An edge-computing-oriented power meter multi-source data fusion processing system and method

Abstract

The application provides an electric energy meter multi-source data fusion processing system and method for edge computing, and applies to the technical field of electric energy meter data processing. Based on an embedded CPU+AI accelerator card heterogeneous architecture, a multi-source data acquisition unit on the edge side is built, a plurality of communication interfaces and power communication protocols are used to access a plurality of types of terminals such as electric energy meters, energy storage units and charging piles, heterogeneous protocol analysis and multi-sampling frequency data alignment are completed, and power data acquisition, breakpoint continuation and local caching are realized. A full decoupling architecture of metering and management is adopted, a metering unit independently outputs legal metering results, and a management unit completes data preprocessing, classified storage and abnormal marking. Relying on a multi-dimensional data fusion algorithm, load fitting and state analysis are carried out, and computing resources are scheduled according to hardware priority; data trusted evidence is realized by double partition storage and hash tree, and uploaded to the cloud according to data change strategy, and finally cross-terminal data fusion, abnormal diagnosis and load analysis are completed.

Description

Technical Field

[0001] This invention relates to the field of electricity meter data processing technology, and in particular to an electricity meter multi-source data fusion processing system and method oriented towards edge computing. Background Technology

[0002] In the current power Internet of Things (IoT) and edge computing scenarios, a large number of terminals of various types, such as electricity meters, energy storage units, and charging piles, are connected. The data sources are scattered, the communication protocols are diverse, and the sampling frequencies are inconsistent, which brings difficulties to the unified processing of data in the distribution area. Existing acquisition systems usually use a single architecture to process metering data and management data, failing to achieve effective decoupling between legal metering and operation management. This makes them susceptible to interference and tampering of metering data, and makes it difficult to meet the independence and security requirements of legal metering in the power industry.

[0003] At the data processing level, traditional solutions mostly focus on cloud computing, with the edge only performing simple forwarding. This results in problems such as large data transmission volumes, insufficient real-time performance, and inability to process data offline. Furthermore, the lack of a unified protocol parsing and timing alignment mechanism for multi-source data makes it difficult to effectively correlate and integrate data from different devices, leading to low accuracy in load analysis and anomaly diagnosis.

[0004] In terms of storage and security, existing edge storage is not managed according to data importance, metering data and management data are stored together, and there is a lack of anti-tampering and reliable evidence storage methods; data uploading mostly adopts the full synchronization method, which consumes a lot of communication resources and is inefficient.

[0005] Furthermore, the lack of priority strategies for edge node hardware resource scheduling makes it difficult to guarantee processing efficiency in high-concurrency computing scenarios such as data fusion and load analysis. Overall, there is a lack of a complete multi-source data fusion and processing system covering "terminal access, protocol parsing, metering decoupling, edge computing, trusted storage, and lightweight cloud deployment," which fails to meet the efficient, reliable, and secure operation and management requirements of intelligent distribution transformer areas.

[0006] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] According to one aspect of this application, a method for multi-source data fusion processing of electricity meters oriented towards edge computing is provided, comprising: constructing an edge-side multi-source data acquisition unit based on an embedded CPU+AI accelerator card heterogeneous architecture and RS485 / Ethernet / NB-IoT / LoRa interfaces, adapting to DL / T645 / DL / T698 / Modbus-RTU / IEC104 / MQTT protocols, accessing multiple types of terminal devices including electricity meters, energy storage units, and charging piles, and completing heterogeneous protocol parsing and multi-sampling frequency data alignment; extracting electricity meter voltage, current, energy consumption, and distribution terminal switch status data through a protocol parsing engine, supporting breakpoint resume and local caching, storing data to local flash memory in the event of network outage, and automatically re-transmitting after network recovery; adopting a fully decoupled architecture design for metering and management, with the metering unit independently completing legal metering data acquisition and outputting tamper-proof metering results; and the management unit through... Edge computing nodes perform data preprocessing, classifying and storing data according to data type, filtering out abnormal data based on preset thresholds and marking them with feature tags; employing multi-dimensional data fusion algorithms, combined with the physical location and operational data of nearby devices, they calculate the load fluctuation impact factor and perform management-side load fitting and status correction, scheduling computing resources according to the priority order of AI accelerator card > DSP > CPU; deploying a dual-partition storage architecture on edge nodes to separate read and write operations, allocating storage resources hierarchically according to data importance, generating hash values ​​for metering records through hash algorithms, constructing a hash tree, and uploading the root node as a verification value to the blockchain network; selecting full synchronization or incremental push mode based on data change frequency and correlation threshold, transmitting only changed data and verification information to the cloud platform, and synchronously updating the local causal relationship graph; performing cross-terminal data correlation fusion, anomaly diagnosis and judgment, and load characteristic analysis, generating mixed output data in a unified format.

[0008] Another aspect of this application discloses a multi-source data fusion processing system for electricity meters oriented towards edge computing, comprising: an edge heterogeneous acquisition unit, used to build acquisition hardware based on an embedded CPU+AI accelerator card heterogeneous architecture and multiple types of interfaces such as RS485 / Ethernet / NB-IoT / LoRa, adapting to DL / T645 / DL / T698 / Modbus-RTU / IEC104 / MQTT power communication protocols, connecting to multiple types of terminal devices such as electricity meters, energy storage units, and charging piles, and completing heterogeneous protocol parsing and multi-sampling frequency data alignment processing; a protocol parsing and caching unit, used to extract voltage, current, power consumption, and distribution terminal switch status operation data of electricity meters through a protocol parsing engine, providing breakpoint resume and local caching capabilities, writing data to local flash memory in the event of a network outage, and automatically performing retransmission operations after the network is restored; and a metering and management decoupling unit, used to adopt a fully decoupled architecture of metering and management, whereby the metering unit independently completes the acquisition of legal metering data and outputs tamper-proof metering results, and the management unit executes the data on the edge computing node. The system includes: a data preprocessing unit, a data categorization and storage unit, and a data filtering and feature labeling unit based on preset thresholds; a multi-dimensional data fusion unit, which uses multi-dimensional data fusion algorithms to combine physical location with the operating data of nearby devices, calculates the load fluctuation impact factor, performs management-side load fitting and status correction, and schedules computing resources according to the hardware priority of AI accelerator card > DSP > CPU; a dual-partition trusted storage unit, which deploys a dual-partition storage architecture at edge nodes to achieve read / write operation separation, allocates storage resources according to data importance, generates hash values ​​for metering records using hash algorithms, constructs a hash tree, and uploads the root node as a verification value to the blockchain network; a cloud synchronization update unit, which selects full synchronization or incremental push mode based on data change frequency and correlation thresholds, transmitting only changed data and verification information to the cloud platform, and synchronously updating local causal relationship graph data; and a cross-terminal collaborative analysis unit, which performs cross-terminal data correlation fusion, anomaly diagnosis and judgment, and load characteristic analysis, generating mixed output data in a unified format.

[0009] According to another aspect of this application, an electronic device includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-described edge computing-oriented multi-source data fusion processing method for electricity meters by executing the executable instructions.

[0010] According to another aspect of this application, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a second processor, implements the above-described edge computing-oriented multi-source data fusion processing method for electricity meters.

[0011] The beneficial technical effects of this application are as follows: Based on an embedded CPU + AI accelerator card heterogeneous architecture, an edge-side multi-source data acquisition unit is constructed. It connects to terminals such as electricity meters, energy storage, and charging piles through various communication interfaces and power protocols, completing heterogeneous protocol parsing and multi-sampling frequency data alignment, achieving data acquisition, breakpoint resumption, and local caching. The system adopts a fully decoupled architecture for metering and management. The metering unit independently outputs legal metering results, while the management unit completes data preprocessing, classification storage, and anomaly marking. It relies on multi-dimensional data fusion algorithms to achieve load fitting and state analysis, scheduling computing resources according to the priority order of AI accelerator card > DSP > CPU. It uses dual-partition read-write isolated storage and hash trees and blockchain to achieve trusted storage of metering data. It adaptively selects full synchronization or incremental push upload to the cloud based on the data change frequency. Finally, it completes cross-terminal data fusion, anomaly diagnosis, and load analysis, outputting standardized data, suitable for edge computing and multi-source data collaborative processing scenarios in smart distribution substations.

[0012] This application decouples legal metrology from management analysis, enabling independent acquisition and read-only storage of metrological data to prevent tampering and interference, thus meeting legal metrology requirements. At the edge, protocol parsing, data cleaning, fusion computing, and anomaly diagnosis are completed, reducing cloud load and improving real-time and offline processing capabilities. Computing resources are scheduled according to hardware priority, adapting to different edge hardware and improving data fusion and load analysis efficiency. Dual-partition storage, hash verification, and blockchain notarization ensure data integrity and traceability. Incremental push reduces communication overhead, and breakpoint resume ensures no data loss, improving system stability. Unified access, parsing, and output support multi-device collaboration and intelligent management of distribution areas.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0014] Figure 1 The flowchart illustrates a multi-source data fusion processing method for electricity meters oriented towards edge computing, provided in an embodiment of this application. Figure 2 This illustration shows a schematic diagram of the structure of a multi-source data fusion processing system for electricity meters oriented towards edge computing, provided in an embodiment of this application. Detailed Implementation

[0015] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0016] In one embodiment, this application also proposes a multi-source data fusion processing method for electricity meters oriented towards edge computing. Figure 1A schematic flowchart of a multi-source data fusion processing method for electricity meters oriented towards edge computing, according to an embodiment of this application, is shown.

[0017] The S101 is an edge-side multi-source data acquisition unit built on an embedded CPU+AI accelerator card heterogeneous architecture and RS485 / Ethernet / NB-IoT / LoRa interfaces. It is compatible with DL / T645 / DL / T698 / Modbus-RTU / IEC104 / MQTT protocols and can connect to various types of terminal devices, including electricity meters, energy storage units, and charging piles, to complete heterogeneous protocol parsing and multi-sampling frequency data alignment.

[0018] In one implementation, combining the unified management requirements of edge computing power terminal assets with national geospatial positioning rules, a two-dimensional matching mechanism of cyberspace assets and geographic elements is introduced. This mechanism constructs a hierarchical mapping model adapted to power IoT terminals such as electricity meters, energy storage units, charging piles, and edge computing gateways, achieving a stable association between network objects and geographic space. This mechanism is applicable to the fields of power IoT terminal data acquisition, edge computing data processing, power information management, and distribution automation, and meets the requirements for the research, development, application, and implementation of power metering and distribution terminal related products.

[0019] Based on the network space asset information collected by edge computing nodes, network space object data such as device number, network address, communication port, connection relationship, and protocol type of terminals such as electricity meters, energy storage units, and charging piles are extracted; at the same time, geographic space boundary data such as distribution area, transformer area number, site geographic boundary, line topology, installation location, and distribution transformer affiliation are collected to form a basic dataset to be matched that is uniform in format, traceable in source, and meets the power system data specifications.

[0020] A precise string matching algorithm is employed, using the power system standard transformer area code, site code, equipment asset code, and country / region code as unique matching keys to perform a one-to-one direct mapping between cyberspace assets and geospatial elements. The matching process does not alter the original data or affect legal metrology results, maintaining consistency with the fully decoupled architecture of metrology management. A fuzzy matching algorithm based on edit distance is used to calculate the similarity of strings such as country / region names, standard names, abbreviations, and English variants. A similarity threshold is set to achieve fault-tolerant matching, improving the matching success rate in complex naming and multi-alias scenarios, and ensuring the stability and applicability of geographic mapping across multiple scenarios. A topology association verification algorithm is used to verify the consistency of matched results based on network connection relationships such as distribution line affiliation, communication link connections, power supply range, and terminal topology level, correcting isolated erroneous matches and ensuring that the geospatial mapping is consistent with the actual physical topology, providing accurate location information for subsequent data processing.

[0021] The final mapping relationship, after precise matching, fuzzy matching, and topology verification, is written into the local causal relationship graph, which solidifies the association structure of network address, device number, transformer area code, geographical boundary, and topology affiliation, forming stable and reusable network space asset-geographic element association data. This provides a unified location basis for subsequent cross-terminal data fusion, load status analysis, anomaly diagnosis, and reliable data storage, and does not intervene in the metering unit or modify the legal metering results throughout the process, which meets the architectural design requirements of complete decoupling of metering and management.

[0022] The edge computing gateway hardware architecture is as follows: The main controller uses a quad-core ARM Cortex-A53 embedded CPU, equipped with two TOPSN PUA accelerator cards and a 32-bit floating-point DSP coprocessor; it has 8GB of local NAND flash memory, with a 6GB read-only metering partition and a 2GB read-write management partition; the communication interface electrical specifications are as follows: RS485: fixed baud rate of 9600, differential level transmission, a maximum of 32 electricity meters can be connected to a single bus. Ethernet: 100M adaptive, independent VLAN isolation for metering services. NB-IoT / LoRa: uplink peak speed of 50kbps, LoRa communication distance ≤2km. The maximum concurrent access terminal limit for a single edge gateway is: 64 electricity meters + 3 energy storage devices + 8 charging piles, and the excess will be automatically distributed to adjacent edge nodes.

[0023] The unified reference time granularity is fixed at 5 minutes; the sampling difference judgment rule is: if the difference between the device sampling interval and the reference is greater than 1 minute, it is judged as asynchronous data; the time alignment interpolation rule is as follows: input the reference time axis T[0,5,10...min], the device's original sampling array D(t,val), traverse each reference time point Ti; find the two most recent original sampling data D_prev and D_next before and after Ti; time difference Δt=Ti-t_prev, total interval ΔT=t_next-t_prev; alignment value val_i=val_prev+(val_next-val_prev) Δt / ΔT; if only one-sided sampling data exists, the nearest value is used for filling. For high-frequency equipment (charging pile sampling every 1 second), the average is automatically downsampled to a 5-minute granularity. For low-frequency energy storage equipment, linear interpolation is used to fill in data during periods when there is no data.

[0024] Fuzzy matching is considered a successful match if the edit distance similarity threshold is ≥0.8; if the similarity is 0.5~0.8, it enters the manual review queue; if it is <0.5, it is discarded directly; if the equipment area code and line number do not match, it is considered an isolated erroneous mapping, the association is automatically deleted, and an exception log is written.

[0025] Time-sharing polling scheduling is adopted: RS485 electricity meter cycle is 10s, Ethernet energy storage is 30s, and NB-IoT charging pile is 60s; when multiple messages arrive at the same time, they are queued according to the priority of "metering data > energy storage > charging pile", conflicting messages are buffered for 1s and then retransmitted, and three consecutive conflicts mark the terminal communication as abnormal.

[0026] S102 extracts data on electricity meter voltage, current, electricity consumption, and power distribution terminal switch status through a protocol parsing engine. It supports breakpoint resume and local caching. In the event of a network outage, data is stored in local flash memory and automatically retransmitted after the network is restored.

[0027] In one implementation, combining the multi-source data acquisition requirements of edge computing energy meters with the power communication protocol parsing rules, a protocol frame parsing algorithm, a breakpoint resume timing verification algorithm, and a local flash memory cache management mechanism are introduced to achieve stable extraction, offline caching, and automatic retransmission of energy meter and distribution terminal operating data after network recovery. This is applicable to implementation scenarios related to power IoT terminal data acquisition, edge computing gateway data processing, and power information acquisition and communication transmission, and meets the technical implementation requirements of power metering terminals and data acquisition devices. Raw data frames of DL / T645, DL / T698, Modbus-RTU, IEC104, and MQTT protocols are received through communication interfaces such as RS485, Ethernet, NB-IoT, and LoRa. Protocol frame header verification algorithms and CRC cyclic redundancy check algorithms are used to verify the integrity and legality of the data frames. Error frames and interference frames with mismatched frame header identifiers, incorrect check codes, or abnormal data lengths are directly filtered out, and only legal and valid data frames are retained for subsequent parsing processes.

[0028] The protocol parsing engine is activated, and the legal data frames are parsed field by field using a field location and extraction algorithm according to the data format and field structure specified in the corresponding power communication protocol. This accurately extracts key operational data such as voltage, current, power, power output, distribution terminal switch status, and acquisition timestamp, ensuring that the parsing results are completely consistent with the protocol standard and providing a standard data source for subsequent data processing. The parsed valid operational data is written to the memory cache in chronological order of acquisition time. A time-series breakpoint marking algorithm is used to add a globally unique sequence number, acquisition timestamp, and cache physical location marker to each data entry, forming a continuous and traceable cached data chain. This provides precise location information for resuming interrupted transmissions, ensuring that the data sequence is not disordered and the location is traceable.

[0029] When an edge node detects a network disconnection or communication link anomaly, it automatically triggers an offline storage switchover mechanism. All data in the memory cache that has not yet been uploaded to the cloud, along with breakpoint marker information, is synchronously written to the edge node's local flash memory and stored in partitions according to the acquisition time sequence, ensuring that data is not lost or overwritten during network outages. Once the network is reconnected or communication returns to normal, a breakpoint comparison algorithm reads the breakpoint sequence number stored in the local flash memory and compares it with the latest received sequence number returned by the cloud platform. Based on the sequence number difference, the range of unuploaded data is determined, and the unuploaded data in the flash memory is automatically uploaded to the cloud platform sequentially, starting from the breakpoint location.

[0030] After the retransmission is completed, the cloud platform verifies the integrity of the received data and returns a confirmation command. The edge node clears the successfully uploaded data from the flash memory based on the confirmation information, and at the same time updates the data upload status in the local causal relationship graph, completing the entire closed loop of data collection, protocol parsing, real-time caching, offline storage, and breakpoint resumption, ensuring that the data is uploaded completely and reliably.

[0031] The frame verification rules are as follows: power frames use CRC16-Modbus polynomial 0xA001; frames with a length <10 bytes or >256 bytes are directly judged as interference frames and discarded; frames with a checksum that does not match the locally calculated value are marked as abnormal frames and stored separately in the fault log partition.

[0032] The breakpoint sequence number rules are as follows: encoding format: station area number-device type-timestamp (ms), globally unique; local flash memory offline cache for a maximum of 7 days of data, older data beyond 7 days is automatically overwritten and a hash verification backup is retained. Breakpoint retransmission interaction specifications: cloud receipt JSON message sample is as follows: {"last_seq":"TQ001-M-1782201360000","ack":1,"miss_seq_list":[]}. Missing sequence numbers in the miss_seq_list are read at the edge and retransmitted in batches, with a maximum of 200 records uploaded at a time; during retransmission, the hash of the current data is compared, and if the hashes match, duplicate data is skipped to avoid duplicate entries into the cloud database.

[0033] S103 adopts a fully decoupled architecture design for measurement and management. The measurement unit independently completes the collection of legal measurement data and outputs tamper-proof measurement results; the management unit performs data preprocessing through edge computing nodes, classifies and stores data according to data type, and filters abnormal data based on preset thresholds and marks them with feature tags.

[0034] In one implementation, considering the decoupling requirements of metering management in the multi-source data fusion processing of edge computing energy meters, an independent legal metering acquisition mechanism and edge-side preprocessing rules are introduced to determine the division of labor for the acquisition and preprocessing of multi-source data from energy meters. Based on the actual needs of multi-source data fusion processing of edge computing energy meters, this solution adopts a design concept that legal metering data and management analysis data are independent and do not interfere with each other, applicable to the research and development and application of power Internet of Things, electricity consumption information collection, smart metering terminals, and distribution automation related systems. Through data classification rules and processing link isolation rules, the collected data is clearly divided, and the processing subject and execution process are determined separately, achieving complete physical and logical separation between metering and management functions. For example, electrical quantity data such as electricity, voltage, and current used for electricity billing are classified as metering data and processed independently by a dedicated metering unit; non-settlement data used for status monitoring and operation analysis are classified as management data and processed by a management unit, ensuring that the two types of data do not interfere with each other.

[0035] In accordance with relevant standards for electricity metering and data security management requirements, data collected by electricity meters for legal metering settlement, such as voltage, current, active power, and reactive power, are designated as legal metering data. These data are processed by independent metering units to ensure the independence of the metering process and the immutability of the results. For example, phase A voltage, phase B current, total forward active power, and reverse reactive power are all legal metering data, collected and calculated solely by the metering unit; the management unit does not participate in the calculations.

[0036] Data such as switch status, equipment operating status, communication status, and environmental status, which are not involved in settlement, are designated as management analysis data. These are received, preprocessed, and analyzed by the management unit within the edge computing node. This clear division of data types and processing flows establishes a clear boundary between the metering and management processes, ensuring independence and standardized data processing for legal metering. For example, circuit breaker open / close status, terminal online / offline status, and equipment ambient temperature are all considered management analysis data. The management unit performs data cleaning, transformation, and feature extraction without affecting the metering process. Through this division of labor, legal metering data is processed in a closed loop within an independent metering unit, while management analysis data is processed within the management unit. The two processes interact only in read-only mode, without data modification or process intervention, fully meeting the system design requirements for complete decoupling of metering and management.

[0037] The method employs a task-based approach to collect data on electricity meter voltage, current, energy consumption, and distribution terminal switch status. The metering unit independently collects legal metering data and outputs the results, which cannot be modified by the management unit. The management unit receives non-metering data. For data such as electricity meter voltage, current, energy consumption, and distribution terminal switch status, this method categorizes the data according to preset task-based rules, dividing it into legal metering data and non-metering management data. These are processed independently by their respective units. Legal metering data, such as voltage, current, active power, and reactive power, are collected, calculated, and output independently by a dedicated metering unit. The management unit has only read-only access and cannot rewrite, modify, or replace the metering results. Non-metering data, such as distribution terminal switch status and equipment operating status, are directly received and processed by the management unit of the edge computing node and do not enter the metering unit's calculation process.

[0038] According to the above division of labor, the metering unit independently collects and calculates the electrical signals from the electricity meter, generates legally mandated metering results, and outputs them in a fixed format. These results serve as the basis for electricity settlement and data storage, and are tamper-proof. For example, the metering unit completes metering calculations based on the collected voltage and current signals, generating a positive active power metering result. Once output, this result remains fixed; the management unit can read it for subsequent analysis, but cannot modify the value. Simultaneously, the management unit receives non-metering data such as switch closing and opening status in real time for status monitoring and anomaly detection. The processing flows for these two types of data are independent and do not interfere with each other.

[0039] A fully decoupled hardware and software architecture for metering management is constructed, comprising a metering acquisition module, edge computing nodes, categorized storage units, and anomaly screening and marking components. Through physical resource partitioning and logical process isolation, separate metering processing and management processing links are defined, allowing for independent configuration, execution, and storage. In the architecture design, the metering acquisition module specifically serves legal metering data processing, only collecting and calculating metering data such as voltage, current, and electricity, without performing any management or analysis operations. The management unit within the edge computing nodes is dedicated to non-metering data processing, data preprocessing, anomaly detection, and load analysis, without intervening in the metering acquisition and calculation stages. The categorized storage units employ partitioning to store metering data and management data separately, preventing tampering or interference caused by mixed data storage. The anomaly screening and marking component operates only within the management link, marking anomalies in management data and read metering results without altering the original metering values.

[0040] Through the above architecture and link setup, the measurement link and the management link operate completely independently. The measurement link focuses entirely on the collection and output of legal measurement data, while the management link focuses on data reception, cleaning, transformation, analysis, and anomaly marking. Only one-way read-only data interaction is allowed between the two links, and there is no control or rewriting relationship, ensuring that the measurement process is independent and the data is safe and reliable.

[0041] The legally mandated metering data source is input into the independent metering unit, completing high-precision data acquisition and output without participating in the management unit's calculations. The legally mandated metering data source output from the electricity meter is connected to the independent metering unit, which independently completes the electrical signal acquisition, metering calculations, and result output. The entire process does not involve any computational interaction with the management unit, nor does it participate in any data processing flow on the management side. The independent metering unit only processes legally mandated metering-related raw signals such as voltage, current, active power, and reactive power. It performs high-precision sampling, error correction, and metering settlement calculations according to electricity metering standards, outputting standard data that meets legal metering requirements. For example, raw analog or digital signals such as phase A voltage and phase B current output from the electricity meter are input into the independent metering unit. The metering unit completes the power accumulation and calculation according to a predetermined metering algorithm, directly generating legally mandated metering results that can be used for settlement.

[0042] The measurement results are used only for local trusted storage and data upload, and do not enter the preprocessing, feature extraction, load analysis and other calculation processes of the management unit. The management unit can only read the measurement results in read-only mode for display and correlation analysis, and must not interfere with the measurement data and calculation process in any way, so as to ensure the independence, impartiality and immutability of legal measurement.

[0043] Non-metering data and the raw metering results output by the metering unit are input to the edge computing node management unit for data cleaning, format conversion, and feature extraction. Non-metering data such as switch status and equipment operating status, along with the raw metering results output by the metering unit, are sent to the edge computing node management unit for unified data preprocessing. This processing does not modify the raw metering results. The management unit first cleans the incoming data, filtering out null values, abnormal jump values, duplicate reports, and other invalid data from switch status data accessed using different protocols and sampling frequencies, retaining only valid and stable status data. For example, invalid data caused by frequent switch status fluctuations is directly discarded by the management unit according to preset rules, retaining only true and valid closing and opening status information.

[0044] After data cleaning, the management unit performs a unified format conversion on all types of data. Non-metering data and metering results from different terminals and following different protocols are converted into a unified data format to ensure the standardization of subsequent data storage and analysis. For example, switch status data in DL / T645 protocol format and Modbus protocol format are uniformly converted into a standard format compatible with the edge computing platform.

[0045] After standardization, the management unit performs feature extraction on the data, extracting operational characteristics such as switch change time, number of changes, and duration from the switch status data. This provides a basis for subsequent anomaly diagnosis and load analysis. The extracted feature data is used only for management analysis and status monitoring, does not participate in legal metrology calculations, and does not change the original metrology results.

[0046] After edge preprocessing, the data is categorized and stored according to data type. Anomaly detection and feature labeling are performed based on preset thresholds, generating data acquisition, preprocessing, storage, and anomaly marking information under a fully decoupled metrology management architecture. Preprocessed data is partitioned and stored according to data type, with legal metrology results and non-metrology operational data stored in separate storage areas. This ensures physical isolation and non-interference between metrology and management data, meeting the requirements of a fully decoupled metrology management architecture. Legal metrology data is stored in a read-only storage area, supporting only reading and not modification; non-metrology data is stored in a read-write storage area for subsequent analysis and querying.

[0047] While storing data, the system performs anomaly detection on the preprocessed data based on preset electrical parameter thresholds and status judgment rules. When data exceeds the normal operating range, it is automatically identified as abnormal data and a corresponding feature tag is added for subsequent operation and maintenance and fault diagnosis. For example, when the system detects that the voltage value exceeds the preset 240V threshold, it is identified as an overvoltage anomaly and the data is labeled "overvoltage anomaly"; when the current exceeds the rated range, it is labeled "overcurrent anomaly"; when the switch status fluctuates frequently, it is labeled "switch fluctuation anomaly".

[0048] By classifying and storing data and marking anomalies, a complete information chain is formed, including data collection, parsing, preprocessing, classification and storage, and anomaly detection. All anomaly markers only apply to the management and analysis process and do not modify or affect the original legal metrology results. Ultimately, standardized, reliable, and traceable data processing results are generated under a fully decoupled metrology management architecture.

[0049] The metering unit and the management unit communicate via an SPI read-only bus, with unidirectional transmission (metering → management) and no reverse write channel. The metering partition is equipped with a hardware write protection lock, which permanently disables the write pin after power-on and can only be unlocked to write basic parameters during the production verification stage. The communication message is accompanied by a metering verification MAC code, and if the management unit tampers with the message, it will be directly discarded by the metering core.

[0050] The electrical anomaly quantification thresholds and classification labels are as follows: Level 1 Alarm (Electricity Theft / Severe Fault): Voltage <176V or >264V, hourly power consumption sudden change ±80%, reverse power consumption >30% of forward power, labeled [High-Risk Electricity Theft / Loss of Voltage]. Level 2 Alarm (Equipment Fault): Current exceeds 120% of rated current, phase loss, labeled [Overload / Phase Loss]. Level 3 Alarm (Minor Fluctuations): Voltage fluctuation within ±10%, only recorded, not pushed to maintenance personnel.

[0051] The metering unit has a built-in fixed error correction coefficient, which is fixed and stored in read-only memory after factory verification; a metering verification log is automatically generated every 24 hours and stored in the read-only partition, and the log cannot be deleted; the original metering data uses the national cryptographic SM3 to generate a verification fingerprint, which meets the requirements for metering traceability and can be used as legal evidence in electricity bill disputes.

[0052] The partitioned storage balancing configuration is as follows: Total flash memory 8GB: 6GB metered read-only (circular erase / write balancing algorithm, evenly distributing the number of erase / write operations to each block to avoid local aging); 2GB management read / write, with automatic cleanup of non-metered logs older than 30 days when the usage exceeds 80%.

[0053] S104 uses a multi-dimensional data fusion algorithm to combine the physical location and the operating data of nearby devices to calculate the load fluctuation impact factor and carry out management-side load fitting and status correction, and schedules computing resources according to the priority of AI accelerator card > DSP > CPU.

[0054] In one implementation, combining the multi-source data fusion processing requirements of edge computing energy meters with the characteristics of multi-dimensional data fusion algorithms, physical location association rules and load fluctuation analysis mechanisms are introduced to build a management-side data fusion and load status analysis process adapted to energy meter metering scenarios. This process operates independently within the management unit of the edge computing node, solely for power grid operation status analysis and load management, without participating in legal metering-related calculations or affecting metering results. Following the management-side analysis process construction rules, functions such as multi-source data fusion, load status fitting, and equipment operation diagnosis are uniformly delegated to the management unit for execution. The entire processing is completely isolated from the metering unit and does not modify the original legal metering data, thus forming a collaborative analysis and processing process compatible with various power terminals such as energy meters, energy storage units, and charging piles.

[0055] This process analyzes main energy meters, adjacent similar metering devices, charging piles, energy storage units, and other load devices within the same power distribution area. It establishes a load status analysis mechanism solely for management-side operational analysis, without accessing or interfering with the internal data collection and calculation processes of the metering units, and without affecting legally mandated metering results, ensuring the independence and legality of the metering process. For example, within the same power distribution area, the management unit only collects and analyzes the external output data of energy meters, energy storage charging and discharging data, and charging pile electricity consumption data. Through multi-dimensional data fusion, it determines the regional load status. All analysis is completed within the management chain, independent of and without interference from the legally mandated metering processes of the metering units.

[0056] In accordance with the multi-source data fusion specification for electricity meters, physical location association matching is performed on voltage, current, electricity consumption, and the operating data of nearby equipment to complete the data association preprocessing before fusion. This ensures that data from different devices within the same area can be accurately matched and collaboratively analyzed. A physical location topology association algorithm is used, with transformer area number, distribution line number, and installation geographic information as matching criteria, to uniformly associate and bind data from various devices within the same physical area, forming a complete regional data set from dispersed data collection.

[0057] Specifically, the system first reads the corresponding transformer substation number, its associated power distribution line, and installation location information for each device. Then, it groups the operating data of devices belonging to the same substation, such as electricity meters, energy storage units, and charging piles, together to complete the matching of data from the same source area. For example, the system integrates the voltage, current, power, power, and operating status data of the main electricity meter, adjacent electricity meters, energy storage units, and charging piles belonging to the same substation according to their physical location, forming a unified regional dataset. This provides a well-organized and reliable data source for subsequent load fluctuation analysis and multi-dimensional data fusion. The entire association process only organizes and aligns the data; it does not change the original metering values ​​or affect the independence of legal metering results.

[0058] Based on the calculation logic of the load fluctuation impact factor, data features of neighboring equipment in the same area are extracted, a load fluctuation impact model is constructed, and a deviation analysis between load fluctuation features and metering data is completed. Following the calculation logic of the load fluctuation impact factor, features are extracted from the operating data of neighboring equipment within the same physical area. Based on the extracted features, a load fluctuation impact model is constructed, and a deviation analysis between load fluctuation features and metering data is completed. This step employs a load fluctuation feature extraction algorithm and a deviation comparison analysis algorithm to extract key features such as load change rate, current fluctuation amplitude, power surge frequency, and operating state switching duration from the operating data of neighboring equipment. Through feature quantification calculation, the degree of impact and fluctuation deviation of load fluctuations of neighboring equipment on the main energy meter's metering data are obtained.

[0059] In actual operation, the system first collects status and electrical data of charging piles starting and stopping, energy storage units charging and discharging, and other equipment within the same area. It then extracts fluctuation characteristics such as current fluctuation range, power increase / decrease rate, and state switching frequency, using these characteristics as input parameters to construct a load fluctuation impact model. Through model calculation and deviation comparison, the system can determine the short-term deviation and pattern of the main energy meter's metering data caused by surrounding load fluctuations. For example, when charging piles frequently start and stop, or when energy storage units perform high-power charging and discharging, it causes rapid fluctuations in line current and power. Through feature extraction and model calculation, the system can accurately analyze the short-term fluctuation deviations in the main energy meter's current and power metering data caused by such fluctuations, providing a basis for subsequent load fitting and state correction on the management side. The entire analysis process is only used for management-side state diagnosis and does not change the original legal metering results.

[0060] A multi-dimensional data fusion algorithm is used to weight and fuse the original metering data with load fluctuation influencing factors for load condition fitting and operational diagnosis, without modifying the original legal metering results. The fusion result is used solely for load condition fitting and equipment operation diagnosis within the transformer substation, without altering the original legal metering results or affecting the legality and accuracy of the metering data. The weighted multi-dimensional data fusion algorithm, according to a preset weight ratio, weights the original legal metering data with the previously calculated load fluctuation influencing factors to obtain load fitting data that more closely reflects actual operating conditions, thus reflecting the true electricity consumption status of the transformer substation.

[0061] Specifically, the system uses original legally measured electricity, voltage, and current data as a foundation, combined with the fluctuation degree reflected by load fluctuation factors, and performs fusion calculations according to weights to fit the actual load operation status of the distribution area. For example, by fusing the original legally measured electricity data output by the electricity meter with load fluctuation factors from nearby charging piles and energy storage units according to set weights, the system can more accurately reflect the actual electricity load of the distribution area, thereby identifying potential problems such as abnormal electricity consumption and overload. The above fusion and fitting process is only executed within the management unit, and the generated results are only used for operation diagnosis, status analysis, and anomaly judgment. It does not make any changes to the original legally measured values ​​output by the metering unit, ensuring the independence and reliability of the metering data.

[0062] Following the hardware execution priority order of AI accelerator card > DSP > CPU, the computational resources for fusion computing and load analysis tasks are scheduled and allocated, generating multi-dimensional data fusion, load status analysis, and computational resource scheduling information. The multi-dimensional data fusion computing and load status analysis tasks are uniformly scheduled and resource-allocated according to the hardware execution priorities of AI accelerator cards, DSPs, and CPUs, generating corresponding fusion analysis results and resource scheduling information. A hardware resource priority scheduling algorithm is employed, which determines available resources in real time based on the edge node hardware configuration and prioritizes high-performance acceleration units for computation, ensuring analysis efficiency.

[0063] In terms of scheduling rules, the system prioritizes allocating computational tasks such as data fusion and load analysis to the AI ​​accelerator card. When the AI ​​accelerator card is unavailable or busy, it automatically switches to the DSP to perform the relevant computations. If the edge node is not configured with a DSP or the DSP resource utilization is too high, the embedded CPU will undertake the computational tasks, and NEON instructions will be enabled for optimized processing according to the CPU architecture. When the CPU does not support NEON instructions, standard pure C code will be used to execute the computation, ensuring that the computational tasks can be completed normally under various hardware configurations.

[0064] In actual operation, edge nodes first monitor the working status of the AI ​​accelerator card in real time. If the AI ​​accelerator card is idle and available, tasks such as multi-dimensional data fusion and load fluctuation analysis are immediately assigned to the AI ​​accelerator card for execution. If the AI ​​accelerator card is detected to be busy or not deployed, the tasks are automatically scheduled to the DSP. In edge nodes without acceleration hardware, the CPU directly completes the entire calculation process, ensuring uninterrupted and latency-free data processing. Through the above hierarchical scheduling mechanism, data fusion and load analysis can be efficiently completed under different hardware configurations, without affecting the independent operation of the legal metrology unit.

[0065] Formula for calculating the impact factor of load fluctuation ,in, This represents the difference in current fluctuation. The power change rate is represented by SwitchNum, and the number of times the device is started and stopped is represented by SwitchNum. The weights for peak hours (8:00-22:00) are α=0.6, β=0.3, and γ=0.1, respectively, while the weights for valley hours are α=0.2, β=0.6, and γ=0.2.

[0066] The computing power scheduling quantification threshold is set as follows: if the AI ​​acceleration card utilization rate is >85%, switch to DSP; if the DSP is fully loaded, switch to CPU; the maximum number of data segments per computing power task is 50 to prevent memory overflow. The lightweight AI diagnostic model adopts a 3-layer CNN lightweight network, with a model size of <2MB and a local NPU inference time of ≤15ms; the model is updated incrementally every 7 days in the cloud, and the update process does not interrupt the metering data collection.

[0067] S105 deploys a dual-partition storage architecture at edge nodes to separate read and write operations, allocates storage resources according to data importance, generates hash values ​​for metering records through a hash algorithm, constructs a hash tree, and uploads the root node as a verification value to the blockchain network.

[0068] In one implementation, combining the trusted storage requirements of multi-source data from edge computing electricity meters with the security verification rules for metering data, a dual-partition read-write isolation mechanism and a data importance-based hierarchical allocation strategy are introduced to achieve edge node storage architecture deployment and trusted verification processing of metering data. Based on the actual requirements of tamper-proof power metering data and reliable storage in edge offline scenarios, a storage architecture isolation deployment rule is adopted to divide the storage resources of edge nodes into mutually independent areas, establishing a read-write separation and importance-based edge storage system. This ensures that legal metering data is not tampered with or accidentally written during storage, guaranteeing the security and integrity of metering data. For the actual operating scenarios of edge computing nodes in distribution areas, independent read-only partitions and read-write partitions are constructed. Simultaneously, a multi-level data storage strategy is established to implement classified secure storage of electricity meter legal metering data and equipment operating status data, with different storage permissions and methods for different types of data.

[0069] Specifically, legal measurement data is stored in a read-only storage partition, allowing only read operations and prohibiting any write or modification operations to ensure the immutability of the measurement data. Management data such as switch status, load data, and operation logs are stored in a read-write partition, supporting normal updates and queries. This partitioning and hierarchical storage method achieves physical isolation and secure storage between measurement and management data, meeting the architectural requirements of complete decoupling of measurement management and adapting to the reliable data preservation needs of edge nodes in complex scenarios such as offline operation and power outages.

[0070] A dual-partition storage architecture is deployed on edge computing nodes, implementing storage area isolation based on read / write operation type and allocating storage resources according to data importance level to establish a hierarchical storage mechanism for edge metering data. Within each edge computing node, a dual-partition storage architecture is deployed, physically and logically isolating storage areas according to read / write operation type and allocating storage resources according to data importance level, thereby establishing a stable and reliable hierarchical storage mechanism for edge metering data. The system employs a dual-partition read / write isolation algorithm and a data importance-based allocation algorithm to divide the flash memory hardware of the edge node into two independent areas: a read-only metering partition and a read / write management partition. Then, corresponding storage space is allocated according to data type and security level. Legal metering data, with the highest security requirements, is allocated to the read-only metering partition; important operational data is allocated secondarily; and ordinary status data, with lower security requirements, is uniformly allocated to the read / write management partition.

[0071] In actual deployment, edge nodes divide their built-in flash memory into two parts: a read-only partition and a read-write partition. Data related to legal metering, such as electricity, voltage, and current, is stored uniformly in the read-only partition. This partition only allows read access and does not accept any write, modification, or deletion operations, ensuring the immutability of metering data. Management data, such as switch status, load curves, operation logs, and communication records, is stored in the read-write partition, supporting normal data update, query, and overwrite operations. Simultaneously, the system automatically configures partition capacity based on data importance, prioritizing the space stability and data security of the read-only metering partition, while the read-write management partition dynamically adapts its storage capacity according to actual operational needs. This achieves the edge data storage effect of categorized storage, secure isolation, and efficient use.

[0072] Using legally mandated electricity meter readings as the core input, this system integrates hash calculation rules and data verification parameters to construct a hash value generation model for single-record metering data. This model is used to achieve tamper-proof and reliable verification of metering data. The system employs the SHA-256 hash algorithm to perform unified hash calculations on key information in the legally mandated metering records, including electricity consumption values, data collection timestamps, meter numbers, and transformer area numbers. Through encrypted calculation, a unique hash value of fixed length is generated as the digital fingerprint of this metering record, used for subsequent data integrity verification and anti-counterfeiting identification.

[0073] In practice, the system first extracts the complete information from the legally compliant metering records, then combines it into a string to be encrypted according to a pre-defined format. The SHA-256 algorithm is then used for hash calculation to obtain a unique and irreversible hash value. For example, for the positive active power record of an electricity meter at a specific moment, the system combines the power value, collection time, meter number, and transformer area number, performs a hash operation, generates a unique hash value, and stores it bound to the metering data. If the original metering data is tampered with, the recalculated hash value will be inconsistent with the original hash value. The system can quickly identify data anomalies, ensuring that legally compliant metering data is not illegally modified during storage and transmission, thus guaranteeing the authenticity and authority of the metering results.

[0074] The hash sequence of metering data is processed by constructing a hash tree hierarchy and extracting the root node. Combined with blockchain on-chain verification rules, a trusted evidence storage process is determined and integrated with the hash calculation results. Through tree-structured construction and on-chain verification values, a trusted storage execution chain for edge metering data is generated, forming a dual-partition storage and blockchain evidence storage system adapted to multi-source data from electricity meters. The Merkle hash tree construction algorithm is used to hierarchically combine the hash values ​​of multiple metering data within the same period to gradually generate a hash tree structure. The hash value of the top-level root node is extracted as the unified verification value for the entire batch of data. Then, according to blockchain evidence storage specifications, the root hash value is uploaded to the platform to complete the trusted evidence storage of batch data.

[0075] The system constructs a hash tree by combining the hash values ​​of electricity meter readings over consecutive time periods. For example, it combines the hash values ​​of multiple meter readings collected within an hour in pairs, calculates layer by layer, and generates a Merkle hash tree. Finally, it extracts a unique root hash value and uploads it to the blockchain. When data verification is required later, the root hash value is used to quickly verify whether the entire batch of data has been tampered with. Any alteration to any original data will result in a mismatch in the root hash value, thus achieving efficient and reliable verification of batch metering data and further ensuring the security of legally compliant metering data.

[0076] For dual-partition read-write isolation, the read-only metering partition uses a hardware mutex lock, and any write command initiated by any management process is directly intercepted by the hardware; metering data migration only allows full partition backup export, and prohibits individual modification; flash memory is automatically balanced in the background every quarter.

[0077] The national standard SM3 hash concatenation specification is as follows: the concatenation order is meter number#collection timestamp#forward electricity#station code#voltage and current, the separator is fixed#, empty fields are filled with 0000, SM3 is used to replace SHA256 to generate metering fingerprints, and SM2 signature encryption is used for transmission.

[0078] The Merkle tree construction rules are as follows: leaf nodes have a maximum of 50 meter hashes per batch; when the length of the hash array is odd, the last hash is copied to make it even; parent nodes are generated by SM3 merging two pairs of each layer, and the unique root hash at the top layer is used as the on-chain verification value.

[0079] The root hash is uploaded to the consortium blockchain of the distribution area using HTTPS+SM2 signature at the edge. If the upload fails 3 times, the root hash is cached locally and retried in the next collection cycle. If the upload fails 12 times in a row, a level 1 operation and maintenance alarm is generated and pushed to the platform.

[0080] S106, based on the data change frequency and correlation threshold, selects full synchronization or incremental push mode, only transmitting changed data and verification information to the cloud platform, and synchronously updating the local causal relationship graph.

[0081] In one implementation, combining the lightweight upload requirements of multi-source data from edge computing energy meters with local data association management rules, a data change frequency detection mechanism and an association degree threshold determination strategy are introduced. An adaptive selection process for full synchronization and incremental push is constructed to ensure reliable data upload while reducing data transmission volume between edge nodes and the cloud. The system employs a data change frequency statistical algorithm to monitor the rate of change of various data such as voltage, current, power consumption, and switch status in real time, classifying the data into steady-state and transient data. For example, the power consumption data of the energy meter has a long update interval and changes slowly, belonging to steady-state data; the switch status changes instantaneously with the on-site operating conditions, changing frequently, belonging to transient data. Simultaneously, a data association degree threshold calculation algorithm is used to compare the similarity between the currently collected data and the data saved in the previous period, determining whether the data has undergone a valid change based on a preset association degree threshold. For example, comparing the current voltage value with historical data, if the fluctuation amplitude is less than 1%, it is determined that no valid change has occurred and no upload is required; if the fluctuation amplitude is greater than 5%, it is determined that a valid change has occurred and upload is required.

[0082] The system automatically selects an appropriate data synchronization mode based on a comprehensive judgment of data change frequency and correlation threshold. When data changes frequently or the correlation is below the threshold, incremental push mode is immediately adopted; when data is stable for a long period and the correlation is above the threshold, full synchronization is performed at fixed intervals. For example, when the load in the distribution area is stable and the data does not fluctuate significantly, full synchronization can be performed once a day; when data changes abruptly, such as when a charging pile starts up or the switch status changes, incremental push is triggered immediately. After determining the transmission mode, the system uses an incremental data extraction algorithm to extract only the changed data entries and their corresponding hash verification information to form a lightweight transmission data packet. For example, only the status data of a switch changing from closed to open and its corresponding hash value are pushed; other unchanged data is not transmitted repeatedly, effectively reducing transmission pressure.

[0083] After data upload is complete, the cloud platform verifies the integrity and legality of the received data. Once verification is successful, the data is stored in the database, and an update confirmation command is returned to the edge nodes. The edge nodes update their local causal relationship graphs based on the confirmation information, ensuring that the data on the edge side is consistent with the cloud platform, thus completing the closed loop of the entire process of data collection, upload, verification, and update.

[0084] The incremental judgment quantification thresholds are as follows: incremental push is immediately triggered when the power fluctuation is ≥0.5kWh, the voltage fluctuation is ≥5V, or the switch status changes; if the fluctuation is below the threshold and there is no change for 24 consecutive hours, only the full synchronization at 0:00 every day is performed.

[0085] The incremental data packet standard uses JSON format as an example: {"time":"2026-06-23 10:00","meterId":"0100123456","data":{"E":1256.32,"U":220.5},"sm3_hash":"xxxx","change_flag":1}. It employs a lightweight graph database for storage, with nodes representing devices / areas and edges representing physical ownership and communication links. Historical data older than 90 days is automatically cleaned up, retaining only the most recent 90 days for traceability.

[0086] S107 performs cross-terminal data association and fusion, anomaly diagnosis and judgment, and load characteristic analysis to generate mixed output data in a unified format.

[0087] In one implementation, combining the cross-terminal collaborative requirements of multi-source data fusion processing for edge computing energy meters with the practical logic of unified processing of multiple devices, a heterogeneous terminal access standard, a multi-protocol parsing mechanism, and a multi-sampling frequency timing alignment strategy are introduced to build a multi-type terminal data collaborative processing system suitable for power field applications. This system takes energy meters, energy storage units, and charging piles within the same distribution substation as collaborative processing objects. Through unified access, unified parsing, unified timing alignment, and unified analysis, it integrates originally scattered, heterogeneous, and asynchronous data into standardized data that can be directly used for management and analysis, ensuring that multi-source data can be compatible, correlated, and fused.

[0088] During implementation, the system establishes standardized communication links for different devices according to a unified terminal access specification, enabling access to edge nodes regardless of the interface and protocol used by the device. For example, electricity meters access via RS485, energy storage units via Ethernet, and charging piles via NB-IoT can all be identified by edge nodes and incorporated into unified management. Simultaneously, the system employs a multi-protocol adaptive parsing mechanism to uniformly parse protocol data from different devices, converting data in different formats such as DL / T645, Modbus, and MQTT into an internal unified format. Furthermore, a multi-sampling frequency time-series alignment strategy aligns high-frequency and low-frequency sampled data onto the same time axis, allowing data collected at different times to be compared and calculated within the same time period.

[0089] For example, the sampling data from electricity meters (once every 1 minute), energy storage units (once every 5 minutes), and charging piles (on real-time sampling data) are all aligned to a 5-minute time base to ensure consistent data timing during subsequent load analysis and anomaly diagnosis. Edge nodes can centrally manage and jointly analyze multiple types of equipment within the distribution area, providing complete and reliable data support for load monitoring, anomaly diagnosis, and operation optimization without modifying legal metering data.

[0090] Leveraging the terminal access capabilities of edge computing nodes, the system enables unified access for various types of terminal devices, including electricity meters, energy storage units, and charging piles, establishing stable and reliable data acquisition and transmission links to meet the collaborative data collection needs of multiple devices. The system employs a multi-interface link adaptation algorithm, automatically matching different communication interfaces such as RS485, Ethernet, NB-IoT, and LoRa based on the communication methods of the terminal devices, completing link negotiation and parameter configuration, and quickly establishing a stable connection.

[0091] In actual deployment, edge nodes can automatically select the appropriate interface based on the device type: connecting to the electricity meter via the RS485 interface to achieve stable data acquisition; connecting to the energy storage unit via the Ethernet interface to ensure high-speed transmission of data during high-power operation; and connecting to the charging pile via the NB-IoT interface to meet the needs of long-distance, low-power data upload. All types of devices use independent links for data transmission, ensuring no interference and guaranteeing stable operation of each acquisition link without data loss or crosstalk even in scenarios with multiple terminals accessing simultaneously.

[0092] Through this multi-interface adaptive access method, edge nodes can be compatible with different types and communication methods of power terminals on site, and can complete unified access without separate configuration, simplifying the on-site construction and commissioning process, and providing a stable and reliable data source for subsequent protocol parsing, data fusion and status analysis.

[0093] Following the unified parsing rules for heterogeneous protocols, the system performs protocol parsing and format conversion on raw data from different terminals using multiple protocols, and performs time alignment on data from different sampling frequencies using a unified time base. This ensures that multi-source data has a basis for fusion and comparison. The system employs a multi-protocol adaptive parsing algorithm to automatically identify various power communication protocols such as DL / T645, Modbus-RTU, and MQTT. It parses valid information such as voltage, current, power consumption, switch status, and operating status field by field according to protocol specifications, and uniformly converts data of different formats into the system's internal standard format to avoid data fusion problems caused by protocol differences.

[0094] Simultaneously, a time-axis interpolation alignment algorithm is employed to perform time-series normalization on data from different sampling frequencies. For example, electricity meters sample at 1-minute intervals, energy storage units sample at 5-minute intervals, and charging piles sample in real time. The system uses 5 minutes as a unified time benchmark, performs averaging on high-frequency sampling data, and performs supplementary point alignment on low-frequency sampling data, ensuring that all data correspond and match at the same point in time. Through unified protocol parsing and time-series alignment processing, multi-source data that originally had different protocols, was asynchronous, and had different frequencies are transformed into standardized data with the same format and time axis, providing an accurate and consistent data foundation for subsequent cross-terminal correlation and fusion, load analysis, and anomaly diagnosis.

[0095] Based on the cross-terminal data collaborative processing flow, multi-source data that has completed time-series alignment is correlated and fused. Then, based on data characteristics, anomaly diagnosis and load characteristic analysis are carried out without modifying the original legal and metrological data throughout the process. The system adopts cross-terminal correlation and fusion algorithms and multi-dimensional anomaly diagnosis algorithms. Combining the equipment's area, physical location, and real-time operating status, it integrates scattered data into a unified analysis dataset, while automatically identifying abnormal operating conditions and extracting area load characteristics.

[0096] In practical processing, the aligned electricity meter data, energy storage unit charging and discharging power data, and charging pile current and power data are correlated and integrated to form comprehensive load data based on the same time and the same distribution area. The system determines whether the equipment is operating abnormally by comparing the changes in power and current and their duration. For example, when the charging pile current continuously exceeds the rated range and the duration exceeds the set threshold, it can be diagnosed as an overload abnormality of the charging pile, and the abnormal period and abnormal value will be automatically marked.

[0097] Meanwhile, based on the fused load curve, the system extracts key characteristics such as peak load occurrence time, maximum load value, and average load for the transformer substation, providing a basis for substation operation management. The entire analysis process is completed independently within the management unit and is used only for status monitoring and operation optimization, without affecting the legal metering results of the metering unit, ensuring both metering independence and analytical practicality.

[0098] The full-scenario anomaly judgment rules are as follows: For charging piles: continuous current > 100% of rated current for 1 hour → overload. For energy storage: sudden change in charging / discharging power ±80% → equipment failure. For distribution areas: synchronous anomalies of 3 or more devices → line failure. The cloud standard outputs complete fields, including distribution area number, meter number, 5-minute aligned metering value, energy storage charging / discharging capacity, charging pile load, anomaly tag, SM3 checksum hash, load fitting value, and collection timestamp. The cloud dynamic parameter update process is as follows: the cloud distributes the threshold, AI model, and weight parameter encrypted package; after edge SM2 verification, it automatically overwrites the local configuration; the entire update process does not restart the metering and acquisition link; if the update fails, it automatically rolls back the old parameters.

[0099] Following the data interaction standards between edge computing nodes and the cloud platform, the results after fusion, diagnosis, and analysis are output in a standardized manner, forming a standardized output execution chain adapted to collaborative processing of multiple types of terminals. The system uses a data format standardization algorithm to uniformly convert fused data, including metering results, equipment status, anomaly markers, and load characteristics, into a compatible format that the cloud platform can directly parse, completing a closed-loop output from the edge to the cloud.

[0100] The system unifies the cross-terminal fusion results into JSON format data. This data includes complete information such as transformer substation number, equipment identifier, data collection time, legal metering results, switch status, anomaly tags, and load characteristics, and is output to the cloud platform. For example, it integrates the electricity meter readings, energy storage charging / discharging status, charging pile operating status and anomaly diagnosis results, and peak load characteristics within the same transformer substation into a single standard data point. After uploading to the cloud, this data can be directly used for transformer substation operation monitoring, load management, and anomaly maintenance without secondary processing. The entire output process adheres to power system data interaction standards, ensuring a unified data structure, standardized fields, and accurate timing. While guaranteeing data integrity, it also meets the access and display requirements of the cloud platform, providing reliable data support for intelligent transformer substation management.

[0101] Furthermore, this application aims to address the issue of damaged flash read-only partitions: metering data is automatically backed up to a backup Flash partition, and restored via hash verification after hardware replacement during maintenance. When the SM3 hash does not match, a high-risk anomaly is immediately flagged, the upload of this data is suspended, and the original data collection message is pushed to supervisory personnel for review. When the edge device experiences a power outage, the flash memory breakpoint sequence number is automatically read upon power-up, and data transmission resumes from the moment before the power outage, with no data loss. When the blockchain connection is interrupted for an extended period, the root hash is locally cached for 30 days, and data is uploaded to the blockchain in batches upon link restoration.

[0102] The standard deployment for the distribution area is as follows: a single ARM+NPU edge gateway covers one distribution area; the metering unit uses an independent metering chip (compliant with the State Grid's legal metering chip specifications), with physical isolation between the metering and management chips, and the whole unit has the function of metering anti-tampering seal triggering log.

[0103] In one implementation, such as Figure 2 As shown, this application also provides a multi-source data fusion processing system for electricity meters oriented towards edge computing, including: The edge heterogeneous acquisition unit 201 is used to build acquisition hardware based on the heterogeneous architecture of embedded CPU + AI accelerator card and multiple types of interfaces such as RS485 / Ethernet / NB-IoT / LoRa. It is compatible with DL / T645 / DL / T698 / Modbus-RTU / IEC104 / MQTT power communication protocols, and can connect to various types of terminal devices such as electricity meters, energy storage units, and charging piles to complete heterogeneous protocol parsing and multi-sampling frequency data alignment processing. The protocol parsing and caching unit 202 is used to extract the voltage, current, power consumption and power distribution terminal switch status operation data of the energy meter through the protocol parsing engine, and provide breakpoint resume and local caching capabilities. In the case of network outage, the data is written to the local flash memory and the retransmission operation is automatically performed after the network is restored. The metrology management decoupling unit 203 is used to adopt a fully decoupled architecture of metrology and management. The metrology unit independently completes the collection of legal metrology data and outputs tamper-proof metrology results, while the management unit performs data preprocessing and storage by type at the edge computing node, and completes abnormal data screening and feature labeling based on preset thresholds. The multi-dimensional data fusion unit 204 is used to apply a multi-dimensional data fusion algorithm, combine the physical location-related operating data of nearby devices, calculate the load fluctuation impact factor, and carry out management-side load fitting and state correction, and complete the scheduling of computing resources according to the hardware priority of AI accelerator card > DSP > CPU. The dual-partition trusted storage unit 205 is used to deploy a dual-partition storage architecture at edge nodes to achieve read and write operation separation, allocate storage resources according to data importance, generate metering record hash values ​​and build a hash tree through a hash algorithm, and upload the root node as a verification value to the blockchain network. The cloud synchronization update unit 206 is used to select full synchronization or incremental push mode based on the data change frequency and correlation threshold, and only transmits the changed data and verification information to the cloud platform to synchronously update the local causal relationship graph data. The cross-terminal collaborative analysis unit 207 is used to perform cross-terminal data association and fusion, anomaly diagnosis and judgment and load characteristic analysis, and generate mixed output data in a unified format.

[0104] The computer-readable storage medium provided in the above embodiments of this application and the multi-source data fusion processing method for electricity meters oriented towards edge computing provided in the embodiments of this application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the applications stored therein.

[0105] The various embodiments in this application are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for the embodiments evaluating the multi-source data fusion processing method for edge computing electricity meters, electronic devices, electronic equipment, and readable storage media, since they are substantially similar to the embodiments of the multi-source data fusion processing method for edge computing electricity meters described above, relevant parts can be referred to in the description of the embodiments of the multi-source data fusion processing method for edge computing electricity meters described above.

Claims

1. A method for multi-source data fusion processing of electricity meters oriented towards edge computing, characterized in that, include: Based on the heterogeneous architecture of embedded CPU+AI accelerator card and RS485 / Ethernet / NB-IoT / LoRa interface, an edge-side multi-source data acquisition unit is built, which is compatible with DL / T645 / DL / T698 / Modbus-RTU / IEC104 / MQTT protocols and can connect to various types of terminal devices including electricity meters, energy storage units and charging piles, and complete heterogeneous protocol parsing and multi-sampling frequency data alignment. The protocol parsing engine extracts data on the voltage, current, energy consumption, and switch status of the power meter and distribution terminal. It supports breakpoint resume and local caching. Data is stored in local flash memory when the network is down and automatically retransmitted after the network is restored. The system adopts a fully decoupled architecture for measurement and management, with each measurement unit independently completing the collection of legal measurement data and outputting tamper-proof measurement results. The management unit performs data preprocessing through edge computing nodes, classifies and stores data according to data type, filters abnormal data based on preset thresholds, and marks it with feature tags; By using a multi-dimensional data fusion algorithm and combining the physical location-related operating data of nearby devices, load fluctuation impact factors are calculated and management-side load fitting and status correction are carried out. Computing resources are scheduled according to the priority order of AI accelerator card > DSP > CPU. A dual-partition storage architecture is deployed at the edge nodes to separate read and write operations. Storage resources are allocated hierarchically according to data importance. A hash algorithm is used to generate hash values ​​for metering records, a hash tree is constructed, and the root node is uploaded to the blockchain network as a verification value. Based on the frequency of data changes and the correlation threshold, select the full synchronization or incremental push mode, and only transmit the changed data and verification information to the cloud platform to update the local causal relationship graph synchronously. Perform cross-terminal data association and fusion, anomaly diagnosis and judgment, and load characteristic analysis to generate mixed output data in a unified format.

2. The method as described in claim 1, characterized in that, The system adopts a fully decoupled architecture for measurement and management, with each measurement unit independently completing the collection of legal measurement data and outputting tamper-proof measurement results. The management unit performs data preprocessing through edge computing nodes, classifies and stores data according to data type, filters out abnormal data based on preset thresholds, and labels it with feature tags, including: In response to the decoupling requirements of metering management for multi-source data fusion processing of edge computing electricity meters, an independent acquisition mechanism for legal metering and edge-side preprocessing rules are introduced to determine the division of labor for the acquisition and preprocessing of multi-source data from electricity meters. The data acquisition process for electricity meter voltage, current, energy consumption, and distribution terminal switch status is divided among different departments. The metering unit independently completes the acquisition of legal metering data, outputs the legal metering results, and the management unit is not allowed to modify them. The management unit receives non-metering data. A fully decoupled architecture consisting of a metering acquisition module, edge computing nodes, classification storage units, and anomaly screening and marking components is built, and an independently operating metering and management processing link is set up; The legal metrology data source is input into the independent metrology unit to complete high-precision acquisition and metrology data output, without participating in the management unit's calculations; Non-metering data and raw metering results output by metering units are input into the edge computing node management unit to perform data cleaning, format conversion and feature extraction operations; The data after edge preprocessing is classified and stored according to data type. Abnormal data is judged and feature labels are marked according to preset thresholds, generating data collection, preprocessing, storage and abnormal marking information under the fully decoupled architecture of metering management.

3. The method as described in claim 1, characterized in that, Utilizing a multi-dimensional data fusion algorithm, combined with operational data from neighboring devices related to physical location, the system calculates load fluctuation impact factors and performs management-side load fitting and status correction. Computational resources are scheduled according to the priority order: AI accelerator card > DSP > CPU, including: Combining the multi-source data fusion processing requirements of edge computing electricity meters with the characteristics of multi-dimensional data fusion algorithms, physical location association rules and load fluctuation analysis mechanisms are introduced to build a management-side data fusion and load status analysis process adapted to electricity meter metering scenarios. In accordance with the multi-source data fusion specification for electricity meters, physical location association matching processing is performed on voltage, current, power consumption and operation data of nearby equipment to complete the data association preprocessing before fusion; Based on the calculation logic of load fluctuation impact factors, data characteristics of adjacent equipment in the same area are extracted, a load fluctuation impact model is constructed, and the deviation analysis between load fluctuation characteristics and metering data is completed. The original measurement data and load fluctuation influencing factors are weighted and fused by a multi-dimensional data fusion algorithm for load state fitting and operation diagnosis, without modifying the original legal measurement results. Based on the hardware execution priority of AI accelerator card > DSP > CPU, the computing resource scheduling and allocation for fusion computing and load analysis tasks are completed, generating multi-dimensional data fusion, load status analysis and computing resource scheduling information.

4. The method as described in claim 1, characterized in that, A dual-partition storage architecture is deployed at edge nodes to separate read and write operations. Storage resources are allocated hierarchically according to data importance. A hash algorithm is used to generate hash values ​​for metering records, a hash tree is constructed, and the root node is uploaded to the blockchain network as a verification value. This includes: Combining the trusted storage requirements of multi-source data from edge computing energy meters with the security verification rules for metering data, a dual-partition read-write isolation mechanism and a data importance-based hierarchical allocation strategy are introduced to realize the deployment of edge node storage architecture and trusted verification processing of metering data. Deploy a dual-partition storage architecture on edge computing nodes, perform storage area isolation according to read and write operation types, complete the targeted allocation of storage resources according to the importance level of data, and establish a hierarchical storage mechanism for edge metering data; Using legal metering records of electricity meters as the core input object, and integrating hash calculation rules and data verification parameters, a hash value generation model for a single metering data is constructed. By constructing a hash tree hierarchy and extracting the root node model to process the hash sequence of metering data, and combining it with the blockchain on-chain verification rules to determine the trusted evidence storage process, and integrating it with the hash calculation results, a trusted storage execution chain for edge metering data is generated through tree structure and on-chain verification values, forming a dual-partition storage and blockchain evidence storage system adapted to multi-source data of electricity meters.

5. The method as described in claim 4, characterized in that, Perform cross-terminal data association and fusion, anomaly diagnosis and judgment, and load characteristic analysis to generate mixed output data in a unified format, including: Combining the cross-terminal collaborative requirements of multi-source data fusion processing of edge computing energy meters with the unified processing logic of multi-device data, heterogeneous terminal access specifications, multi-protocol parsing mechanisms and multi-sampling frequency timing alignment strategies are introduced to carry out collaborative fusion and analysis processing of multi-type terminal data. Based on the terminal access capabilities of edge computing nodes, it connects to various types of terminal devices such as electricity meters, energy storage units, and charging piles to establish a stable data acquisition and transmission link. According to the unified parsing rules for heterogeneous protocols, protocol parsing and format conversion are performed on the raw data of multiple protocols from different terminals, and timing alignment is performed on the data of multiple sampling frequencies according to a unified time base. Based on the cross-terminal data collaborative processing flow, the aligned multi-source data is correlated and fused, and anomaly diagnosis and load characteristic analysis are completed based on data characteristics; Based on the data interaction standards of edge computing and cloud platforms, output cross-terminal fused data in a unified format to form a collaborative processing and standardized output execution chain that adapts to multiple types of terminals.

6. A multi-source data fusion processing system for electricity meters oriented towards edge computing, characterized in that, The system includes: The edge heterogeneous acquisition unit is used to build acquisition hardware based on the heterogeneous architecture of embedded CPU + AI accelerator card and multiple types of interfaces such as RS485 / Ethernet / NB-IoT / LoRa. It is compatible with DL / T645 / DL / T698 / Modbus-RTU / IEC104 / MQTT power communication protocols and can connect to various types of terminal devices such as electricity meters, energy storage units, and charging piles to complete heterogeneous protocol parsing and multi-sampling frequency data alignment processing. The protocol parsing and caching unit is used to extract the voltage, current, power consumption and power distribution terminal switch status data of the energy meter through the protocol parsing engine. It provides breakpoint resume and local caching capabilities. In the case of network outage, the data is written to the local flash memory and the retransmission operation is automatically performed after the network is restored. The metrology management decoupling unit is used to adopt a fully decoupled architecture of metrology and management. The metrology unit independently completes the collection of legal metrology data and outputs tamper-proof metrology results, while the management unit performs data preprocessing, classifies and stores data by type, and completes abnormal data screening and feature labeling based on preset thresholds at the edge computing node. The multi-dimensional data fusion unit is used to apply multi-dimensional data fusion algorithms, combine physical location-related data of nearby devices, calculate load fluctuation impact factors, and carry out management-side load fitting and state correction. It also completes computing resource scheduling according to the hardware priority of AI accelerator card > DSP > CPU. The dual-partition trusted storage unit is used to deploy a dual-partition storage architecture at edge nodes to achieve read and write operation separation, allocate storage resources according to data importance, generate metering record hash values ​​through a hash algorithm and build a hash tree, and upload the root node as a verification value to the blockchain network. The cloud-based synchronization update unit is used to select full synchronization or incremental push mode based on the data change frequency and correlation threshold, and only transmits the changed data and verification information to the cloud platform to synchronously update the local causal relationship graph data. The cross-terminal collaborative analysis unit is used to perform cross-terminal data association and fusion, anomaly diagnosis and judgment, and load characteristic analysis, generating mixed output data in a unified format.

7. An electronic device, characterized in that, include: First processor; and memory for storing executable instructions of the first processor; The first processor is configured to execute the multi-source data fusion processing method for electricity meters oriented towards edge computing as described in any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the second processor, it implements the multi-source data fusion processing method for electricity meters oriented towards edge computing as described in any one of claims 1 to 5.

Images