Power grid fault diagnosis and centralized meter reading method based on multi-agent system

Abstract

The invention discloses a power grid fault diagnosis and centralized meter reading method based on a multi-agent system, and the method comprises the steps: collecting the operation parameters of power equipment and the reading of an electric energy meter, achieving the distributed synchronous obtaining of the operation parameters and the reading of the electric energy meter, and avoiding the data aggregation delay and single-point collection bottleneck in a centralized architecture; according to the method, branch-level preliminary screening is performed in combination with a power grid topology directed graph and operation parameters, and a diagnosis conclusion is output based on a multi-source parameter fusion result, so that the problem of diagnosis interruption caused by failure of a center node is solved, and the real-time performance and accuracy of fault positioning are improved; in addition, local aggregation, structured storage and index generation are carried out on the electric energy reading, so that wide-area communication dependence is reduced, link congestion is relieved, and the continuous execution capability of a meter reading task under the condition of communication limitation or local fault is guaranteed; collaborative autonomy of fault diagnosis and meter reading functions under a unified multi-agent framework is integrally realized, and the robustness, expansibility and engineering practicability of the system are remarkably enhanced.

Description

Technical Field

[0001] This invention relates to the field of intelligent management and information processing technology for power systems, specifically to a method for power grid fault diagnosis and centralized meter reading based on a multi-agent system. Background Technology

[0002] In recent years, the application of distributed artificial intelligence methods in power systems has gradually attracted attention. Multi-agent systems, due to their autonomy, cooperation, and flexibility, can effectively overcome the limitations of centralized architectures, providing new solutions for handling complex tasks in power grids. However, existing distributed artificial intelligence systems lack comprehensive collaborative processing capabilities. Therefore, how to achieve integrated optimization in the power grid environment has become a pressing technical challenge. Summary of the Invention

[0003] The purpose of this invention is to provide a power grid fault diagnosis and centralized meter reading method based on a multi-agent system. Targeting the operation monitoring and management scenarios of smart grids, this method utilizes a distributed agent collaboration and optimization mechanism to achieve rapid diagnosis and isolation of power grid faults, and significantly improves the efficiency and reliability of centralized meter reading for large-scale users.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: Firstly, a method for power grid fault diagnosis and centralized meter reading based on a multi-agent system is provided, the method comprising: Collect the operating parameters of each power device in the target power grid and the power readings of each electricity meter; Based on the operating parameters of the power equipment and the directed graph of the power grid topology, faulty power grid branches are screened from the target power grid, and the fault diagnosis results of each faulty power grid branch are determined according to the target fusion results of the operating parameters of the target power equipment. The target power equipment is all the power equipment on the faulty power grid branch. The energy readings are aggregated, and the aggregation results are stored in a structured manner using a relational database. A data index is generated, and application support is provided based on the aggregation results and historical data.

[0005] Secondly, a power grid fault diagnosis and centralized meter reading system based on a multi-agent system is provided, the system comprising: The data acquisition module is used to collect the operating parameters of each power device in the target power grid and the energy readings of each energy meter; The fault diagnosis module is used to filter faulty power grid branches from the target power grid based on the operating parameters of the power equipment and the directed graph of the power grid topology, and to determine the fault diagnosis results of each faulty power grid branch according to the target fusion result of the operating parameters of the target power equipment. The target power equipment is all the power equipment on the faulty power grid branch. The centralized meter reading module is used to aggregate the various electricity readings, store the aggregation results in a structured manner through a relational database, generate a data index, and provide application support based on the aggregation results and historical data.

[0006] In summary, the present invention has at least one of the following beneficial technical effects: This application provides a method for power grid fault diagnosis and centralized meter reading based on a multi-agent system. The method includes collecting operating parameters of power equipment and electricity meter readings, achieving distributed synchronous acquisition of both, thus avoiding data aggregation delays and single-point acquisition bottlenecks in centralized architectures. It combines a directed graph of the power grid topology with operating parameters for branch-level preliminary screening, and outputs diagnostic conclusions based on the fusion results of multi-source parameters, overcoming the diagnostic interruption problem caused by central node failure and improving the real-time performance and accuracy of fault location. Furthermore, it performs local aggregation, structured storage, and index generation of electricity readings, reducing wide-area communication dependence, alleviating link congestion, and ensuring the continuous execution capability of meter reading tasks under communication constraints or localized fault conditions. Overall, it achieves collaborative autonomy of fault diagnosis and meter reading functions within a unified multi-agent framework, significantly enhancing the system's robustness, scalability, and engineering practicality. Attached Figure Description

[0007] Figure 1 A flowchart illustrating the steps of a power grid fault diagnosis and centralized meter reading method based on a multi-agent system provided in this application; Figure 2 A flowchart illustrating the steps of another power grid fault diagnosis and centralized meter reading method based on a multi-agent system provided in this application; Figure 3 A flowchart illustrating the steps of another power grid fault diagnosis and centralized meter reading method based on a multi-agent system provided in this application; Figure 4 A flowchart illustrating the steps of another power grid fault diagnosis and centralized meter reading method based on a multi-agent system provided in this application; Figure 5 A schematic diagram of the structure of a smart grid management system provided in this application; Figure 6 A schematic diagram of another smart grid management system provided in this application; Figure 7 A schematic diagram of another smart grid management system provided in this application; Figure 8 A schematic diagram of another smart grid management system provided in this application; Figure 9 A schematic diagram of another smart grid management system provided in this application; Figure 10This is a schematic diagram of another intelligent management system for multiple power grids provided in this application.

[0008] Explanation of reference numerals in the attached diagram: 1000. Smart power grid management system; 100. Data acquisition module; 200. Fault diagnosis module; 300. Centralized meter reading module; 201. Phasor Measurement Unit; 202. Diagnostic Unit; 301. Data Management Unit; 302. Data Aggregation and Forwarding Unit; 303. Storage and Indexing Unit; 304. Analysis Unit. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0010] like Figure 1 As shown, Figure 1 This application provides a flowchart of a method for power grid fault diagnosis and centralized meter reading based on a multi-agent system. The steps of this method include: Step S20: Collect the operating parameters of each power device in the target power grid and the power readings of each power meter; The core of this application lies in decoupling modular intelligent agent division of labor and distributed data flow to simultaneously support two key business functions: highly reliable fault diagnosis and highly efficient centralized meter reading.

[0011] The target power grid can be a distribution network or transmission network subsystem under a specific voltage level to be monitored and managed. Its scope can be dynamically defined according to dispatching zones, geographical areas or asset ownership, such as 10kV distribution network feeder sections, 35kV substation outgoing line bays or 110kV ring network units. For example, data can be collected by devices such as voltmeters, ammeters, frequency acquisition instruments, and phase acquisition instruments installed at various key nodes of the target power grid, power equipment, and electricity meters.

[0012] For fault diagnosis of the target power grid, the system can collect operating parameters of various power equipment and key nodes in the target power grid. Operating parameters include, but are not limited to, the opening and closing status of circuit breakers / disconnectors, transformer oil temperature, effective value of bus voltage, harmonic content of feeder current, frequency deviation, power factor, etc.

[0013] For centralized meter reading of a target power grid, data can be collected by reading the electricity meter readings of each user. These readings include total forward active power, total reverse active power, four-quadrant reactive power, maximum demand, and the timestamp of the occurrence. For example, this data can be parsed using the DLMS / COSEM protocol after collection.

[0014] Step S30: Based on the operating parameters of the power equipment and the directed graph of the power grid topology, the faulty power grid branches are screened from the target power grid, and the fault diagnosis results of each faulty power grid branch are determined according to the target fusion results of the operating parameters of the target power equipment. The target power equipment is all the power equipment on the faulty power grid branch. Among them, the directed graph of the power grid topology is an abstraction of the target power grid into a directed graph. This graph model can be jointly constructed by the static topology library of the SCADA system and the dynamic topology identification module of WAMS, and supports real-time updates and version snapshot management of topology change events (such as switching operations, line commissioning and decommissioning).

[0015] The target fusion result can be for the same faulty branch. The system integrates and outputs multi-source information on the operating parameters of all target power equipment. This fusion result is directly mapped to the fault type identification framework Θ = {short-circuit fault, open-circuit fault, ground fault, equipment aging, malfunction interference}, ultimately outputting the diagnostic conclusions corresponding to each fault branch eᵢ, such as " "A-phase metallic short circuit, confidence level 92.3%".

[0016] The target power equipment is explicitly defined as all equipment connected to the branch (e.g., if branch e_i connects circuit breaker CB-101 and transformer T-202, then CB-101 and T-202 are the target power equipment), ensuring that subsequent fusion diagnostics cover all observable variables within the fault influence domain. As an optional embodiment, a graph neural network can be used instead of explicit topology rule matching, modeling G=(V,E) as a graph structure, using node embedding vectors to represent equipment states, and learning the branch fault probability distribution through a message passing mechanism, thereby adapting to active distribution network scenarios with frequent topology changes.

[0017] Step S40: Aggregate the various energy readings, store the aggregation results in a structured manner using a relational database, generate a data index, and provide application support based on the aggregation results and historical data.

[0018] Aggregation can be performed under dual constraints of time and space dimensions to reduce data. The purpose of aggregation is to suppress noise such as that caused by channel attenuation through data fusion. The data index can include a basic tree index and also support time-series range queries (such as "average daily electricity consumption of a certain community in Q3 2024") and multi-dimensional association queries (such as "fluctuation rate of meter readings corresponding to voltage over-limit periods"). Optionally, the centralized meter reading module 300 can integrate a time-series database (such as TimescaleDB) as an extended storage engine for relational databases to perform cold and hot data separation management of high-frequency raw readings. Hot data is stored in a PostgreSQL in-memory table, and cold data is automatically compressed and archived to columnar storage.

[0019] Relational databases can support JSONB fields, partitioned tables, parallel queries, and materialized views; when using structured storage, aggregation results are written to the core table.

[0020] Generating a data index refers to creating a composite index on the core table and simultaneously building an inverted index for quickly retrieving specific user groups (such as "all residential users whose peak electricity consumption exceeds 500kWh").

[0021] This application relies on a multi-agent hierarchical data acquisition architecture to complete preliminary perception and protocol adaptation of power equipment operating parameters and electricity meter readings at the edge, avoiding bandwidth pressure and central node computing bottlenecks caused by massive raw data backhaul. It uses a directed graph of the power grid topology as a physical constraint framework to quickly delineate the scope of fault impact. Furthermore, by implementing spatiotemporal dual aggregation and weighted adaptive mechanisms on electricity readings, it preserves macroscopic load trend characteristics while suppressing the impact of random noise and occasional outliers. Through standardized modeling and multi-dimensional indexing using a relational database, the aggregation results can directly serve billing, scheduling, and analysis systems, significantly shortening the data value conversion path. Therefore, this embodiment simultaneously solves the technical problems mentioned in the background art, such as slow fault diagnosis response, low accuracy, low efficiency of centralized meter reading, and susceptibility to communication bottlenecks, achieving the technical effects of short fault branch identification time and high diagnostic accuracy.

[0022] Optionally, the method further includes: A sliding window-based mean filtering method and outlier removal strategy are used to denoise the operating parameters of each power device and the energy readings of each energy meter.

[0023] Among them, the mean filtering method based on a sliding window refers to applying a length of [missing information] to the raw collected data in the form of a continuous time series. A sliding window is used, and the arithmetic mean is calculated within each window. This average is then used to replace the original sampling point at the center of the window, thereby achieving smooth suppression of high-frequency random noise. Sliding window length. The value is an integer, ranging from 3 to 101, and its specific value is determined in conjunction with the dynamic response characteristics of the power equipment and the sampling frequency: when the sampling frequency is 1 kHz, With a time window width of 20 ms, it can effectively filter out 50 Hz power frequency harmonics and their odd harmonics; when the sampling frequency is 100 Hz... It corresponds to a 110 ms window width and is suitable for typical fault disturbances with transient process durations on the order of 100 ms.

[0024] Outlier removal strategies refer to identifying and correcting / removing data points that significantly deviate from the normal operating range based on statistical principles or domain prior knowledge. For example, this includes two parallel implementation modes: the first is outlier detection based on the three-standard-deviation criterion, which involves calculating the mean of the local sequence after sliding window filtering. with standard deviation , will satisfy data points The first type is an outlier, which is replaced by the median or the interpolated result of a nearby time interval within the window; the second type is a hard-limiting discrimination based on physical constraint thresholds, that is, setting differentiated threshold ranges for different types of power equipment—for example, the reasonable range for the phase angle difference of the voltage of a 10 kV distribution network branch is... If the measured value exceeds this range, it is considered an anomaly caused by communication errors or sensor drift, and is directly marked as invalid, triggering the proxy re-sampling mechanism. The two modes can be enabled independently or used in cascade: first, perform a coarse screening using physical thresholds, and then apply statistical criteria to the remaining data for fine screening.

[0025] This application implements sliding window mean filtering and outlier removal simultaneously on power equipment operating parameters and electricity meter readings, solving the problems of fault diagnosis deviation and meter reading error caused by measurement errors, communication noise and external disturbances affecting the original collected data. It significantly improves the quality consistency and spatiotemporal continuity of input data, reduces the probability of conflict in fusion judgment, thereby improving the fault location accuracy and centralized meter reading success rate, and enhancing the robustness and engineering practicality of multi-agent systems in complex electromagnetic environments.

[0026] like Figure 2 As shown, this application also provides an optional method embodiment for screening faulty grid branches from a target power grid based on power equipment operating parameters and a directed graph of the power grid topology. This method embodiment includes: Step S201: Determine the voltage at both ends of each power grid branch of the target power grid from the operating parameters of each power equipment, and calculate the voltage phase angle difference based on the voltage at both ends of each power grid branch. Step S202: Compare each voltage phase angle difference with a preset threshold, and determine the grid branch with the comparison result being the target result as the faulty grid branch. The target result is that the voltage phase angle difference exceeds the preset threshold.

[0027] This application may first introduce a graph theory modeling method based on topological constraints to represent the power grid as a directed graph. ,in A set of nodes, representing a busbar or substation. Let be the set of edges, representing power grid branches or circuit breakers. For each power grid branch... The voltage phase angle difference between the two nodes is obtained through the phasor measurement unit 201. If it exceeds the preset threshold If the faulty branch is not found, then the branch is considered a faulty branch. in, This represents the candidate set of faulty power grid branches. The physical meaning of the formula is that when a power grid experiences faults such as short circuits, grounding, or open circuits, it causes abrupt changes in power flow distribution, with the phase angle difference significantly deviating from the normal operating range. By comparing with a threshold, abnormal branches can be quickly identified, narrowing down the scope of fault diagnosis. This method not only improves the accuracy of boundary discrimination but also reduces the computational overhead of subsequent multi-source fusion.

[0028] To further enhance the accuracy of the identification, this application can also introduce a topology traversal method based on breadth-first search (BFS) in the process of identifying faulty power grid branches. By starting from the identified faulty power grid branches and expanding the search layer by layer outward, the fault propagation path model can be dynamically updated, enabling rapid location and boundary convergence of the fault area. This process effectively avoids misjudgments caused by single-point measurement errors, thereby improving the robustness of the identification.

[0029] like Figure 3 As shown, this application also provides an optional method embodiment for obtaining the target fusion result, the method embodiment including: Step S301: The operating parameters of the two target power devices are fused through the basic credibility allocation function, and the fusion result is processed by conflict dispersion according to the conflict factor correction mechanism to obtain the initial fusion result; Step S302: The initial fusion result is optimized using a hybrid algorithm of particle swarm optimization and gray wolf optimization to obtain the target fusion result.

[0030] The fusion of the operating parameters of two target power devices through a basic credibility allocation function means that for any two target power devices (such as circuit breaker A and transformer B) on any faulty power grid branch, their operating parameter sequences, such as voltage amplitude, current phase, zero-sequence component, and transient overvoltage duration, are extracted respectively. After sliding window filtering and outlier removal, they are mapped to the same identification framework (i.e., the set of all possible fault states of the power grid branch, such as {normal, short circuit, open line, ground fault, equipment aging}).

[0031] Subsequently, based on the support strength of each parameter for different fault states, a basic probability assignment (BPA) function is constructed, denoted as... and ,in ; Two independent sources of evidence and The fusion can be represented as: The above formula reflects the cumulative consistency of evidence from different sources at their intersection; that is, when the intersection of two subsets is... At that time, their product was included in Credibility.

[0032] Optionally, the initial fusion result is obtained by performing conflict dispersion processing on the fusion result according to the conflict factor correction mechanism, including: An identification framework encompassing all possible fault states of the target power grid is constructed. Based on the conflict factor correction mechanism, the operating parameters with conflicts exceeding the conflict threshold in the fusion results are distributed to different subsets of the identification framework to obtain the initial fusion results.

[0033] Traditional DS evidence theory is prone to distorting the fusion results when highly conflicting evidence is present. This invention addresses this by introducing a conflict factor correction mechanism, which adjusts the conflict level... Adaptive allocation is performed to reasonably distribute conflicting parts to different subsets of the identification framework, thereby effectively mitigating the negative impact of conflicting evidence.

[0034] To further improve the accuracy and convergence speed of the fusion, this application employs a hybrid algorithm combining Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) to optimize the BPA parameters. PSO simulates the cooperative search of particles in a swarm, achieving global optimization through iterative updates of velocity and position. Its iterative formula is as follows: in, For particle velocity, For location, For the individual's optimal position, To be the globally optimal position For inertial weights, As a learning factor, The result is a random number. The advantage of PSO is its fast convergence speed, but it is prone to getting trapped in local optima.

[0035] GWO simulates the social hierarchy and hunting behavior of gray wolves, utilizing... The core update formula for the three types of individuals guiding the search direction is: in, This is the current solution vector. These represent the three currently optimal individuals. The weight parameters are dynamically adjusted. GWO's advantage lies in its strong global exploration capability, but its convergence speed is relatively slow. By combining the fast convergence characteristics of PSO with the global search capability of GWO, this application achieves efficient and stable parameter optimization, thereby ensuring the accuracy and real-time performance of multi-source fusion diagnosis.

[0036] When a fault occurs in the target power grid branch, the system can receive synchronous operating parameters from two or more power devices (such as adjacent bus PTs, line CTs, and switch status auxiliary contacts) on the faulty power grid branch. These parameters are first converted into a multi-source BPA representation of the fault state space. When a significant difference in the support strength of different device parameters for the same fault type is detected (e.g., one device displays a voltage drop while another does not respond), the system automatically triggers conflict factor calculation. If the threshold is exceeded, the conflict quality is distributed across multiple reasonable fault hypotheses to avoid over-concentration of diagnostic conclusions on a single state. Subsequently, a hybrid optimization algorithm searches for the optimal combination of BPA parameters within a limited number of iterations (e.g., 100 generations). The output target fusion result retains the physical meaning of the original data while significantly reducing the risk of misjudgment due to sensor bias, communication delay, or model mismatch. Therefore, this solves the information conflict problem caused by spatiotemporal asynchrony, dimensional differences, and measurement errors in the operating parameters of multi-source power devices, improving the accuracy and robustness of identifying complex fault modes (such as high-resistance grounding and intermittent arc short circuits).

[0037] Optionally, the method further includes: Each electrical energy reading is verified and compressed, and the processed readings are cached and managed through a FIFO queue mechanism.

[0038] Among them, data verification of each power reading refers to performing multi-dimensional consistency verification on the original power readings in accordance with the power industry communication protocol specifications (such as DL / T645–2007, DLMS / COSEM or IEC 62056 series standards), including but not limited to: frame header / frame tail identifier verification (such as 0x68, 0x16), length field comparison with actual load length, address field legality judgment (such as whether the meter address is in the pre-registered whitelist), data field CRC16 or MD5 digest verification, timestamp monotonically increasing check, and power value reasonableness threshold judgment (such as daily increment not exceeding rated capacity × 24h).

[0039] Compression processing refers to the lossless or quasi-lossless reduction of the volume of original data while ensuring the semantic integrity and traceability of electricity readings. Specific implementations may include: using differential encoding (DeltaEncoding) to eliminate redundancy in continuous time-series readings, followed by further compression using run-length encoding (Run-Length Encoding) or dictionary encoding (LZ77); replacing fixed-length integers with variable-length integer encoding for structured fields (such as meter ID and timestamps); and replacing repeated storage of recurring protocol template fields (such as fixed message headers and unit identifiers) with template extraction and index referencing. As an optional implementation, the compression algorithm supports hot-swappable configuration: the Zstandard (zstd) algorithm is enabled by default (balancing compression ratio and decompression speed), and it can be switched to the Smaz lightweight algorithm, which is more suitable for embedded terminals, or the Brotli algorithm with a higher compression ratio can be enabled on the main station side. All compression formats retain the original data metadata and verification tags, ensuring that the original metering value can be 100% restored after decompression.

[0040] The processed readings are cached and managed using a FIFO queue mechanism. This means that data packets that have completed verification and compression are written sequentially into a first-in-first-out (FIFO) memory queue with a fixed depth or dynamic scaling capability, according to their arrival time. The queue can be implemented based on a circular buffer, supporting atomic enqueue / dequeue operations and avoiding lock contention. The queue depth can be dynamically configured according to system resources (e.g., default 1024 records, maximum support 8192 records). When the queue is full, a backpressure mechanism is triggered, returning a "busy" status signal to the upstream acquisition unit, suspending the injection of new data instead of discarding data. As an optional implementation, the FIFO queue supports priority enhancement extension: an emergency flag bit is superimposed on the basic FIFO, allowing high-priority meter reading requests (such as real-time calls for suspected electricity theft users) to be inserted at the head of the queue, achieving low-latency, penetrating processing of critical data while maintaining the strict temporal order of regular meter reading data.

[0041] This approach avoids the problem of unreliable historical data caused by missing verification in traditional centralized meter reading, overcomes the network congestion and storage expansion problems caused by uncompressed direct transmission, and eliminates the risk of timing errors and packet loss caused by processing rate fluctuations through the FIFO mechanism. Ultimately, without increasing the burden on the central system, it significantly improves the data integrity, link adaptability and operational robustness of the meter reading system.

[0042] like Figure 4 As shown, this application also provides an optional method embodiment for generating a data index, the method embodiment including: Step S401: Perform weighted summation on each energy reading according to a preset weighting factor to obtain the target reading; Step S402: The target readings are stored in a structured manner using a relational database to generate a data index.

[0043] Where, assuming the first Each electricity meter at time The collected data is The process by which the concentrator weights and aggregates the data and uploads it to the data aggregation and forwarding unit 302 can be represented as: in, The number of electricity meters to be connected. For the first Weighting factors for each electricity meter, This represents the communication noise term. The formula shows that the centralized meter reading process not only considers the weighted contributions of multi-source data but also effectively suppresses the noise impact caused by the communication channel.

[0044] In centralized meter reading systems, target readings undergo multiple stages of forwarding (electricity meter → data acquisition terminal → solar concentrator → main station), network transmission, and database writing. This process carries the risk of data distortion due to packet loss, timing discrepancies, abnormal intermediate node caching, malicious tampering, or hardware failures. Specifically, this manifests as: duplicate entries of the same user's readings, inverted or abrupt timestamps, sudden numerical changes exceeding physically reasonable ranges (e.g., daily electricity consumption exceeding 10 times the theoretical peak), significant deviations from historical trends (e.g., zero readings during summer air conditioning load periods), and incorrect field formats (e.g., missing decimal places, incorrect unit identifiers). If these issues are not identified and intercepted before being entered into the database, they will directly contaminate the core metering data in the relational database, leading to deviations in downstream electricity bill calculations, inaccurate electricity consumption behavior analysis, ineffective energy-saving recommendations, and even user complaints or scheduling decision errors.

[0045] Therefore, the method also includes: Before the target readings are stored in a structured manner, data verification and consistency checks are performed on the target readings.

[0046] Data verification refers to verifying the integrity, authenticity, and tamper-proof nature of the target readings. For example, the data verification process may include: The SHA-256 hash algorithm is used to generate a digest value from the original byte stream of the target reading (including timestamp, meter ID, reading value, and checksum fields), and this digest is compared with the digest obtained after decrypting the digital signature attached to the message by the data aggregation and forwarding unit 302. If they do not match, it is determined that the data has been tampered with or corrupted in the transmission link. In addition, verification can also include basic format verification: parsing the JSON or XML message structure to confirm that required fields (such as `meter_id`, `reading_value`, and `timestamp`) are not empty and have compliant types (`reading_value` is a floating-point number, `timestamp` conforms to the ISO 8601 format), and their length is within a preset threshold (such as `meter_id` ≤ 16 characters, `reading_value` precision ≤ 4 decimal places). As an optional implementation, data verification can also be replaced with hash verification based on the national cryptographic SM3 algorithm, or a lightweight CRC32 verification combined with a sequence number increment mechanism can be used to achieve low-overhead verification in resource-constrained edge nodes.

[0047] Consistency checks refer to verifying the logical rationality of a target reading and its context. For example, this can include three sub-checks: The first is a timing consistency check, verifying whether the current `timestamp` is later than the timestamp of the most recent successful data entry for that meter ID, and that the time difference does not exceed a preset window (e.g., 5 minutes) to prevent replay attacks or clock drift from overwriting old data; the second is a numerical consistency check, comparing the `reading_value` with the historical mean and standard deviation of the meter over the same time period (±1 hour) for the past 7 days. If it exceeds the range of `mean ± 3σ`, it is marked as abnormal; the third is a topology consistency check, combining the operating parameters of upstream / downstream equipment in the directed graph of the power grid topology (e.g., when the corresponding circuit breaker is in "open" state, all downstream meter readings should be 0) to perform cross-source logical verification. As an optional implementation, the numerical judgment threshold in the consistency check can be dynamically adjusted: for example, by introducing a sliding window to adaptively update `mean` and `σ`, or by using an Isolation Forest model to detect outliers in batch target readings, replacing the fixed threshold rule.

[0048] The above process solves the problems of integrity loss and logical distortion that may occur during the transmission and writing of target readings, significantly improving the credibility and security of metering data, and ensuring the accuracy and reliability of subsequent application services such as user bills, electricity consumption pattern analysis and energy-saving suggestions generated based on the data.

[0049] Optionally, the method further includes: Calculate electricity charges and obtain user bills based on target readings and corresponding historical data. Based on the target reading and the corresponding historical data, machine learning algorithms are used to analyze users' electricity consumption patterns and generate optimized power dispatching decisions. Energy-saving recommendations are generated based on the target reading and the corresponding historical data.

[0050] The target reading refers to the aggregated result generated after weighted summation. The historical data corresponding to the target reading refers to a set of structured historical readings that have the same time dimension and user identifier as the current target reading, and whose time span covers at least, for example, 12 consecutive months.

[0051] Both the target reading and the corresponding historical data can be stored in the same relational database and can be retrieved through data indexing.

[0052] For example, a built-in electricity billing engine could be invoked to perform composite rule calculations on the target reading, including segmented billing, time-of-use billing, power factor adjustment fees, and basic electricity fees, based on current national or local electricity pricing policies. This could include identifying the billing cycle to which the target reading belongs, extracting the cumulative electricity consumption for each time period (peak, flat, valley, and peak) within that cycle, matching the corresponding time-of-use electricity price standard, overlaying the power factor adjustment coefficient and government fund surcharges, and finally generating a structured bill file containing detailed items, total amount, payment deadline, and electronic invoice QR code, which would then be pushed to the user's app, SMS platform, or business hall system. Optionally, the electricity billing logic could be replaced with a pluggable rule engine that supports dynamic electricity pricing interfaces, allowing operators to update rate parameters and billing formulas online through a configuration interface without modifying the underlying code.

[0053] Alternatively, an engineering pipeline for electricity consumption behavior characteristics can be constructed, extracting time-series statistical features (daily average, peak-to-valley difference, load factor), periodic features (weekly cycle pattern, monthly trend slope), abrupt change features (load jump amplitude, duration), and correlation features (temperature rise load elasticity coefficient, holiday offset) from target readings and historical data. These features are then input into a trained supervised or unsupervised model. The machine learning algorithms employed include, but are not limited to: clustering algorithms for user load profiling and grouping; classification models for predicting the direction and amplitude of short-term (e.g., 15–60 minutes) load fluctuations; and time-series prediction models for generating refined load curves for a future time period (e.g., 24–72 hours). Based on the above analysis results, optimized power dispatching decisions are generated.

[0054] Dispatch decisions include, for example, pushing executable dispatch commands to the dispatch center such as revised load forecasts for different zones, suggested lists of users invited to demand response (sorted by response potential), energy storage charging and discharging timing instructions, and suggestions for coordinated adjustment of distributed photovoltaic output. As a variant, electricity consumption pattern analysis can also employ an online learning mechanism, with model parameters continuously fine-tuned as new target readings arrive, ensuring adaptive tracking capability against user behavior drift.

[0055] For example, based on electricity consumption pattern analysis results and user profile tags, a rule engine or collaborative filtering recommendation system can be triggered to generate personalized, actionable, and evidence-based energy-saving intervention plans. For instance, for commercial users experiencing peak midday activity, it is recommended to raise the air conditioning setting temperature from 26℃ to 28℃, and the corresponding theoretical energy savings and electricity cost savings estimates are provided; for residential users active at night, it is recommended to extend the electric vehicle charging period from 22:00–24:00 to 00:00–02:00, and to incorporate time-of-use pricing arbitrage calculations; for industrial users with a load factor consistently below 30%, the risk of equipment operating under no-load conditions is highlighted, and motor energy efficiency diagnostics are recommended. All recommendations are accompanied by explanations of the basis (e.g., "based on a comparison of your average daily electricity consumption curve over the past three months"), an implementation difficulty rating (low / medium / high), and an expected energy-saving effect range (kWh / month, yuan / month).

[0056] The target readings serve as a unified data base, supporting three applications simultaneously: electricity billing (emphasizing metering accuracy and policy compliance), electricity consumption pattern analysis (emphasizing the depth of time-series modeling and the foresight of decision-making), and energy-saving suggestion generation (emphasizing problem traceability and the operability of measures). Historical data not only provides a statistical benchmark but also enables cross-domain data linkage between "diagnosis, meter reading, and energy efficiency" through fault event marking. Meanwhile, the structured index of the relational database (three-dimensional modeling based on user ID + timestamp + event type) ensures data consistency and query efficiency during concurrent access by multiple tasks.

[0057] Through the above-described steps, this application achieves the following: First, it deeply mines the intrinsic value of electricity data beyond simply collecting and storing it in a structured manner, overcoming the limitations of traditional meter reading systems that only record data without analysis or use it. Second, it solves the problems of long calculation cycles, high error rates, and lagging policy adaptation associated with manual calculations through automated electricity billing calculations, improving the response speed and compliance of billing services. Third, it overcomes the technical bottlenecks of low load forecasting accuracy and fragmented demand response resources through machine learning-driven electricity consumption pattern analysis, providing high-confidence load forecast curves and flexible adjustment resource pools for day-ahead / intra-day dispatching plans. Fourth, it addresses the lack of targeted energy efficiency management and limited intervention methods on the user side through data-driven personalized energy-saving suggestions, promoting active user participation in demand-side response and forming a new energy service ecosystem of two-way interaction between the power grid and users, thereby comprehensively improving the service intelligence level and energy utilization efficiency of the smart grid.

[0058] like Figure 5 As shown, Figure 5 A structural block diagram of a smart grid management system 1000 provided in this application is shown below. Figure 5 As shown, the system includes a data acquisition module 100, a fault diagnosis module 200, and a centralized meter reading module 300.

[0059] The data acquisition module 100 is used to collect the operating parameters of each power device in the target power grid and the power readings of each power meter; The fault diagnosis module 200 is used to filter faulty power grid branches from the target power grid based on the operating parameters of the power equipment and the directed graph of the power grid topology, and to determine the fault diagnosis results of each faulty power grid branch according to the target fusion result of the operating parameters of the target power equipment. The target power equipment is all the power equipment on the faulty power grid branch. The centralized meter reading module 300 is used to aggregate the various electricity readings, store the aggregation results in a structured manner through a relational database, generate a data index, and provide application support based on the aggregation results and historical data.

[0060] Optionally, the fault diagnosis module 200 is further configured to perform noise reduction processing on the operating parameters of each power device and the energy readings of each energy meter using a sliding window-based mean filtering method and an outlier removal strategy.

[0061] like Figure 6 As shown, Figure 6 The structural block diagram of another smart grid management system 1000 provided in this application is shown, wherein the fault diagnosis module 200 includes a phasor measurement unit 201; The phasor measurement unit 201 is used to determine the voltage at both ends of each power grid branch of the target power grid from the operating parameters of each of the power devices, and to calculate the voltage phase angle difference based on the voltage at both ends of each power grid branch. Each voltage phase angle difference is compared with a preset threshold, and the power grid branch whose comparison result is the target result is determined as the faulty power grid branch, wherein the target result is that the voltage phase angle difference exceeds the preset threshold.

[0062] like Figure 7 As shown, Figure 7 The structural block diagram of another smart grid management system 1000 provided in this application shows that the fault diagnosis module 200 further includes a diagnosis unit 202; The diagnostic unit 202 is used to fuse the operating parameters of the two target power devices in pairs through a basic confidence allocation function, and to perform conflict dispersion processing on the fusion result according to the conflict factor correction mechanism to obtain an initial fusion result; The initial fusion result is optimized using a hybrid algorithm combining particle swarm optimization and gray wolf optimization to obtain the target fusion result.

[0063] Optionally, the diagnostic unit 202 is specifically used to construct an identification framework that includes all possible fault states of the target power grid, and to distribute the operating parameters in the fusion result with conflicts higher than the conflict threshold to different subsets of the identification framework according to the conflict factor correction mechanism.

[0064] like Figure 8 As shown, Figure 8 The structural block diagram of another smart grid management system 1000 provided in this application is shown, wherein the centralized meter reading module 300 includes a data management unit 301; The data management unit 301 is used to perform data verification and compression processing on each of the power readings, and to cache and manage the processed readings through a FIFO queue mechanism.

[0065] like Figure 9 As shown, Figure 9 This is a structural block diagram of another smart grid management system 1000 provided in this application. The centralized meter reading module 300 further includes a data aggregation and forwarding unit 302 and a storage index unit 303. The data aggregation and forwarding unit 302 is used to perform weighted summation on each of the energy readings according to a preset weighting factor to obtain a target reading, and forward the target reading to the storage index unit 303. The storage index unit 303 is used to perform structured storage of the received target reading through the relational database to generate the data index.

[0066] Optionally, the storage index unit 303 is further configured to perform data verification and consistency check processing on the target reading.

[0067] like Figure 10 As shown, Figure 10 The structural block diagram of another smart grid management system 1000 provided in this application shows that the centralized meter reading module 300 further includes an analysis unit 304; The analysis unit 304 is used to calculate the electricity bill based on the target reading and the historical data corresponding to the target reading, and obtain the user bill. Based on the target reading and the historical data corresponding to the target reading, machine learning algorithms are used to analyze the user's electricity consumption patterns and generate optimized power dispatching decisions. Energy-saving recommendations are generated based on the target reading and the corresponding historical data.

[0068] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A power grid fault diagnosis and centralized meter reading method based on a multi-agent system, characterized in that, The method comprises: Collecting the operation parameters of each power device in the target power grid and the electric energy readings of each electric energy meter; Filtering fault power grid branches from the target power grid based on the power device operation parameters and the power grid topology directed graph, and determining the fault diagnosis results of each fault power grid branch according to the target fusion results of the operation parameters of the target power device, the target power device being all power devices on the fault power grid branch; Aggregating each electric energy reading, structurally storing the aggregation results in a relational database, generating a data index, and applying support based on the aggregation results and historical data.

2. The method of claim 1, wherein, The method further comprises: Using a mean filtering method based on a sliding window and an outlier rejection strategy to denoise the operation parameters of each power device and the electric energy readings of each electric energy meter.

3. The method of claim 1, wherein, The filtering of fault power grid branches from the target power grid based on the power device operation parameters and the power grid topology directed graph comprises: Determining the voltages at both ends of each power grid branch of the target power grid from the operation parameters of each power device, and calculating the voltage phase angle difference according to the voltages at both ends of each power grid branch; Comparing each voltage phase angle difference with a preset threshold, and determining the power grid branch with a target result as the fault power grid branch, the target result being that the voltage phase angle difference exceeds the preset threshold.

4. The method of claim 1, wherein, The method further comprises: Fusing the operation parameters of each pair of target power devices by a basic credibility allocation function, and dispersing conflicts in the fusion results according to a conflict factor correction mechanism to obtain an initial fusion result; Optimizing the initial fusion result by a hybrid algorithm of particle swarm optimization and grey wolf optimization to obtain the target fusion result.

5. The method of claim 4, wherein, The dispersing of conflicts in the fusion results according to the conflict factor correction mechanism to obtain the initial fusion result comprises: Constructing an identification framework containing all possible fault states of the target power grid, and dispersing the operation parameters with conflicts higher than a conflict threshold in the fusion results to different subsets of the identification framework according to the conflict factor correction mechanism to obtain the initial fusion result.

6. The method of claim 1, wherein, The method further comprises: Performing data verification and compression processing on each electric energy reading, and performing buffer management on the processed readings by a FIFO queue mechanism.

7. The method of claim 1, wherein, The structurally storing of the aggregation results in a relational database to generate a data index comprises: Weighted sum processing of each electric energy reading according to a preset weight factor to obtain a target reading; Structurally storing the target reading in the relational database to generate the data index.

8. The method of claim 7, wherein, The method further comprises: Before structurally storing the target reading, performing data verification and consistency check processing on the target reading.

9. The method of claim 1, wherein, The method further comprises: Calculating electricity charges according to the target reading and historical data corresponding to the target reading to obtain user bills; Analyzing the user's power consumption mode by a machine learning algorithm according to the target reading and historical data corresponding to the target reading to generate an optimized power dispatching decision; Generating energy saving suggestions according to the target reading and historical data corresponding to the target reading.

10. A smart grid management system, characterized by, The system comprises: A data collection module is configured to collect operation parameters of each power device in a target power grid and electric energy readings of each electric energy meter. A fault diagnosis module is configured to filter fault power grid branches from the target power grid based on the operation parameters of the power devices and a power grid topology directed graph, and determine fault diagnosis results of each of the fault power grid branches according to a target fusion result of operation parameters of target power devices, the target power devices being all power devices on the fault power grid branches. A centralized meter reading module is configured to aggregate each of the electric energy readings, store an aggregation result in a structured manner through a relational database, generate a data index, and provide application support based on the aggregation result and historical data.

Images