeneous data sources; performing timestamp correction and alignment on the real-time data based on the dynamic time windows of the heterogeneous data sources to obtain time-aligned data; determining the spatial relationships between heterogeneous data sources based on the topological neighborhoods of device nodes in the distribution network equipment topology model; filtering low-frequency background data from the unified data object based on these spatial relationships, fusing the low-frequency background data and high-frequency data from the time-aligned data to obtain a fused sample; performing a quality score on the fused sample based on its fusion consistency, and adjusting the reliability score of the heterogeneous data sources based on the quality score.

[0022] Example 1 To better understand the above technical solutions, a detailed explanation will be provided below with reference to the accompanying drawings and specific implementation methods. Figure 1 As shown in the figure, this is an exemplary flowchart of a method for adaptive acquisition and spatiotemporal fusion of multi-source heterogeneous distribution network dispatch data according to this embodiment of the application, including the following steps: In step S1, the raw data from heterogeneous data sources in the power distribution network are obtained and standardized to generate a unified data object.

[0023] In this embodiment, obtaining the raw data from heterogeneous data sources in the distribution network and performing standardized transformation to generate a unified data object can be achieved through the following steps: Obtain raw data from heterogeneous data sources in the power distribution network; Convert the field names, types, and units of the original data into unified standard fields, standard types, and standard units; Records with field conflicts, missing key data, inconsistent units, and out-of-bounds values ​​after conversion are cleaned, repaired, and quality tags are generated to obtain a unified data object.

[0024] It should be noted that, in this application, the raw data refers to unstandardized data records output by different heterogeneous data sources in the distribution network according to their respective field structures, data types, and dimensions; the standard field refers to data fields used to uniformly carry the same business meaning in different heterogeneous data sources; the standard type refers to data types that limit the storage format of standard fields according to unified data coding rules; the standard unit refers to the unit of measurement uniformly set for similar observations according to physical quantity type and dimension; the quality mark refers to the marking information used to record field conflicts, key missing values, inconsistent dimensions, and numerical out-of-bounds values, as well as the availability status of the corresponding records; the unified data object refers to data records organized using standard fields, standard types, and standard units and carrying quality marks, and the unified data object includes equipment identifier, source system, data type, original timestamp, observation value or business content, spatial attribution information, and quality mark.

[0025] In practical implementation, firstly, data records output by the scheduling automation system, edge acquisition terminal, equipment ledger system, maintenance defect system, inspection log system, and environmental monitoring system are received through pre-configured data access interfaces. The interface identifier, source system code, communication protocol type, message subject or service address, terminal address, and message header characteristics corresponding to each data access interface are read. The source system of the data record is identified primarily based on the interface identifier and source system code. When the interface identifier is missing or the same data access interface carries multiple types of data, the source system is further identified by combining the communication protocol type, message subject or service address, terminal address, and message header characteristics, and then processed according to the corresponding... The communication protocol and message structure are parsed item by item to extract the original fields. The source identification result, original field name, original field type, original unit, original timestamp, and original value are retained along with the data record. The data record that has completed source identification and message parsing is used as the original data of the heterogeneous data source. Next, a pre-established source field mapping table is called. According to the mapping relationship between the original field name and the standard field, the plant number, interval number, measurement point number, acquisition time, measurement type, measurement value, and quality code are written into the matching standard fields. And according to the type conversion rules configured in the source field mapping table, the time field, numerical field, status field, and text field are converted. The data is converted to the corresponding standard type. Then, the standard unit dictionary is queried according to the physical quantity type. The original units and standard units are compared in terms of dimensions. If the dimensions match, the proportional coefficient and offset recorded in the standard unit dictionary are used to complete the unit conversion. If the dimensions do not match, the original values ​​are retained and written to the pending cleaning state. The data records that have completed field mapping, type conversion, and unit conversion are used as the converted records. Finally, the converted records are checked field by field. When different heterogeneous data sources write to the same standard field and the field values ​​are inconsistent, the data is selected for retention in the following order: priority of precise matching of unified encoding, priority of authoritative source, priority of newer update time, and priority of higher field completeness. The values ​​are recorded, and the reasons for any unused values ​​and conflicts are noted. When a critical missing value can be filled in by the equipment identification conversion rules, the preset binding relationship between the measurement point and the equipment, or the associated field in the same data record, the filling result is written to the corresponding standard field. If the equipment identification, original timestamp, or observation value cannot be filled in, the corresponding record is marked as invalid. When there is a unit inconsistency, the unit conversion is stopped and the original value is retained. When the observation value exceeds the rated range of the equipment or the range allowed by the business, the original value is retained. The anomaly types and their repair status corresponding to field conflicts, critical missing values, unit inconsistencies, and numerical out-of-bounds are written to the quality flag, where the repair status includes not occurring, repaired, and not repaired.The system reads the repair status of each quality anomaly in the quality tags and, according to a pre-configured quality status mapping table, sets the quality status coefficient corresponding to the non-occurring state to 1, the quality status coefficient corresponding to the repaired state to a preset repair coefficient, and the quality status coefficient corresponding to the unrepaired state to 0. The preset repair coefficient is set based on the degree of information retention between the cleaned and repaired record and the original record. The quality status coefficients corresponding to field conflicts, key missing values, inconsistent units, and out-of-bounds values ​​are multiplied by their corresponding quality weights, and the products are accumulated. The accumulated result is used as the data quality score for that record. The sum of all quality weights is 1, and the quality weight corresponding to a key missing value is not lower than the quality weights corresponding to the other quality anomalies. When a record is marked as invalid because the device identifier, original timestamp, or observation cannot be completed, the data quality score for that record is set to 0. The cleaned and repaired standard fields, standard types, standard units, quality tags, and data quality scores are encapsulated according to a unified data structure, and the encapsulated data record is used as a unified data object.

[0026] In step S2, a collection control factor is calculated based on the state awareness characteristics of the heterogeneous data source. The collection cycle is adjusted using the collection control factor, and real-time data from the heterogeneous data source is obtained.

[0027] In this embodiment, the calculation of the acquisition control factor based on the state-aware characteristics of heterogeneous data sources can be achieved through the following steps: Obtain device type score, topology location score, load importance score, and historical defect score for devices corresponding to heterogeneous data sources; The degree of data variability is calculated based on observations from heterogeneous data sources and statistical characteristics within a historical sliding window; The service importance of the equipment is calculated based on the equipment type score, the topology location score, the load importance score, and the historical defect score. Determine the low-frequency data update status of the corresponding devices of the heterogeneous data source and the event triggering intensity of the heterogeneous data source; The data change rate, the device service importance, the low-frequency data update status, and the event trigger intensity are used as state awareness features of heterogeneous data sources. They are then weighted and summed according to preset weight coefficients to obtain the acquisition control factor.

[0028] It should be noted that, in this application, the equipment type score refers to the score set according to the role of the equipment type of the heterogeneous data source in the distribution network dispatching business; the topology location score refers to the score set according to the feeder location, transformer location, and upstream and downstream connection location of the heterogeneous data source in the distribution network equipment topology model; the load importance score refers to the score set according to the load level and power supply range of the load associated with the heterogeneous data source; the historical defect score refers to the score set according to the historical defect level, historical defect occurrence, and unresolved defect status of the heterogeneous data source; the statistical characteristics within the historical sliding window refer to statistical parameters composed of the average and standard deviation of the effective observations within the historical sliding window; and the data variability refers to the current observation value of the heterogeneous data source relative to the average observation value within the historical sliding window. The standardization deviation of the data is as follows: the equipment service importance is represented by the equipment type score, topology location score, load importance score, and historical defect score, which together quantify the importance of equipment scheduling services; the low-frequency data update status is the status value used to identify whether low-frequency data related to the corresponding equipment of the heterogeneous data source has been added, modified, or changed in status; the event trigger intensity is the degree of impact of switch changes, protection actions, fault alarms, and over-limit alarms on the collection requirements of the heterogeneous data source; the status awareness feature is the heterogeneous data source status description feature composed of data change degree, equipment service importance, low-frequency data update status, and event trigger intensity; the preset weight coefficient is the coefficient used to configure the participation ratio of each status awareness feature in the collection control factor; the collection control factor is the quantitative parameter used to adjust the collection cycle of the heterogeneous data source.

[0029] In specific implementation, firstly, the device identifiers of the corresponding devices from heterogeneous data sources are read from the unified data object. Based on these device identifiers, the device ledger and historical defect records are associated. From the device ledger, the device type, feeder location, transformer area location, upstream and downstream connection locations, associated load level, and power supply range are read. From the historical defect records, the historical defect level, number of defects, and defect elimination status within the statistical period are read. Functional identifiers are set according to the power transmission, segmented control, interconnection and transfer, branch power supply, and status monitoring functions corresponding to the device type. Functional identifiers that the device actually possesses are recorded as valid states, and functional identifiers that the device does not possess are recorded as invalid states. Then, corresponding functions are configured for the valid functional identifiers according to the degree of participation of each function in scheduling control. Weighting is performed by summing the products of each effective function identifier and its corresponding function weight, and the resulting function contribution value is used as the equipment type score. Based on feeder location, transformer substation location, and upstream / downstream connection location, the section number of the equipment from the feeder power supply side, the section number of the equipment from the transformer substation entrance, and the number of directly connected upstream / downstream equipment are extracted. Each indicator is converted to percentile rank according to its position among similar equipment. Specifically, the feeder section number and transformer substation section number undergo reverse percentile conversion, while the number of directly connected upstream / downstream equipment undergoes forward percentile conversion. The three percentile ranks are then arithmetically averaged, and the average result is used as the topology location score. Associated load levels are converted to load level indicators according to power supply load classification rules, and extracted from the power supply range. Power supply capacity, number of power users, or number of covered load objects are used as indicators of power supply range. The load level indicators and power supply range indicators are vector-normalized using an approximation ideal solution ranking method. The optimal values ​​of each indicator form positive ideal solutions, and the worst values ​​form negative ideal solutions. The Euclidean distance between the equipment load indicator vector and the positive and negative ideal solutions is calculated separately. The ratio of the distance from the equipment load indicator vector to the negative ideal solution to the sum of its distances to the positive and negative ideal solutions is used as the load importance score. Historical defect levels are converted into defect impact coefficients according to defect severity. For eliminated defects, the defect impact coefficient is corrected using an exponential decay function based on the time interval between defect elimination and the current time. For uneliminated defects… The corresponding defect impact coefficient is retained. The corrected defect impact coefficients of each defect record within the statistical period are accumulated, and the accumulated result is converted into a ratio with the set defect score upper limit. The converted result is used as the historical defect score. The obtained results are used as the equipment type score, topology location score, load importance score, and historical defect score of the corresponding equipment of the heterogeneous data source, respectively. Next, the current observation value is read according to the acquisition time sequence of the heterogeneous data source, and the observation values ​​that fall into the historical sliding window and are not marked as invalid by the quality label are selected from the continuous historical records before the current observation value. After arranging them in the order of timestamp, the mean and standard deviation are calculated. The mean and standard deviation are used as the statistical characteristics within the historical sliding window, and the data variability is calculated according to the following formula: in, For the first A heterogeneous data source at time The degree of data change; This refers to the sequence number of the heterogeneous data source; This refers to the acquisition time corresponding to the current observation. For the first A heterogeneous data source at time Observed values; For the first The average of valid observations from heterogeneous data sources within a historical sliding window; The duration of the historical sliding window; This represents the time interval corresponding to the historical sliding window. For the first The standard deviation of valid observations from heterogeneous data sources within a historical sliding window; A positive constant is set to prevent the standard deviation from being zero, and its value is set according to the observation accuracy of the heterogeneous data source; the standardized deviation obtained by dividing the absolute difference between the current observation and the mean by the sum of the standard deviation and the positive constant is used as the data variability; then, the equipment type score, topology location score, load importance score, and historical defect score are multiplied by their corresponding preset weight coefficients, and the equipment business importance is calculated according to the following formula: in, For the first The importance of device services corresponding to each heterogeneous data source; This refers to the sequence number of the heterogeneous data source; For the first Device type ratings for devices corresponding to heterogeneous data sources; For the first Topological location scoring of devices corresponding to heterogeneous data sources; For the first Load importance score for each heterogeneous data source corresponding to the device; For the first Historical defect scores of devices corresponding to each heterogeneous data source; Preset weighting coefficients for equipment type scoring; Preset weighting coefficients for topological location scoring; Preset weighting coefficients for load importance scores; Preset weighting coefficients are assigned to historical defect scores. Each preset weighting coefficient is configured based on the proportion of the corresponding score's impact on distribution network dispatching services. The cumulative result obtained by multiplying each score by its corresponding preset weighting coefficient is used as the equipment service importance. Subsequently, equipment ledgers, defect records, maintenance records, inspection records, operation mode sheets, and environmental information are queried according to equipment identifiers. The record version, update time, and status fields of each low-frequency data are read, and the current record is compared field by field with the previous valid record. When there is a new record, a modification of the record content, or a change in the status field, the low-frequency data update status is assigned the value corresponding to the update status. When there is no new, modified, or status change, the low-frequency data update status is assigned the value corresponding to the non-updated status. The assigned status value is then used as... This involves identifying low-frequency data update status; simultaneously, reading switch changes, protection actions, fault alarms, and limit exceedance alarms occurring within the current collection period, and using a combination of event type, event level, and impact range as search criteria to retrieve corresponding scores from a pre-configured event scoring table. Protection actions, fault alarms, and critical switch changes correspond to higher scores, while general status indicators and ordinary alarms correspond to lower scores. When multiple events exist within the same collection period, the event with the highest score is selected from highest to lowest event level, and this selected score is used as the event trigger intensity. Finally, data change rate, equipment business importance, low-frequency data update status, and event trigger intensity are used as status awareness features of the heterogeneous data source, and each is multiplied by its corresponding preset weight coefficient to calculate the collection control factor using the following formula: in, For the first A heterogeneous data source at time The acquisition control factor; This refers to the sequence number of the heterogeneous data source; To collect the calculation time corresponding to the control factor; For the first A heterogeneous data source at time The degree of data change; For the first The importance of device services corresponding to each heterogeneous data source; For the first Each heterogeneous data source corresponds to a device at time Low-frequency data update status; For the first A heterogeneous data source at time The intensity of the event triggering; The preset weighting coefficients corresponding to the degree of data variability; Preset weighting coefficients corresponding to the importance of equipment operations; Preset weighting coefficients for low-frequency data update status; The preset weight coefficients are the corresponding weights for the intensity of event triggering. Each preset weight coefficient is configured according to the proportion of the impact of data changes, the importance of equipment scheduling services, low-frequency data updates, and discrete events on the collection requirements. The cumulative result obtained by multiplying each state perception feature by its corresponding preset weight coefficient is used as the collection control factor.

[0030] In this embodiment, adjusting the acquisition period and obtaining real-time data from heterogeneous data sources through the acquisition control factor can be achieved using the following steps: Obtain the basic collection cycle corresponding to heterogeneous data sources; The actual acquisition cycle of the heterogeneous data source is calculated based on the basic acquisition cycle and the acquisition control factor. Real-time data from heterogeneous data sources is acquired according to the stated actual acquisition cycle.

[0031] It should be noted that the basic acquisition period mentioned in this application refers to the acquisition time interval pre-configured by the heterogeneous data source when the period is not adjusted according to the state-aware characteristics; the actual acquisition period refers to the period calculated based on the basic acquisition period and the acquisition control factor and used to control the current acquisition time interval of the heterogeneous data source; and the real-time data refers to the current operating data of the device obtained from the heterogeneous data source according to the actual acquisition period.

[0032] In practice, firstly, the corresponding acquisition configuration is read according to the source system and data type of the heterogeneous data source, and the pre-set initial acquisition time interval is extracted from the acquisition configuration. When the same heterogeneous data source contains multiple data types, the initial acquisition time interval corresponding to each data type is read separately, and the initial acquisition time interval matching the data type of the current acquisition task is used as the basic acquisition cycle corresponding to the heterogeneous data source. Secondly, the acquisition control factor obtained in the previous steps and the pre-configured adjustment coefficient are read, and the basic acquisition cycle, acquisition control factor, and adjustment coefficient are substituted into the following formula to calculate the actual acquisition cycle of the heterogeneous data source: in, For the first A heterogeneous data source at time The actual collection cycle; This refers to the sequence number of the heterogeneous data source; To calculate the current moment of the actual data acquisition cycle; For the first The basic collection cycle corresponding to each heterogeneous data source; This is an adjustment coefficient used to configure the adjustment range of the acquisition control factor on the basic acquisition cycle; For the first A heterogeneous data source at time The acquisition control factor is used; the constant term 1 in the denominator is used to make the actual acquisition period equal to the basic acquisition period when the acquisition control factor is zero; the acquisition time interval obtained according to the above formula is used as the actual acquisition period of the heterogeneous data source; finally, the actual acquisition period is written into the acquisition task timer corresponding to the heterogeneous data source, and the actual acquisition period is used as the timing interval between two adjacent acquisition tasks; for heterogeneous data sources using the polling method, when the acquisition task timer reaches the actual acquisition period, a data read request is sent through the corresponding data access interface, and the data records returned by the heterogeneous data source are received; for heterogeneous data sources actively uploaded by the edge acquisition terminal, the actual acquisition period is written into the sampling period configuration and upload period configuration of the edge acquisition terminal, and the edge acquisition terminal performs data sampling and uploads data records according to the actual acquisition period; the device identifier, data type, original timestamp, and observation value are read from the received data records, and the data records containing the above content are used as the real-time data of the heterogeneous data source.

[0033] In step S3, the real-time data is timestamped and aligned according to the dynamic time window of the heterogeneous data source to obtain time-aligned data.

[0034] In this embodiment, the real-time data is timestamped and aligned according to the dynamic time window of the heterogeneous data source to obtain time-aligned data. This can be achieved through the following steps: A linear correction relationship is established based on the historical timestamp sequence of heterogeneous data sources, and the original timestamps of the real-time data are fundamentally corrected based on the linear correction relationship. Determine the reliability score of heterogeneous data sources; A dynamic time window for heterogeneous data sources is generated based on the communication latency fluctuations and data types of the heterogeneous data sources, as well as the reliability score. The dynamic time window is used to perform timestamp matching and alignment on the real-time data after basic correction to obtain time-aligned data.

[0035] It should be noted that, in this application, the historical timestamp sequence refers to the time record sequence obtained by arranging the original timestamps generated by heterogeneous data sources in the continuous historical acquisition period according to the acquisition sequence; the linear correction relationship refers to the linear mapping relationship between the original timestamps and the unified reference time; the basic correction refers to the time correction process of converting the original timestamps of real-time data to the unified reference time axis using the linear correction relationship; the reliability score refers to the data credibility score obtained by quantifying the data integrity rate, communication latency, packet loss rate, and outlier ratio of heterogeneous data sources; the communication latency fluctuation refers to the dispersion of each data transmission latency of heterogeneous data sources relative to the average transmission latency within the continuous observation period; the data type refers to the data category divided according to the data record generation method and time change form; the dynamic time window refers to the timestamp matching range dynamically configured according to the communication latency fluctuation, data type, and reliability score; and the time-aligned data refers to the data obtained after matching the basic-corrected real-time data to the unified time axis and completing the observation aggregation or event attachment according to the data type.

[0036] In practice, firstly, historical time samples are read from the most recent sliding statistical window of the heterogeneous data source. The sliding statistical window uses the most recent 10 minutes or at least 30 valid records. Each historical time sample includes the device sending time, the system receiving time, and the available standard time synchronization time. For heterogeneous data sources that do not provide a standard time synchronization time, the system receiving time minus the median communication delay within the sliding statistical window is used as the reference time. Historical time samples with invalid time formats, duplicate records, non-monotonic device sending times, system receiving times earlier than device sending times, and communication delays deviating from the median communication delay by more than three times the median absolute deviation are removed. The remaining historical time samples are then processed. The original timestamp is used as the independent variable, and the corresponding reference time is used as the dependent variable. Sample weights are configured according to communication delay fluctuations, message verification results, and reliability scores of the previous time window. Historical time samples with smaller communication delay fluctuations, normal message verification, and higher reliability scores in the previous time window are given larger weights. The weighted least squares method is used to construct the weighted sum of squared residuals between the original timestamp and the reference time. The clock drift correction coefficient and time offset correction amount that minimize the weighted sum of squared residuals are calculated. The mapping relationship between the clock drift correction coefficient and the time offset correction amount is used as a linear correction relationship, and the original timestamp of the real-time data is fundamentally corrected according to the following formula: in, For the first The first heterogeneous data source uploaded The timestamp of each real-time data point after basic correction; This refers to the sequence number of the heterogeneous data source; This refers to the record sequence number of real-time data in the heterogeneous data source; For the first The first heterogeneous data source uploaded The original timestamp of each real-time data point; For the first Clock drift correction coefficients for heterogeneous data sources are used to correct the cumulative drift of the local clock of the heterogeneous data source relative to the unified reference time axis. For the first The time offset correction amount for each heterogeneous data source is used to compensate for the fixed time deviation of the local clock of the heterogeneous data source relative to the unified reference time axis. The timestamps converted according to the linear correction relationship are written into the corresponding real-time data, and the real-time data after the timestamp conversion is used as the real-time data after basic correction. The clock drift correction coefficient and the time offset correction amount are refitted according to the preset update cycle, which is 5 minutes. When the correction residual exceeds the preset threshold, the refitting is performed in advance. When there are insufficient effective historical time samples, the linear correction relationship of the previous update cycle is continued to be used. Secondly, within the most recent time window, the number of data records that should arrive from the heterogeneous data source is counted according to the actual acquisition cycle, and the number of data records that are actually received and not marked as invalid by the quality label is counted. The ratio of the actual received number to the number that should arrive is mapped to the data integrity score. The system reception of each data record is calculated. The time difference between the original time and the corrected timestamp is used to match the median of the time difference with the allowable communication delay range corresponding to the heterogeneous data source. A communication delay score is obtained by assigning a higher score to lower communication delays. When data records carry data sequence numbers, the number of lost records is counted based on consecutive gaps in the data sequence numbers. When no data sequence numbers are carried, the number of lost records is counted based on the difference between the expected number and the actual number received. The ratio of lost records to expected numbers is converted into a packet loss rate score, with a higher score for lower packet loss rates. Quality markers in the unified data object are read, and the number of data records with out-of-bounds values, duplicates, abnormal timestamps, or abnormal formats is counted. The ratio of abnormal data records to actual received numbers is converted into an outlier ratio score, with a higher score for lower outlier ratios. A reliability score is then calculated using the following formula: in, For the first A heterogeneous data source at time Reliability score; This refers to the calculation time corresponding to the reliability score; For the first Data integrity score of heterogeneous data sources within the most recent time window; For the first Communication latency scores for heterogeneous data sources within the most recent time window; For the first Packet loss rate score for each heterogeneous data source within the most recent time window; For the first A score showing the proportion of outliers from heterogeneous data sources within the most recent time window; The weighting coefficients corresponding to the data completeness score; These are the weighting coefficients corresponding to the communication delay score; The weighting coefficients corresponding to the packet loss rate score; The weighting coefficients for the outlier ratio scores are configured according to the data types of the heterogeneous data sources and the requirements of distribution network dispatching services. The cumulative result obtained by multiplying each score by its corresponding weighting coefficient is used as the reliability score of the heterogeneous data source. Next, the system reception time and the timestamp after basic correction for each real-time data point within the most recent time window are read, and the time difference between the two is used as the communication delay for each data transmission. The standard deviation of the communication delay sequence is calculated, and the calculated standard deviation is used as the communication delay fluctuation. Data type correction items are read from a pre-configured data type correction table according to the data type. For continuous high-frequency telemetry data, the data type correction items are configured according to the actual acquisition cycle; for switch changes, protection actions, and fault alarms, the data type correction items are configured according to the time deviation allowed by the event association. The communication delay fluctuation, reliability score, and data type correction items are substituted into the following formula to generate the dynamic time window for the heterogeneous data source: in, For the first A heterogeneous data source at time The width of the dynamic time window; The base time window for heterogeneous data sources; For the first Communication latency fluctuations of heterogeneous data sources within the most recent time window; This indicates the communication latency of real-time data; For the first A heterogeneous data source at time Reliability score; For the first Data type correction items corresponding to each heterogeneous data source; This is the adjustment coefficient corresponding to the communication delay fluctuation. This is the adjustment coefficient corresponding to the reliability score; The adjustment coefficient corresponding to the data type correction term is used; the dynamic time window is widened when the communication delay fluctuation increases and narrowed when the reliability score improves. The time range obtained according to the above formula is used as the dynamic time window of the heterogeneous data source; finally, the pre-configured unified time axis is read. For continuous high-frequency telemetry data, the absolute time difference between the timestamp after basic correction and each time point in the unified time axis is calculated one by one. When the absolute time difference satisfies the following formula, the corresponding real-time data is used as the candidate observation value for that time point: in, For the first The first heterogeneous data source uploaded The timestamp of each real-time data point after basic correction; To unify the timeline of the first A point in time; To unify the sequence numbers of time points in the timeline; For the first At a given time point, heterogeneous data sources The corresponding dynamic time window width; when there is only one candidate observation within the same dynamic time window, that candidate observation is taken as the time point. The aligned observations; when multiple candidate observations exist within the same dynamic time window, time-decay weighted aggregation is performed according to the following formula: in, For the first Aligning heterogeneous data sources to a point in time Observed values; For the first The first heterogeneous data source uploaded The observation value of the real-time data; For the first The timestamp of each real-time data point after basic correction; To unify the target time point in the timeline; The candidate observation number that falls within the dynamic time window; This is the time decay coefficient, used to control the rate at which the weight of a candidate observation decreases as the absolute time difference between it and the target time point increases; The natural exponential function is used; the observations obtained by time-decay weighted aggregation are used as the aligned observations of continuous high-frequency telemetry data at the target time point; for switch changes, protection actions, and fault alarms, event records whose timestamps fall within the dynamic time window after basic correction are attached to the corresponding time point or time slice. When the same event occurs repeatedly in multiple heterogeneous data sources, deduplication is performed according to event type, device identifier, and time difference, and the event record with the smallest time difference is retained; for switch status, the most recent valid status is used before a new switch change occurs; the real-time data after observation aggregation, event attachment, and status maintenance are organized according to a unified time axis, and the organized data is used as time-aligned data.

[0037] In step S4, the spatial association between heterogeneous data sources is determined based on the topological neighborhood of the equipment nodes in the power distribution network equipment topology model.

[0038] In this embodiment, determining the spatial association between heterogeneous data sources based on the topological neighborhood of device nodes in the distribution network equipment topology model can be achieved through the following steps: Map the device identifiers corresponding to each data point in the time-aligned data to the device nodes in the distribution network topology model; The topological neighborhood of a device node is determined based on its feeder connection relationship, transformer substation affiliation relationship, and upstream and downstream adjacency relationship in the distribution network topology model. The spatial association between heterogeneous data sources is determined based on the overlapping and adjacency relationships between the topological neighborhoods of the corresponding device nodes.

[0039] It should be noted that, in this application, the device node refers to the node object corresponding to the distribution network equipment in the distribution network topology model; the overlapping relationship refers to the overlapping relationship of sets of the same device nodes in the topological neighborhood of the device node corresponding to the heterogeneous data source; the adjacency relationship refers to the adjacency relationship between the topological neighborhoods of the device node corresponding to the heterogeneous data source, where there is no overlap, but the device nodes in the two topological neighborhoods have a direct electrical connection; and the spatial association relationship refers to the spatial connection between heterogeneous data sources established based on the overlapping and adjacency relationships between topological neighborhoods.

[0040] In practice, firstly, the device identifiers corresponding to each data point in the time-aligned data are read one by one. These device identifiers are then used as search criteria in the device mapping table, prioritizing precise matching based on unified asset codes, scheduling equipment codes, and pre-defined binding relationships between measurement points and devices. If no unique match is found, the range of candidate device nodes is narrowed down using terminal address, feeder, transformer area number, and device type. If multiple candidate device nodes still exist, an identifier consistency score is obtained based on the matching results of unified asset codes, scheduling equipment codes, or pre-defined binding relationships. Finally, a classification score is obtained based on the matching results of device type, device name, and rated parameters. The topology consistency score is obtained based on the matching of feeder, transformer area, and upstream / downstream adjacency relationships. The three scores are weighted according to weights of 0.50, 0.25, and 0.25 respectively. When the highest comprehensive matching score is not lower than 0.85, and the difference between the highest and second-highest comprehensive matching scores is not lower than 0.10, the candidate device node corresponding to the highest comprehensive matching score is mapped to the corresponding data in the time-aligned data to the device node in the distribution network topology model. When the highest comprehensive matching score is between 0.65 and 0.85, or the difference between the highest and second-highest comprehensive matching scores is not lower than 0.10, the candidate device node corresponding to the highest comprehensive matching score is mapped to the device node in the distribution network topology model. When the difference between the two comprehensive matching scores is less than 0.10, the corresponding data is marked as pending review. When the highest comprehensive matching score is lower than 0.65, or when the candidate device node crosses incompatible feeders or transformer areas, the corresponding data is marked as unmapped and not included in the subsequent spatial association establishment process. Secondly, the topological neighborhood of the device node is determined based on the feeder connection relationship, transformer area affiliation relationship, and upstream and downstream adjacency relationship of the device node in the distribution network topology model. Finally, according to the source identifier of the heterogeneous data source, the topological neighborhood of the device node corresponding to each heterogeneous data source is read, and the device node identifiers in the two topological neighborhoods are set together. The intersection operation treats the relationship between two topological neighborhoods as an overlap relationship when the intersection contains at least one device node identifier. When the intersection is empty, it queries whether there is a direct electrical connection between a device node in one topological neighborhood and a device node in another topological neighborhood. If a direct electrical connection exists, the relationship between the two topological neighborhoods is treated as an adjacency relationship. If no direct electrical connection exists, no spatial connection is established between the corresponding heterogeneous data sources. The heterogeneous data sources with overlapping or adjacency relationships and their corresponding device nodes are associated with records, and these associated records are used as the spatial association relationship between the heterogeneous data sources.

[0041] In this embodiment, determining the topological neighborhood of a device node based on its feeder connection relationship, transformer substation affiliation relationship, and upstream / downstream adjacency relationship in the distribution network topology model can be achieved through the following steps: Construct an electrical connection diagram between equipment nodes based on their feeder connection relationships, transformer substation affiliation relationships, and upstream and downstream adjacency relationships in the distribution network topology model; Select a device node as the selected device node. In the electrical connection diagram, extend a preset topological distance outward along the connection path with the selected device node as the center. All device nodes falling within the preset topological distance are determined as the topological neighborhood of the selected device node.

[0042] It should be noted that, in this application, the feeder connection relationship refers to the connection relationship formed by the device node according to its feeder and actual electrical wiring; the substation affiliation relationship refers to the device affiliation relationship formed by the device node according to its substation; the upstream and downstream adjacency relationship refers to the direct upstream or downstream connection relationship formed by the device node along the power supply direction of the feeder; the electrical connection diagram refers to a graph structure with device nodes as graph nodes, direct electrical connections between device nodes as graph connection edges, and recording substation affiliation relationships and upstream and downstream adjacency relationships; the selected device node refers to the device node that serves as the starting point for topology distance expansion in the electrical connection diagram; the connection path refers to the node connection sequence that starts from the selected device node and passes through the graph connection edges between device nodes in sequence; the preset topology distance refers to the maximum number of graph connection edges allowed to be traversed along the connection path starting from the selected device node; and the topology neighborhood refers to the set of device nodes consisting of the selected device node and the device nodes that can be reached from the selected device node within the preset topology distance.

[0043] In practice, firstly, the device identifier, feeder, transformer substation, upstream device identifier, and downstream device identifier corresponding to each device node in the distribution network topology model are read. Each device node is written into the graph node set, and its feeder and transformer substation are written into the attribute fields of the corresponding graph node. For two device nodes with direct electrical connection in the feeder connection relationship, the upstream device node is pointed to the downstream device node according to the upstream and downstream adjacency relationship, and the corresponding graph connection edge is established. For device nodes with the same transformer substation affiliation relationship, the feeder connection records within the transformer substation are further queried. Graph connection edges are only established when the feeder connection record indicates that there is a direct electrical connection between the two device nodes. Device nodes with the same transformer substation affiliation relationship but no direct electrical connection are not established. After writing the graph nodes and graph connection edges item by item, the graph structure containing device nodes, feeder connection relationships, transformer substation affiliation relationships, and upstream and downstream adjacency relationships is used as the electrical connection between device nodes. The process involves several steps: First, selecting a device node from the electrical connection diagram. The topological distance of the selected device node is set to zero and added to the sequence of nodes to be traversed. A breadth-first traversal is used to sequentially read the device nodes in the sequence. Adjacent device nodes are searched along the upstream and downstream connections of the current device node, and their topological distances are incremented by one. If an adjacent device node has not yet been traversed and its topological distance does not exceed a preset topological distance, it is added to the sequence. If the topological distance of an adjacent device node reaches the preset topological distance, it is retained but no further expansion is initiated from that node. Traversed device nodes are not added back to the sequence. After traversing all device nodes in the sequence, the selected device node and all device nodes whose topological distance does not exceed the preset topological distance are considered the topological neighborhood of the selected device node.

[0044] In step S5, low-frequency background data is filtered out from the unified data object according to the spatial correlation, and the low-frequency background data and the high-frequency data in the time-aligned data are fused to obtain a fused sample.

[0045] In this embodiment, the low-frequency background data can be filtered from the unified data object based on the spatial correlation relationship using the following steps: Extract high-frequency data from the time-aligned data; Determine the spatial candidate set of high-frequency data based on the spatial correlation; Low-frequency background data that fall into the spatial candidate set are extracted from the unified data object.

[0046] It should be noted that, in this application, the high-frequency data refers to data continuously acquired at a relatively short acquisition cycle and used to record the operating status or discrete events of power distribution network equipment; the spatial candidate set refers to a set composed of the equipment nodes corresponding to the high-frequency data and the equipment nodes that have spatial correlation with the equipment nodes; and the low-frequency background data refers to data with an update frequency lower than that of the high-frequency data and used to provide equipment ledgers, defect records, maintenance records, inspection logs, operation mode sheets or environmental information.

[0047] In practice, firstly, the data type and device identifier of each piece of time-aligned data are read one by one, and the data type is matched with the data type dictionary pre-configured during standardization conversion. When the data type belongs to voltage, current, active power, reactive power, frequency, temperature, switch status, protection action, or alarm information, the corresponding data, its device identifier, and the aligned timestamp are retained, and the retained data is regarded as high-frequency data. Secondly, for each piece of high-frequency data, the device node mapped to the distribution network topology model is read through the device identifier, and the device node is used as the query condition for spatial association. Device nodes with overlapping or adjacent relationships with the topological neighborhood of the device node are retrieved. The device node corresponding to the high-frequency data is added to the search. The obtained device nodes are processed, and duplicate device nodes are removed according to their device identifiers. The deduplicated set of device nodes is used as the spatial candidate set for high-frequency data. Finally, the unified data object is traversed, and data records corresponding to equipment ledgers, defect records, maintenance records, inspection logs, operation mode sheets, and environmental information are filtered according to data type. The device nodes mapped to the distribution network topology model are read through the device identifiers of each data record, and the read device nodes are matched one by one with the device nodes in the spatial candidate set. When the device identifiers match, the corresponding data record, its valid time interval, and data quality score are retained. When the device identifiers do not match, the corresponding data record is excluded, and the retained data record is used as low-frequency background data falling into the spatial candidate set.

[0048] In this embodiment, fusing the low-frequency background data and the high-frequency data in the time-aligned data to obtain the fused sample can be achieved through the following steps: Candidate fusion data with an effective time interval covering the current moment are selected from the low-frequency background data; The fusion weights are generated based on the data quality score, time freshness, topological distance, and source reliability of the candidate fusion data. The candidate fusion data and the high-frequency data in the time-aligned data are weighted and fused according to the fusion weight, and the samples are combined to obtain fused samples.

[0049] It should be noted that, in this application, the effective time interval refers to the start and end time range within which the business content in the low-frequency background data remains valid for the distribution network equipment; the current moment refers to the time point corresponding to the high-frequency data in the time-aligned data on a unified time axis; the candidate fusion data refers to low-frequency background data whose effective time interval covers the current moment and can participate in high-frequency data fusion; the data quality score refers to the data availability score set based on the field completeness, timestamp validity, reasonableness of numerical range, duplication, and source reliability of the candidate fusion data; the time freshness refers to the proximity of the most recent update time of the candidate fusion data to the current moment; the topological distance refers to the number of graph connection edges contained in the connection path between the device node corresponding to the high-frequency data and the device node corresponding to the candidate fusion data; the source reliability refers to the credibility of the heterogeneous data source to which the candidate fusion data belongs in terms of data integrity rate, communication latency, packet loss rate, and outlier ratio; the fusion weight refers to the normalized coefficient of the proportion of the candidate fusion data participating in the current high-frequency data fusion; and the fusion sample refers to the data sample obtained by combining the candidate fusion data and high-frequency data according to the fusion weight.

[0050] In practice, firstly, high-frequency data in the time-aligned data is read one by one, and the time point corresponding to the high-frequency data on a unified time axis is taken as the current time. Then, the low-frequency background data obtained by filtering in the previous steps is read according to the equipment identifier. For defect records, maintenance records, and operation mode sheets with effective and expiration times, the current time is compared sequentially with the effective and expiration times respectively. If the current time is not earlier than the effective time and not later than the expiration time, the corresponding low-frequency background data is retained. For defect records without an expiration time, they are retained until the defect elimination record appears. For equipment ledgers, they are retained until the next equipment ledger update. For inspection logs and maintenance records without an expiration time, the sum of the record time and the effective period is taken as the end time of the effective time interval according to the pre-configured validity period. The process involves several steps: First, low-frequency background data whose effective time interval, after time comparison, covers the current moment is selected as candidate fusion data. Second, data quality scores are read from the quality markers carried by the candidate fusion data. The time interval between the most recent update time of the candidate fusion data and the current moment is used as the basis for calculating time freshness, with smaller intervals resulting in higher values. The number of graph edges contained in the shortest connection path between the device nodes corresponding to the high-frequency data and the device nodes corresponding to the candidate fusion data is read from the distribution network topology model and used as the topological distance. The reliability score of the heterogeneous data source to which the candidate fusion data belongs is also read and used as the source reliability. Finally, the data quality score, the time interval between the most recent update time and the current moment, the topological distance, and the source reliability are substituted into the following formula to generate the fusion weight: in, For the current moment Next The candidate fusion data relative to the first The fusion weights of high-frequency data; The sequence number of the high-frequency data; The sequence number of the candidate fusion data; This represents the current moment corresponding to high-frequency data on a unified time axis. For the first Candidate fusion data; For the first Data quality score for each candidate fused data; For the first The time interval between the most recent update time of each candidate fused data item and the current time. This is the time decay coefficient, used to control the rate at which the fusion weights decrease as the time interval increases; For the first The device node corresponding to each high-frequency data point; For the first The device nodes corresponding to each candidate fused data; For device nodes With device nodes Topological distance between them; This is the topology distance attenuation coefficient, used to control the rate at which the fusion weights decrease as the topology distance increases; For the first The reliability of the source of the candidate fusion data; For the current moment The following and the first A candidate fusion dataset that matches a set of high-frequency data points; For the first candidate fusion dataset Candidate fusion data; The candidate fusion data index used when summing the candidate fusion data set; For the first Data quality score for each candidate fused data; For the first The time interval between the most recent update time of each candidate fused data item and the current time. For the first The device nodes corresponding to each candidate fused data; For device nodes With device nodes Topological distance between them; For the first The reliability of the source of each candidate fusion data is assessed; the denominator is used to normalize the weight calculation results corresponding to the candidate fusion data so that the sum of the fusion weights of the candidate fusion data is 1; the normalization coefficient obtained according to the above formula is used as the fusion weight; finally, the candidate fusion data is encoded according to the field types in the unified data object, where numerical fields are normalized according to the rated range of equipment, the permitted range of business, or the historical statistical range, categorical fields are one-hot encoded or fixed-length encoded according to the preset category dictionary, level fields are mapped to numerical values ​​according to the business level order and normalized, time fields are converted into numerical values ​​relative to the current time, the remaining valid time, and whether it is in the valid time interval, and text fields are extracted according to the power distribution equipment business dictionary to extract equipment location, defect type, severity, handling measures, and handling results, and the appearance and frequency of keywords are encoded as fixed-length numerical values; the encoding results of each field are arranged in a fixed field order, and then the candidate fusion data is weighted and fused according to the following formula: in, For the current moment The following and the first The weighted fusion result of candidate fusion data matched with high-frequency data; The sequence number of the high-frequency data; The current moment; In order to be with the first A candidate fusion dataset that matches a set of high-frequency data points; For the first candidate fusion dataset Candidate fusion data; The sequence number of the candidate fusion data; For the first The candidate fusion data relative to the first The fusion weights of high-frequency data; For the first Numerical representations obtained after field encoding of candidate fused data; This involves summing the products of each candidate fusion data point and its corresponding fusion weight in the candidate fusion data set; using the summation result as the weighted fusion result of the candidate fusion data; subsequently, encoding the high-frequency data in the time-aligned data according to the numerical range and field order matching the candidate fusion data; calculating the field difference, field ratio, or normalized product for high-frequency data fields and candidate fusion data fields with the same physical meaning; extracting the corresponding encoded values ​​according to the pre-configured field association relationship and performing element-wise product according to the above calculation results; using the above calculation results as the interaction result between high-frequency data and candidate fusion data; and completing the combination according to the following formula: in, For the first High-frequency data at the current moment The corresponding fusion samples; The sequence number of the high-frequency data; The current moment; For the first Numerical representations obtained after high-frequency data are encoded using fields; In order to be with the first The weighted fusion result of candidate fusion data matched with high-frequency data; The interactive results are obtained by using field differences, field ratios, normalized products, or element-wise products after mapping, based on the field types of high-frequency data and candidate fusion data. This function performs interactive computations according to pre-configured field relationships; square brackets indicate that the numerical representation of high-frequency data, the weighted fusion result of candidate fusion data, and the interactive result are combined according to a fixed field order; the combined data is used as the fusion sample.

[0051] In step S6, the fused samples are given a quality score for fusion consistency, and the reliability score of the heterogeneous data source is adjusted based on the quality score.

[0052] In this embodiment, the quality score for the fusion consistency of the fused samples, and the adjustment of the reliability score of the heterogeneous data source based on the quality score, can be achieved by the following steps: The quality score of the fused sample is calculated by weighting the scores of each of the four dimensions: data source quality, time alignment quality, spatial correlation quality, and fusion consistency. The quality score is compared with a preset quality threshold. If it is lower than the preset quality threshold, the feedback direction is determined based on the relative difference between each sub-score and its respective threshold. The feedback processing action is executed according to the feedback direction, and after the feedback processing action is executed, the reliability score of the heterogeneous data source is updated based on the average fusion quality score within the time window.

[0053] It should be noted that, in this application, the data source quality refers to the quality of the data participating in the construction of the fusion sample in terms of data integrity, outlier ratio, communication latency, and source reliability; the time alignment quality refers to the effectiveness of the data participating in the fusion falling within the dynamic time window and the degree of deviation of the timestamp from the target time point; the spatial correlation quality refers to the consistency of device objects, topological distance adaptation, and spatial correlation range between the corresponding device nodes of high-frequency data and low-frequency background data; the fusion consistency refers to the degree of consistency between high-frequency data and low-frequency background data in terms of device operating status and business content; the sub-item score refers to the score obtained by quantifying the data source quality, time alignment quality, spatial correlation quality, and fusion consistency respectively; the quality score refers to the comprehensive quality score of the fusion sample obtained by weighting the sub-item scores of the four dimensions; and the preset quality threshold refers to the threshold used to judge... The scoring thresholds for whether the fused samples meet the quality requirements; the respective thresholds represent the trigger thresholds for data source quality, time alignment quality, spatial correlation quality, and fusion consistency configuration, respectively; the relative difference represents the proportion of the difference when the sub-item score is lower than its respective threshold relative to its respective threshold; the feedback direction represents the quality dimension that needs to prioritize the feedback processing action based on the relative difference; the feedback processing action represents actions performed to address deficiencies in data source quality, time alignment quality, spatial correlation quality, or fusion consistency, such as supplementary sampling, retransmission, sampling priority adjustment, dynamic time window adjustment, time alignment re-execution, device object mapping verification, topology neighborhood verification, or fusion weight adjustment; the time window represents a continuous time range extending forward from the current update time of the reliability score to a preset time length; the average fusion quality score represents the average quality score of the fused samples formed by heterogeneous data sources within the time window.

[0054] In practice, firstly, the data sources, time alignment results, spatial relationships, fusion weights, and quality markers involved in the fusion samples are read one by one. The data integrity rate is obtained based on the ratio between the actual received data volume and the data volume that should be received according to the actual collection cycle. The outlier ratio is obtained by statistically analyzing the proportion of data volume corresponding to out-of-bounds values, duplicates, timestamp anomalies, and format anomalies in the quality markers relative to the actual received data volume. The communication delay experienced by the data from transmission to reception and the reliability scores of the heterogeneous data sources to which the data participating in the fusion belong are also read. The data integrity rate and reliability scores are mapped to a unified scoring range according to the direction of higher values, while the outlier ratio and communication delay are also mapped according to the direction of higher values. Lower values ​​are mapped to the same scoring range, and then weighted according to a pre-configured participation ratio. The weighted result is used as the sub-score corresponding to the data source quality. The absolute time difference between the timestamp of the data participating in the fusion (after basic correction) and the target time point is calculated for each data point. The absolute time difference is divided by the corresponding dynamic time window width, and the ratio minus one is used as the time deviation score for a single data point. Data exceeding the dynamic time window has a time deviation score of zero. The average of each time deviation score is then weighted by the proportion of data falling within the dynamic time window to the total amount of data participating in the matching. The weighted result is used as the sub-score corresponding to the time alignment quality. The mapping between high-frequency data and low-frequency background data is checked line by line. The obtained device nodes are configured with the highest device object consistency score when they are mapped to the same device node. When they are mapped to different device nodes, the topological distance between them is read, and the topological distance score is the ratio of the difference between the preset topological distance and the preset topological distance. If the difference exceeds the preset topological distance, the topological distance score is recorded as zero. At the same time, it is checked whether the device node corresponding to the low-frequency background data belongs to the spatial candidate set of high-frequency data. The check result is weighted with the device object consistency score or the topological distance score, and the weighted result is used as the sub-score corresponding to the spatial association quality. Then, the high-frequency data fields and low-frequency background data fields with business associations in the fused sample are analyzed item by item. Perform consistency checks; for unresolved defect records, use rule mapping to associate defect types with temperature, current, active power, voltage, switch status, or alarm fields, and use Z-score anomaly detection to identify abnormal fluctuations in continuous numerical high-frequency data. Use event rule matching to identify switch actions, protection actions, and alarm events. When the anomaly type and direction of change are consistent with the defect record, the corresponding check item is recorded as consistent. For maintenance records, use time interval status matching to extract high-frequency data during or after the maintenance period. When the equipment status corresponding to shutdown, component replacement, parameter adjustment, or power restoration is consistent with the maintenance content, the corresponding check item is recorded as consistent.For inspection logs, word segmentation, state dictionary matching, and negative word recognition methods are used to convert the log text into standard equipment states. The most recent time matching method is used to select high-frequency data within the dynamic time window corresponding to the inspection time, and this high-frequency data is converted into standard equipment states. When the two standard equipment states are consistent, the corresponding inspection item is recorded as consistent. Records that cannot be compared are not counted as valid inspection items. The ratio of the number of consistent inspection items to the total number of valid inspection items is used as the sub-item score for fusion consistency. The sub-item scores of the four dimensions are substituted into the following formula for weighted calculation: in, For the first A fusion sample at time... Quality rating; The sequence number of the fused sample; This refers to the corresponding time points of the fused samples on a unified time axis; Sub-scores corresponding to the quality of the data source; Sub-scores for time alignment quality; Sub-item scores corresponding to spatial correlation quality; The sub-scores for fusion consistency; The weighting coefficient corresponding to the quality of the data source; The weighting coefficients corresponding to time alignment quality; These are the weighting coefficients corresponding to the spatial correlation quality; The weighting coefficients are defined for consistency. Each weighting coefficient is configured based on the degree of attention given to the four dimensions in the distribution network dispatching business scenario. The sum of the four sub-item scores multiplied by their corresponding weighting coefficients is used as the quality score of the fused sample. Next, the quality score is compared with a preset quality threshold using the following formula: in, For the first A fusion sample at time... Quality rating; This refers to the sequence number of the fused sample; This refers to the corresponding time points of the fused samples on a unified time axis; A preset quality threshold is set. When the quality score is not lower than the preset quality threshold, the current quality score of the fused sample is retained. When the quality score is lower than the preset quality threshold, the corresponding sub-item score is subtracted from the respective thresholds of data source quality, time alignment quality, spatial correlation quality, and fusion consistency. The difference is then divided by the corresponding threshold, and the result is used as the relative difference between the corresponding sub-item score and its respective threshold. When only one sub-item score is lower than its respective threshold, the quality dimension corresponding to that sub-item score is used as the feedback direction. When multiple sub-item scores are lower than their respective thresholds, the relative differences are compared, and the quality dimension with the largest relative difference is used as the feedback direction. When multiple relative differences are the same, the feedback direction is selected in the order of data source quality, time alignment quality, spatial correlation quality, and fusion consistency. Finally, when the feedback direction is data source quality, the data integrity rate, packet loss rate, and outlier ratio are read. When the data integrity rate is lower than the preset integrity rate threshold, supplementary data collection is triggered. When the packet loss rate is higher than the corresponding threshold and the heterogeneous data source has caching or retransmission capabilities, retransmission is triggered. When the outlier ratio is higher than the corresponding threshold, the fusion weight of the corresponding heterogeneous data source is reduced. When consecutive missing tests or short-term anomalies occur, the fusion weight is increased. Prioritize short-term acquisition of heterogeneous data sources; when the feedback direction is time alignment quality, adjust the dynamic time window width when the time deviation exceeds the preset deviation threshold or the effective matching ratio is lower than the preset ratio, and re-execute timestamp matching and alignment using the adjusted dynamic time window; when the feedback direction is spatial correlation quality, re-examine the device identifier mapping results corresponding to high-frequency data and low-frequency background data, and check whether the corresponding device nodes fall within the topological neighborhood range; when the feedback direction is fusion consistency, reduce the fusion weight of low-frequency background data that has conflicts, is invalid, or has a low data quality score, and regenerate fusion samples according to the adjusted fusion weight; after each feedback processing action is completed, recalculate the four sub-scores and quality score of the fusion sample. If the recalculated quality score is still lower than the preset quality threshold, continue to execute the feedback processing action corresponding to the next relative difference. If the preset quality threshold is still not reached after completing each feedback processing action, mark the fusion sample as a sample to be reviewed; after the feedback processing action is completed, read the fusion samples formed by heterogeneous data sources within the time window with the current update time as the end time, and construct the fusion sample set according to the following formula: in, heterogeneous data source The set of fusion samples formed within the time window; Identifier for heterogeneous data sources; This is the current update time for the reliability score; The preset time length for the time window; This refers to the start and end time range corresponding to the time window. This refers to the time point corresponding to the fused sample within the time window; For a moment The corresponding number One fusion sample; The vertical line represents the sequence number of the fused samples within the time window; the condition to the right of the vertical line indicates that only samples from heterogeneous data sources will be included. The fused samples of the data are included in the fused sample set; then the average fusion quality score within the time window is calculated according to the following formula: in, heterogeneous data source Average fusion quality score within the time window; Identifier for heterogeneous data sources; This is the current update time for the reliability score; heterogeneous data source The set of fusion samples formed within the time window; This represents the number of fused samples contained in the fused sample set. For a moment The corresponding number One fusion sample; For the first A fusion sample at time... Quality rating; The sequence number of the fusion sample participating in the summation; This refers to the time point corresponding to the fused sample within the time window; This involves summing the quality scores of each fused sample within the fused sample set; dividing the sum of quality scores by the number of fused samples yields the average fused quality score within the time window; finally, the reliability score of the heterogeneous data source is updated according to the following formula: in, heterogeneous data source Reliability score at the next update time; Identifier for heterogeneous data sources; The next update moment after the current time window evaluation is completed; heterogeneous data source Reliability score at the current update time; This is a smoothing coefficient used to configure the proportion of the current reliability score in the updated results; This represents the proportion of the average fusion quality score in the updated results; heterogeneous data source The average fusion quality score within the time window; the product of the current reliability score and the smoothing coefficient is then combined with the average fusion quality score. The products are added together, and the result is used as the reliability score after the heterogeneous data source is updated.

[0055] Therefore, this application can perform a quality score on the fusion consistency of the fused samples and adjust the reliability score of heterogeneous data sources based on the quality score. Firstly, by standardizing and converting the original data from heterogeneous data sources in the distribution network to generate unified data objects, differences in data format, field names, device identifiers, and data types between different business systems can be eliminated, avoiding difficulties in correlation caused by inconsistent expression of multi-source data. Secondly, by calculating the acquisition control factor based on the state awareness characteristics of heterogeneous data sources and adjusting the acquisition cycle through the acquisition control factor, the data acquisition process can adapt to the operating state of equipment and data changes, reducing redundant acquisition during stable operation and increasing the density of real-time data acquisition during abnormal fluctuations or critical states, thus enhancing the ability of scheduling data to capture changes in equipment state from the data source. Furthermore, by performing timestamp correction and alignment of real-time data through dynamic time windows and combining the topological neighborhood of device nodes in the distribution network equipment topology model to determine the spatial correlation between heterogeneous data sources, the time misalignment problem caused by sampling cycle differences, communication delays, and clock deviations can be alleviated. This approach reduces erroneous device associations caused by relying solely on text field matching, enabling multi-source data under the same operational event to participate in subsequent processing within a more accurate time reference and device topology scope. Furthermore, by filtering low-frequency background data from a unified data object based on spatial correlations and fusing this low-frequency background data with high-frequency data in time-aligned data, low-frequency background information such as equipment ledgers, defect records, maintenance records, and inspection logs can participate in the construction of high-frequency operational status data within an effective correlation range. This avoids simply splicing failure records, irrelevant equipment records, or low-quality background data into the fused sample, thereby enhancing the fused sample's ability to reflect the actual operational status of the distribution network. Finally, by performing a quality score on the consistency of the fused sample and adjusting the reliability scores of heterogeneous data sources based on the quality score, consistency issues in the fused sample can be fed back to the data source evaluation stage. This provides a basis for subsequent acquisition cycle correction, time alignment range adjustment, and data source weight reduction, enabling the distribution network dispatch data processing process to have continuous correction capabilities and improving the credibility of the fused sample when entering the fault assessment, risk warning, and dispatch decision-making stages.

[0056] In summary, the technical solution adopted in this application can achieve spatiotemporal consistency fusion of multi-source heterogeneous distribution network dispatch data in complex dispatch scenarios, thereby improving the reliability of distribution network dispatch data fusion results.

[0057] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0059] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A method for adaptive acquisition and spatiotemporal fusion of multi-source heterogeneous distribution network dispatch data, characterized in that, Includes the following steps: Obtain raw data from heterogeneous data sources in the power distribution network, perform standardized transformation, and generate a unified data object; The acquisition control factor is calculated based on the state awareness characteristics of the heterogeneous data source, and the acquisition cycle is adjusted through the acquisition control factor to obtain real-time data from the heterogeneous data source. The real-time data is timestamped and aligned using a dynamic time window based on the heterogeneous data source to obtain time-aligned data. The spatial relationships between heterogeneous data sources are determined based on the topological neighborhood of device nodes in the distribution network equipment topology model; Based on the spatial correlation, low-frequency background data is filtered out from the unified data object, and the low-frequency background data and high-frequency data in the time-aligned data are fused to obtain a fused sample. The fused samples are scored for fusion consistency, and the reliability scores of the heterogeneous data sources are adjusted based on the quality scores.

2. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, Acquiring raw data from heterogeneous data sources in the power distribution network and performing standardized transformation to generate a unified data object specifically includes: Obtain raw data from heterogeneous data sources in the power distribution network; Convert the field names, types, and units of the original data into unified standard fields, standard types, and standard units; Records with field conflicts, missing key data, inconsistent units, and out-of-bounds values ​​after conversion are cleaned, repaired, and quality tags are generated to obtain a unified data object.

3. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, The calculation of acquisition control factors based on the state-aware characteristics of heterogeneous data sources specifically includes: Obtain device type score, topology location score, load importance score, and historical defect score for devices corresponding to heterogeneous data sources; The degree of data variability is calculated based on observations from heterogeneous data sources and statistical characteristics within a historical sliding window; The service importance of the equipment is calculated based on the equipment type score, the topology location score, the load importance score, and the historical defect score. Determine the low-frequency data update status of the corresponding devices of the heterogeneous data source and the event triggering intensity of the heterogeneous data source; The data change rate, the device service importance, the low-frequency data update status, and the event trigger intensity are used as state awareness features of heterogeneous data sources. They are then weighted and summed according to preset weight coefficients to obtain the acquisition control factor.

4. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, Adjusting the acquisition cycle and acquiring real-time data from heterogeneous data sources through the acquisition control factor specifically includes: Obtain the basic collection cycle corresponding to heterogeneous data sources; The actual acquisition cycle of the heterogeneous data source is calculated based on the basic acquisition cycle and the acquisition control factor. Real-time data from heterogeneous data sources is acquired according to the stated actual acquisition cycle.

5. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, The real-time data is timestamped and aligned using a dynamic time window based on the heterogeneous data source, resulting in time-aligned data including: A linear correction relationship is established based on the historical timestamp sequence of heterogeneous data sources, and the original timestamps of the real-time data are fundamentally corrected based on the linear correction relationship. Determine the reliability score of heterogeneous data sources; A dynamic time window for heterogeneous data sources is generated based on the communication latency fluctuations and data types of the heterogeneous data sources, as well as the reliability score. The dynamic time window is used to perform timestamp matching and alignment on the real-time data after basic correction to obtain time-aligned data.

6. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, Determining the spatial relationships between heterogeneous data sources based on the topological neighborhood of device nodes in the distribution network equipment topology model specifically includes: Map the device identifiers corresponding to each data point in the time-aligned data to the device nodes in the distribution network topology model; The topological neighborhood of a device node is determined based on its feeder connection relationship, transformer substation affiliation relationship, and upstream and downstream adjacency relationship in the distribution network topology model. The spatial association between heterogeneous data sources is determined based on the overlapping and adjacency relationships between the topological neighborhoods of the corresponding device nodes.

7. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 6, characterized in that, The topological neighborhood of a device node is determined based on its feeder connection relationship, transformer substation affiliation relationship, and upstream / downstream adjacency relationship in the distribution network topology model. Specifically, this includes: Construct an electrical connection diagram between equipment nodes based on their feeder connection relationships, transformer substation affiliation relationships, and upstream and downstream adjacency relationships in the distribution network topology model; Select a device node as the selected device node. In the electrical connection diagram, extend a preset topological distance outward along the connection path with the selected device node as the center. All device nodes falling within the preset topological distance are determined as the topological neighborhood of the selected device node.

8. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, Filtering low-frequency background data from the unified data object based on the spatial correlation specifically includes: Extract high-frequency data from the time-aligned data; Determine the spatial candidate set of high-frequency data based on the spatial correlation; Low-frequency background data that fall into the spatial candidate set are extracted from the unified data object.

9. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, The fusion of the low-frequency background data and the high-frequency data in the time-aligned data to obtain the fused sample specifically includes: Candidate fusion data with an effective time interval covering the current moment are selected from the low-frequency background data; The fusion weights are generated based on the data quality score, time freshness, topological distance, and source reliability of the candidate fusion data. The candidate fusion data and the high-frequency data in the time-aligned data are weighted and fused according to the fusion weight, and the samples are combined to obtain fused samples.

10. The adaptive acquisition and spatiotemporal fusion method for multi-source heterogeneous distribution network dispatch data as described in claim 1, characterized in that, The process of scoring the fusion consistency of the fused samples and adjusting the reliability score of the heterogeneous data source based on the quality score specifically includes: The quality score of the fused sample is calculated by weighting the scores of each of the four dimensions: data source quality, time alignment quality, spatial correlation quality, and fusion consistency. The quality score is compared with a preset quality threshold. If it is lower than the preset quality threshold, the feedback direction is determined based on the relative difference between each sub-score and its respective threshold. The feedback processing action is executed according to the feedback direction, and after the feedback processing action is executed, the reliability score of the heterogeneous data source is updated based on the average fusion quality score within the time window.

Images