An AI-based meter reading data management and resource scheduling optimization method and system

Abstract

The present invention discloses a method and system for meter reading data management and resource scheduling optimization based on AI, and relates to the technical field of intelligent management of meter reading data. The method comprises: obtaining meter reading data of a meter, wherein the meter has a corresponding metal identification piece with a QR code; managing the meter reading data according to geographical region and user type to form a multi-level data feature system, wherein the user type includes residential type and industrial type; using an AI model to perform trend modeling on the multi-level data feature system to determine the energy consumption trend in the future cycle; dynamically adjusting the meter reading task according to the energy consumption trend and the network load to dynamically schedule human resources, data resources, network resources and hardware resources; wherein the grid topology position coding is used to bind the relationship between the meter reading task and the physical conduction path of the grid. The present application realizes the full-link optimization from meter reading data collection to resource scheduling.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent management of meter reading data, and specifically to an AI-based meter reading data management and resource scheduling optimization method and system. Background Art

[0002] In daily life, meter reading is a fundamental task, regularly collecting energy meter usage data. Meter reading data plays a key role in energy management and forms the core basis for billing. Based on individual users' electricity, water, or gas consumption and corresponding billing standards, accurate calculations of user fees are possible, ensuring reasonable energy billing.

[0003] Traditional manual meter reading requires a meter reader to visit the user's premises at regular intervals, observing the meter dial and recording the data. However, this method requires a large number of meter readers to collect data door-to-door. This results in low efficiency and high labor costs in scenarios where meters are widely distributed or numerous.

[0004] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0005] The embodiments of the present application provide an AI-based meter reading data management and resource scheduling optimization method and system to solve the above technical problems.

[0006] According to one aspect of an embodiment of the present application, an AI-based meter reading data management and resource scheduling optimization method is provided, including: obtaining meter reading data of a meter, the meter having a corresponding QR code metal identification plate, the QR code on the QR code metal identification plate storing the identification of the meter and the grid topology location code; managing the meter reading data according to geographical area and user type to form a multi-level data feature system, the user type including residential type and industrial type; using an AI model to perform trend modeling on the multi-level data feature system to determine the energy usage trend in future cycles; dynamically adjusting the meter reading task according to the energy usage trend and network load conditions to dynamically schedule human resources, data resources, network resources and hardware resources; wherein, the identification of the meter is used to generate a data retransmission token, and the grid topology location code is used to bind the association between the meter reading task and the physical conduction path of the grid.

[0007] According to another aspect of an embodiment of the present application, an AI-based meter reading data management and resource scheduling optimization system is provided, including: a meter reading data acquisition module, used to obtain meter reading data of a meter meter, the meter meter having a corresponding QR code metal identification plate, the QR code on the QR code metal identification plate storing the identification of the meter meter and the grid topology location code; a meter reading data management module, used to manage the meter reading data according to geographical area and user type to form a multi-level data feature system, the user type including residential type and industrial type; a usage trend determination module, used to use an AI model to perform trend modeling on the multi-level data feature system to determine the energy usage trend in the future cycle; a resource scheduling optimization module, used to dynamically adjust the meter reading task according to the energy usage trend and network load conditions to dynamically schedule human resources, data resources, network resources and hardware resources; wherein, the identification of the meter meter is used to generate a data retransmission token, and the grid topology location code is used to bind the association relationship between the meter reading task and the physical conduction path of the grid.

[0008] According to another aspect of an embodiment of the present application, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the above-mentioned AI-based meter reading data management and resource scheduling optimization method through the computer program.

[0009] Based on the embodiments provided in this application, by binding the grid topology location code with the meter reading task and combining a multi-dimensional data feature system with AI trend modeling, full-link optimization from meter reading data collection to resource scheduling is achieved. The advantage lies in integrating the physical characteristics of the power system with the business management process, providing a smarter and more efficient solution for energy management. Specifically:

[0010] The grid topology location code stored in the QR code on the metal identification tag ties meter reading tasks to the physical transmission path of the power grid, aligning meter reading route planning with power transmission routes, reducing wasted trips and improving human resource utilization efficiency. Data retransmission tokens are generated using meter identification to accurately locate and reissue missing data when network transmission anomalies occur. Combined with the meter reading task association established based on the grid topology location code, this ensures the integrity of data collection and transmission, reducing manual review costs. A multi-level data feature system is constructed based on geographic region and user type (residential / industrial). This not only preserves spatial load density and device distribution characteristics, but also captures the differences in load curve shape across different user types, providing more comprehensive feature input for energy usage trend forecasting. Based on AI-predicted energy usage trends and network load, meter reading tasks and resource allocation (manpower, hardware, network, etc.) are dynamically adjusted. For example, data collection in key areas is prioritized during high-load periods, improving overall system responsiveness and resource utilization. Integrating the physical conduction path information of the power grid into meter reading task management links business scheduling with the actual structure of the power grid, providing more direct data support for subsequent power system analysis (such as fault location and load balancing), and breaking the separation between traditional meter reading and power grid operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the embodiments of the present invention and constitute a part of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0012] Figure 1 Flowchart of an optional AI-based meter reading data management and resource scheduling optimization method according to an embodiment of the present application;

[0013] Figure 2 Flowchart of another optional AI-based meter reading data management and resource scheduling optimization method according to an embodiment of the present application;

[0014] Figure 3 A flowchart of an optional dynamic task adjustment and digital twin modeling according to an embodiment of the present application;

[0015] Figure 4 This is a structural diagram of an optional AI-based meter reading data management and resource scheduling optimization system according to an embodiment of the present application.

[0016] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0018] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] According to one aspect of the embodiments of the present application, a method is provided, such as Figure 1 As shown, the method includes:

[0020] S101, obtaining meter reading data of a meter, the meter having a corresponding metal identification plate with a QR code, the QR code on the metal identification plate storing an identification of the meter and a grid topology location code;

[0021] In some embodiments of the present application, meters include, but are not limited to, active energy meters, reactive energy meters, and multi-function energy meters. The meters have built-in operator-certified IoT communication modules. The meter's identification is stored via a QR code on a metal identification plate and is bound to the operator's IMSI number on the SIM card. Before data transmission, the operator's two-way authentication mechanism (such as the AKA protocol) verifies the device's legitimacy.

[0022] The QR code metal identification sheet is a corrosion-resistant metal sheet attached to the meter. Its QR code contains the meter identification (such as MAC address) and the grid topology location code (the format can be area code + line number + node number).

[0023] Methods for acquiring meter reading data include periodic and event-driven acquisition. Exemplarily, the meter acquires meter reading data through a built-in data acquisition module. This module collects real-time electricity parameters (such as voltage, current, and power) and cumulative usage data, and generates raw data frames with timestamps. The meter integrates a carrier-certified IoT communication module and connects to the operator's customized power communication channel for the power industry via a SIM card. The data acquisition module initiates acquisition based on a preset cycle (e.g., once every hour) or event triggers (e.g., sudden changes in usage). Real-time and cumulative usage data are channel-coded and transmitted to the power system master station via the carrier's base station.

[0024] Optionally, the meter has a built-in cache module to temporarily store data when the network is interrupted, and resend it through the power communication channel after communication is restored to ensure data integrity. In some embodiments of the present application, in industrial parks with dense equipment, the grid topology location code is bound to the transmission trunk branch node, and the equipment distribution gradient is superimposed during real-time load density calculation to dynamically adjust the grid code weight. In residential areas, the grid topology location code is associated with the distribution box level, and high-energy-consuming appliances (such as air conditioning clusters) are marked through a label tree to generate a fine-grained load curve. For example, in an industrial park, each meter's QR code contains a device ID (such as "IND-Z3-205") and a topology code (such as "Grid-Branch-12"). The latter corresponds to the park's No. 12 transmission trunk branch. The park is divided into 20 grids (G01-G20). Because the G05 grid contains three high-voltage lines, the equipment distribution density gradient is set to 0.9. The label tree is divided into the "chemical plant → electrolysis workshop → reactor" hierarchy to extract the minute-level load mutation characteristics of the reactor. The harmonic characteristics (hourly level) of the G05 grid are associated with the reactor load curve, the equipment density is corrected, and it is marked as a high-risk area.

[0025] S102, managing meter reading data by geographical region and user type to form a multi-level data feature system, where user types include residential and industrial types;

[0026] S103 uses AI models to perform trend modeling on the multi-layered data feature system to determine energy usage trends in future cycles;

[0027] Optionally, future cycles include but are not limited to different time scales such as hourly level (such as real-time load control), daily level (such as peak and valley period scheduling), weekly level (such as weekly equipment inspection planning), monthly level (such as monthly energy consumption trend forecast), quarterly / annual level (such as seasonal resource allocation), etc., which are dynamically set according to energy scheduling accuracy and management needs.

[0028] Among them, AI models may include but are not limited to any one or more of machine learning models, deep learning models or reinforcement learning models.

[0029] The deep learning model may include, but is not limited to, any one or more of a recurrent neural network, a graph neural network, a transformer model, or a spiking neural network. The spiking neural network may include a causal neural field spiking network model and / or a leaky integral spiking neural network.

[0030] Machine learning models include, but are not limited to, random forests, gradient boosted trees, and hidden Markov models.

[0031] Reinforcement learning models include but are not limited to deep reinforcement learning models (such as deep deterministic policy gradient models and deep Q networks); reinforcement learning algorithms based on policy gradients.

[0032] In a specific embodiment, the AI ​​model adopts a hidden Markov model; the AI ​​model is used to perform trend modeling on a multi-layered data feature system to determine the energy usage trend in future cycles, including: by constructing a probabilistic model including hidden states and observation sequences, to explore the energy consumption pattern transfer rules in different geographical regions and different user types.

[0033] The specific implementation steps are as follows: First, historical meter reading data for residential and industrial users is divided into independent subsets by geographical region. For each subset, a hidden state reflecting the user's electricity usage characteristics is defined (such as "daily electricity usage" and "holiday electricity usage" for residential users, and "peak production period" and "equipment maintenance period" for industrial users). Then, the meter reading data for each time period is used as an observation sequence, and the state transition probability and observation probability matrix are learned through hidden Markov model training. Finally, the dynamic changes of the hidden state are inferred based on the current observation data to achieve trend prediction of energy usage in future cycles.

[0034] It should be noted that the AI ​​model may also employ a causal neural field spike network model, a deep deterministic policy gradient model, or the like, which is not limited in this embodiment. A specific embodiment of employing a causal neural field spike network model to perform trend modeling on a multi-layered data feature system to determine energy usage trends in future cycles will be described later.

[0035] S104, dynamically adjust meter reading tasks based on energy usage trends and network load conditions to dynamically schedule human resources, data resources, network resources, and hardware resources; the meter identifier is used to generate a data retransmission token, and the grid topology location code is used to bind the relationship between the meter reading task and the physical transmission path of the grid.

[0036] In this embodiment, regarding token generation and path selection, when an industrial electricity meter (ID: E-05-3A) loses data due to a network outage, a retransmission token 0x01|GZ-3B→4A|E-05-3A is generated based on its grid topology location code "GZ-110kV-3B." The topology path selection rules are as follows: Priority determination: Based on the risk level of the device health warning period, high-risk device tokens are marked with 0x01 (highest priority). Path optimization: Grid topology coefficients are used to calculate the relay capabilities of adjacent grids (for example, the transmission impedance of grid 4A is 15% lower than that of 3B) and select the optimal relay path.

[0037] The token execution process involves sending a token to the target base station via the SDN controller, triggering the meter in grid 4A to act as a relay node and establish a temporary communication channel. The relay node locates the faulty meter based on the device ID in the token and transmits data in a targeted manner using the LoRa protocol, reducing communication time.

[0038] The data retransmission token is bound to the physical conduction path of the power grid to ensure that the data recovery path matches the physical line impedance and improve the relay success rate.

[0039] In this embodiment, the data retransmission token is a dynamically generated encrypted string containing the device identifier, retransmission priority, and topological path information. It is used to restore meter reading data after a network interruption. The token generation rules include: Priority field: Calculated based on the risk level of the device health warning period (high risk = 0x01, medium risk = 0x02, low risk = 0x03). Topological path: The optimal relay path is generated based on the grid topology location code (e.g., "GZ-110kV-3B → GZ-110kV-4A").

[0040] For example, smart meters deployed in an industrial park upload meter readings every hour. QR code parsing yields the device ID "E-05-3A" and the topology code "GZ-110kV-3B." Based on the topology code "GZ-110kV-3B," the meter is classified as belonging to grid 3B of a local 110kV power grid. The label tree identifies the meter as belonging to the "industrial user → manufacturing → high-energy-consuming equipment" category. An AI model analyzes 30 consecutive days of industrial load data for this grid and predicts that the peak load for the next month will occur between 10:00 AM and 12:00 PM on weekdays. Monitoring of the base station load rate in this area reaches 85%, triggering a scheduling policy. Automatic meter readings originally scheduled for peak hours are postponed to the nighttime off-peak hours, and data from faulty meters is prioritized using a data retransmission token (containing the device ID).

[0041] Based on the embodiments provided in this application, by binding the grid topology location code with the meter reading task and combining a multi-dimensional data feature system with AI trend modeling, full-link optimization from data collection to resource scheduling is achieved. The advantage lies in integrating the physical characteristics of the power system with the business management process, providing a smarter and more efficient solution for energy management. Specifically:

[0042] The grid topology location code stored in the QR code on the metal identification tag ties meter reading tasks to the physical transmission path of the power grid, aligning meter reading route planning with power transmission routes, reducing wasted trips and improving human resource utilization efficiency. Data retransmission tokens are generated using meter identification to accurately locate and reissue missing data when network transmission anomalies occur. Combined with the meter reading task association established based on the grid topology location code, this ensures the integrity of data collection and transmission, reducing manual review costs. A multi-level data feature system is constructed based on geographic region and user type (residential / industrial). This not only preserves spatial load density and device distribution characteristics, but also captures the differences in load curve shape across different user types, providing more comprehensive feature input for energy usage trend forecasting. Based on AI-predicted energy usage trends and network load, meter reading tasks and resource allocation (manpower, hardware, network, etc.) are dynamically adjusted. For example, data collection in key areas is prioritized during high-load periods, improving overall system responsiveness and resource utilization. Integrating the physical conduction path information of the power grid into meter reading task management links business scheduling with the actual structure of the power grid, providing more direct data support for subsequent power system analysis (such as fault location and load balancing), and breaking the separation between traditional meter reading and power grid operation and maintenance.

[0043] As an optional solution, meter reading data can be managed by geographical region and user type to form a multi-level data feature system, including:

[0044] Construct a dynamic index system based on geographic region, user type, and time dimensions. The geographic region dimension is divided into grid codes according to the power grid topology, and the user type dimension is divided according to the label tree hierarchy.

[0045] In this embodiment, the dynamic indexing system utilizes a combined indexing mechanism consisting of geographic grid codes (spatial indexing), user tag trees (semantic indexing), and time slices (temporal indexing), supporting fast retrieval of multi-dimensional features. The tag tree hierarchy is structured as a user classification tree, with the root node being Residential / Industrial, and child nodes broken down into industry type, device configuration, and more.

[0046] Label tree hierarchy: A tree-like classification structure with the root node being "Residential" or "Industrial," and child nodes being further refined by industry and equipment type. Traversal rules: Residential users are categorized into three levels based on their electricity usage habits: "Conventional" / "Peak / Off-Peak" / "Energy Saving"; Industrial users are categorized by industry, process, and equipment type (e.g., "Chemical Industry" / "Electrolysis" / "Rectifier").

[0047] For example, a user's tag path is "Industry → Chemical Industry → Electrolysis → Rectifier". The load curve shows periodic peaks (high load for 10 minutes once every hour), which is associated with the batch production mode of the electrolysis process.

[0048] The user type dimension is divided into label tree levels, and differentiated collection frequency strategies are determined based on the load curve shape;

[0049] Meter reading data is hierarchically reorganized based on a dynamic indexing system. In terms of geographic region, meter reading data is spatially clustered by grid code to generate regional features including real-time load density and device distribution density. In terms of user type, user features are matched through a label tree traversal to generate type features including load curve morphology and device configuration identification.

[0050] The reorganized meter reading data is multi-scale associated with the time dimension to generate a multi-level data feature system including a basic data layer, a dimensional feature layer, and a cross-dimensional fusion layer.

[0051] In a specific implementation, the index is constructed as follows: Geographic Grid: A city's power grid is divided into 10 main grids (e.g., 110kV-A Zone) based on voltage level. Each main grid is subdivided into 100 subgrids (coded in the format A-01 to A-100). User Tag Tree: A four-level tag tree is constructed: Level 1: Residential / Industrial, Level 2 (Industrial): Manufacturing / Commercial / Public Utilities, Level 3 (Manufacturing): Automotive / Electronics / Chemical, and Level 4: Equipment Type (e.g., air compressor / furnace). Spatial Clustering: Data from 32 electricity meters are clustered in subgrid A-23, and the real-time load density is calculated as 58kW / km² and the equipment distribution density is 15 units / km. Tag Matching: The tag tree is traversed to identify a user as "Industry → Manufacturing → Electronics → SMT Machine." The load curve is extracted, showing a periodic pulse pattern (peaks every 15 minutes during working hours). Temporal Correlation: The load data in grid A-23 is sliced ​​and calibrated by hour, and a temperature anomaly code for the day (e.g., "T-35°C") is injected to mark a device overload event.

[0052] Based on the embodiments provided by this application, a dynamic index system of geographical regions, user types, and time dimensions is constructed to achieve multi-dimensional structured management of meter reading data. Specifically, the geographical region dimension is divided into grid codes according to the power grid topology, so that data clustering fits the physical structure of power transmission (for example, the meters in the power supply area of ​​the same substation are classified into the same grid), avoiding the problem of traditional geographical division ignoring the connection relationship of the power grid, and facilitating the rapid location of regional load density and equipment distribution characteristics; the user type dimension is divided into hierarchical levels through a label tree (such as the difference in load curves for residential and industrial users), and the power consumption characteristics of different users can be extracted layer by layer (for example, the equipment configuration identifier of industrial users reflects the distribution of high-energy-consuming equipment), providing a basis for differentiated scheduling; the hierarchically reorganized data forms a coupling relationship of "regional features-type features". For example, combining the real-time load density of a grid with the equipment configuration identifier of the industrial users in the area can more accurately identify the cause of high load and provide more granular input features for subsequent AI trend modeling.

[0053] As an optional solution, the reorganized meter reading data is multi-scale associated with the time dimension to generate a multi-layered data feature system consisting of a basic data layer, a dimensional feature layer, and a cross-dimensional fusion layer, including:

[0054] Based on the time dimension, the reorganized meter reading data is calibrated into four time slices: minutes, hours, days, and months. Each time slice is injected with climate anomaly factor codes and device event flags to generate time series data blocks with spatiotemporal attributes.

[0055] In this embodiment, the climate anomaly factor encoding is a quantitative parameter representing the impact of temperature, humidity, and other factors on load. For example, "T+5" indicates that the temperature is 5°C higher than the historical average. The device event flag is a binary or multi-encoded field used to identify operational anomalies or status changes that occurred in a meter or associated device within a specific time slice.

[0056] In this application, device event flags use a 4-bit encoding scheme: Bit 1 (overload flag): 1 indicates the current exceeds the rated value by 15% for more than 10 minutes; Bit 2 (temperature anomaly): 1 indicates the device case temperature is >70°C; Bit 3 (communication interruption): 1 indicates three consecutive meter reading data upload failures; Bit 4 (harmonic excess): 1 indicates total harmonic distortion (THD) ≥8%. For example, if an industrial meter detects a current overload (Bit 1 = 1) and a harmonic excess (Bit 4 = 1) within an hourly time slice, device event flag 1001 is generated. The climate anomaly factor encoding indicates that the temperature on that day is "T+8" (8°C above the historical average). Combining flag 1001 with the climate code triggers the following actions: Minute-level: Extract the frequency of sudden load changes (5 times / minute) for the meter during the overload period. Hour-level: Analyze the harmonic energy distribution and locate the inverter device in the A-15 grid. Based on the flag bit priority (Bit1>Bit4>Bit3>Bit2), the operation and maintenance work order is automatically dispatched to the A-15 grid, giving priority to overloaded equipment.

[0057] Differentiated feature coupling operations are performed for different time slices. At the minute level, the frequency of load mutations is extracted to form transient features. At the hourly level, harmonic energy distribution features are generated through spectrum analysis. At the daily level, the load peak and valley attenuation rates are calculated. At the monthly level, equipment health trend vectors are generated, forming a time series feature layer that is linked to the dynamic index system.

[0058] The equipment health trend vector reflects the temporal characteristics of equipment performance degradation, and its dimensions include insulation resistance attenuation rate, harmonic distortion growth rate, etc.

[0059] The harmonic energy distribution characteristics in the time series feature layer are spatially convolved with regional characteristics. The load peak and valley attenuation rate is behaviorally correlated with the load curve shape of the user type dimension. The device distribution density is corrected using the device health trend vector to generate a three-dimensional fusion matrix.

[0060] In this embodiment, the spatial convolution operation includes the following: Convolution kernel construction: A dynamic convolution kernel is generated based on the grid topology location code, with the kernel weight determined by the device distribution density gradient. For example, the kernel size is automatically reduced in high-density areas to capture local harmonic characteristics, while the kernel size is increased in low-density areas to smooth noise. Operational mechanism: The harmonic energy distribution characteristics (hourly slices) are convolved with the real-time load density in the regional characteristics. Each element in the output matrix represents the "harmonic propagation intensity per unit device density," which is used to identify harmonic pollution sources caused by specific device groups.

[0061] When correlating the daily load peak-valley decay rate with the load curve shape for the user type dimension, a dynamic time warping algorithm is used to align the time axes and calculate the similarity between the two shapes. When an industrial user's peak-valley decay rate is abnormal, it is automatically linked to the device configuration identifier to determine whether it is caused by a device failure or the addition of a high-energy-consuming device.

[0062] Based on the three-dimensional fusion matrix, a multi-level data feature system is constructed, including the basic data layer, the dimensional feature layer and the cross-dimensional fusion layer.

[0063] In one specific implementation, a residential electricity meter was detected to have three sudden load changes between 08:15 and 08:17 (a frequency threshold alarm). Hourly-level FFT analysis of meter data in a commercial area revealed that the 23rd harmonic component exceeded the standard (total harmonic distortion (THD) = 8.5%). Daily-level analysis calculated the peak-to-valley attenuation rate of a factory's daily load, which was 67% (normal range > 50%), indicating no abnormality. Monthly-level analysis analyzed a 12% decrease in the monthly average insulation resistance of substation equipment, generating a health trend vector of [0.88, 0, 0.12] (representing resistance, temperature, and vibration health, respectively). The hourly harmonic signature of the commercial area was convolved with the device density (25 units / km) of grid A-15, identifying the harmonic source as the UPS connected to meter A-15-07. The daily peak-to-valley attenuation rate (82%) of an electronics factory was associated with the "reflow soldering machine" configuration in its tag tree, indicating that the equipment was not in energy-saving mode. Based on the health trend of the substation, the equipment density of the grid where it is located was revised from 20 units / km to 17 units / km (excluding 3 units awaiting maintenance).

[0064] Based on the embodiments provided in this application, reorganized data is correlated across multiple time scales to address the collaborative analysis of transient faults and long-term health management in electricity meter reading. Specifically, minute-level load mutation frequency captures transient disturbances (such as meter data jumps caused by lightning strikes) and, combined with grid topology coding, quickly locates whether the source of the disturbance comes from upstream lines. Hourly harmonic energy distribution uses spectrum analysis to identify harmonic pollution from nonlinear devices. After correlating with the grid characteristics of "industrial user concentration areas," it can locate the line section where the harmonic source is located (for example, the third harmonic of a 10kV line in an industrial park exceeds the standard). Daily peak-to-valley attenuation rates quantify residents' electricity usage habits (such as extending the evening peak of air conditioning load in summer) and, combined with user-type load curves, assist in predicting seasonal energy valley changes. Monthly equipment health trends use historical data to fit meter error drift or line insulation aging. For example, when the health of a substation outlet meter continues to decline, line maintenance can be scheduled in advance based on the topological location. Cross-dimensional fusion identifies risk scenarios involving "high harmonics + heavy load + equipment aging" (such as accelerated aging of industrial equipment due to harmonics), improving the accuracy of risk warnings.

[0065] As an optional solution, AI models include causal neural field spike network models. Using AI models, trend modeling is performed on a multi-layered data feature system to determine energy usage trends over future cycles, including:

[0066] Mapping the features of the basic data layer, the dimensional feature layer, and the cross-dimensional fusion layer to the nodes of the causal neural field spike network model;

[0067] The spatial field strength of geographic grid nodes is constructed based on grid coding and carries load parameters. The behavioral pulse intensity of user type nodes is generated based on the load curve shape and device configuration identification of the label tree. The periodic pattern of time nodes is modulated by the decay rate of healthy devices and load fluctuation characteristics.

[0068] In this embodiment, the behavioral pulse intensity is a parameter reflecting the intensity of the user's electricity consumption behavior. For industrial users, the behavioral pulse intensity = reference value × (maximum load / rated capacity), where the reference value is 2.0; for residential users, the reference value is 0.5, and only the average daily load is calculated.

[0069] For example, a steel plant has a maximum load of 80MW (rated capacity 100MW) and a pulse intensity of 2.0×0.8=1.6. Its nodes drive the load of adjacent grids to increase in the neural field model. The causal neural field pulse network model is trained based on historical grid topology and load data, with the optimization objective of minimizing the simulation deviation of the grid conduction path.

[0070] In this embodiment, the grid conduction path simulation deviation is the mean square error between the load diffusion path predicted by the model and the actual grid power flow. In some embodiments of the present application, the causal neural field pulse network model is designed for grid topology conduction path modeling, and its nodes are composed of three sets of parameters:

[0071] Geographic Grid Node: The grid code of the power grid topology is converted into three-dimensional spatial coordinates. The initial spatial field strength of each node is calculated by weighting the grid's historical average load density and the equipment impedance parameters (for example, high load density areas are assigned strong spatial field strength, while high impedance areas are compensated for attenuation). The transmission impedance coefficient is designed based on the line loss model specified in IEEE Std 738-2012, "Standard for Calculating Ampacity of Overhead Conductors." For example, for a 110kV overhead line, the impedance coefficient = line resistance / (line resistance + reactance), where resistance and reactance are calculated based on the conductor material (e.g., steel-core aluminum stranded conductor) and cross-sectional area.

[0072] For example, grid A-10 is a 110kV transmission line with a conductor cross-section of 300mm², a resistance of 0.12Ω / km, and a reactance of 0.4Ω / km. The impedance coefficient is 0.12 / (0.12 + 0.4) = 0.23. At a load density of 65kW / km², the spatial field strength is 65 × (1 - 0.23) = 50.05, reflecting the power transmission efficiency of this line.

[0073] User type node: Based on the label tree hierarchy, the load curve of residential / industrial users is decomposed into pulse sequences (for example, the pulse amplitude of industrial users during the morning peak is higher than that of residential users), and the device configuration identifier is converted into pulse width modulation parameters (for example, high-power devices correspond to wide pulses).

[0074] Time node: Map the equipment health trend vector to a periodic pulse attenuation factor. For example, when the health decreases, the pulse amplitude decays according to an exponential curve, simulating the impact of equipment performance degradation on load conduction.

[0075] In this embodiment, the training process of the causal neural field spike network model involves backpropagating the adjustment of field strength weights and pulse modulation parameters based on historical grid topology data, with the goal of minimizing the deviation in grid conduction path simulation. For example, when the deviation between the actual load diffusion rate in a certain area and the model output exceeds a certain threshold, the field strength coupling coefficient between that grid node and its adjacent nodes is increased.

[0076] The spatial clustering results of the grid code are directly used as initialization input for the geographic grid nodes, ensuring that the model's field strength distribution is consistent with the actual device density. The load curve morphology generated by the label tree traversal is encoded as a pulse sequence of user type nodes, strictly binding behavioral modeling to user classification characteristics.

[0077] In some embodiments, for example, node parameter design is as follows: For geographic grid nodes, the field strength calculation formula is: field strength = load density × (1 - transmission impedance coefficient). For example, grid A-10 has a load density of 65 kW / km² and an impedance coefficient of 0.2, resulting in field strength = 65 × 0.8 = 52. For user type nodes, for industrial users, pulse strength = 2.0 × (actual load / rated capacity); for residential users, it = 0.5 × average daily load. For example, for a steel plant with an 80 MW load (rated 100 MW), pulse strength = 2.0 × 0.8 = 1.6. For time nodes, the pulse attenuation factor = the leading term of the health trend vector × the time coefficient. If the monthly attenuation rate of transformer insulation resistance is 12% and the time coefficient is 0.9, the attenuation factor = 0.88 × 0.9 = 0.79.

[0078] The training process of the causal neural field pulse network model includes: when backpropagation adjusts the field strength weight, if the actual load diffusion speed of a line is 20% faster than the model, the field strength coupling coefficient between the line node and the downstream node is enhanced by 10%.

[0079] The physical conduction paths of the power grid are constructed within the causal neural field pulse network model. The load conduction paths between geographic grid nodes are dynamically modeled based on the device distribution density gradient and transmission impedance parameters, simulating the diffusion of electrical energy along the grid topology. The load impact between user-type nodes is constrained through hierarchical relationships in a label tree.

[0080] The device density gradient describes the rate of change in the density of power equipment in geographic space (unit: units / km²·m). Its functions include: serving as a core parameter in the causal neural field spike network model, driving the load conduction path simulation; the gradient direction determines the preferred channel for energy diffusion; constructing an energy potential well model for meter reading tasks within the digital twin engine; the gravitational gradient is positively correlated with the rate of change in device density, directing personnel to areas with high gradients; and enabling a high-frequency data collection strategy in areas with a gradient change rate greater than 0.5, dynamically adjusting communication bandwidth in conjunction with network load heat maps. The network model generates a three-dimensional trend map that includes high-load density areas, abnormal user groups, and equipment health warning periods to determine future energy usage trends.

[0081] For example, in model construction: Geographic Grid Node: Set the initial field strength to 1.5 for grid A-10 (based on a load density of 65 kW / km² and an impedance of 0.2 Ω / km). User Type Node: For an automobile factory, the label tree path is "Industry → Manufacturing → Automobile → Stamping Machine," generating a pulse with an amplitude of 4.2 and a width of 120 ms. Time Node: Based on the transformer health trend vector [0.85, 0.1, 0.05], set the pulse attenuation factor to 0.9 / month. Conduction Path Modeling: The equipment density gradient from grids A-10 to A-11 is +8 units / km, the transmission impedance difference is 0.1 Ω, and the modeled load conduction speed is 5 kW / min. The automobile factory node influences supporting enterprise nodes in the "Manufacturing" sector through the label tree, constraining the influence coefficient to no more than 0.3. Trend Output: Forecast a high load density area (e.g., exceeding a threshold by 15%) in grid A-10 between 2:00 PM and 4:00 PM each day for the next month. A residential load curve in a certain community was detected to be similar to an industrial pulse pattern (e.g., an abnormal correlation coefficient of 0.82), marking it as an abnormal user group. Based on the health of the transformer, an early warning period was generated: maintenance recommended from 09:00 to 12:00 on the 25th of each month.

[0082] Based on the embodiments provided in this application, a causal neural field pulse network model is employed to embed the physical conduction laws of the power grid into an AI architecture. Specifically, the spatial field strength of geographic grid nodes is constructed based on topological encoding (e.g., node field strength reflects its location in the transmission line; the closer to the substation, the less attenuation). The carried load parameters are associated with the transmission impedance (simulating line power loss), enabling the model to simulate the physical process of load diffusion from substation to trunk line to branch line. User-type node behavior pulses are generated hierarchically using a label tree (industrial users have higher pulse intensity than residential users) and are constrained by the grid topology (e.g., load fluctuations of users on a line influence each other through impedance), thus overcoming the limitations of traditional statistical models that ignore physical conduction. The time node periodic model integrates equipment health decay (e.g., the longer load fluctuation period of aging lines) with load fluctuation characteristics. For example, when predicting that an aging line may trip due to load superposition during high-temperature periods, the risk area and time period are simultaneously output. The model is trained with the goal of minimizing grid conduction path deviation, accurately simulating the voltage drop caused by line load, and providing reliable trend predictions for meter reading task scheduling.

[0083] As an optional solution, a causal neural field spike network model is used to generate a three-dimensional trend map that includes high-load density areas, abnormal user groups, and equipment health warning periods, including:

[0084] The geographic grid node load parameters, user type node behavior pulse sequence and time node cycle pattern output by the causal neural field spike network model are mapped into feature tensors of geographic area dimension, user type dimension and time dimension respectively to form an initial feature cube;

[0085] Based on the physical conduction path of the power grid, a spatiotemporal evolution operator is designed to update the initial feature cube and generate a dynamic feature cube. The load diffusion feature matrix in the geographic region dimension is driven by the device distribution density gradient; the behavioral feature propagation matrix in the user type dimension is constrained by the label tree hierarchy; and the temporal feature modulation vector in the time dimension is integrated with the healthy device attenuation rate parameter.

[0086] In some embodiments of the present application, the spatiotemporal evolution operator consists of two submodules:

[0087] Load Diffusion Module: Calculates the load conduction direction based on the equipment distribution density gradient. For every 10% increase in density difference, the conduction speed increases by 5% (simulating the natural diffusion of electrical energy to low-density areas).

[0088] Behavior propagation module: This module uses the hierarchical relationships in the tag tree to constrain the user's scope of influence. For example, an industrial user node can only influence users in the same tag subtree (such as companies in the same industrial chain), avoiding cross-industry misassociation.

[0089] In this embodiment, the dynamic feature cube is a three-dimensional data structure consisting of geography (X-axis), user type (Y-axis), and time (Z-axis), and each unit stores [load value, health level, abnormality probability].

[0090] High-load density areas, abnormal user groups, and equipment health warning periods are extracted from the dynamic feature cube to generate a three-dimensional trend map.

[0091] In some embodiments of the present application, extracting high-load density areas, abnormal user groups, and equipment health warning periods from the dynamic feature cube includes:

[0092] In the geographical dimension of the dynamic feature cube, high load density areas are identified based on the energy density distribution of the load diffusion feature matrix;

[0093] In the user type dimension of the dynamic feature cube, the behavior feature propagation matrix is ​​used to construct the electricity consumption behavior phase space, and the Lyapunov index is calculated to identify abnormal user groups.

[0094] In the time dimension of the dynamic feature cube, based on the time series feature modulation vector, the evolution trend of the equipment failure risk heat map is predicted through the spatiotemporal attention mechanism to determine the equipment health warning period; based on the embodiment provided by this application, a spatiotemporal evolution mechanism of the three-dimensional trend map is generated to realize the dynamic mapping of the physical conduction of the power grid and the data characteristics. Specifically, the initial feature cube integrates the geographical grid load parameters, user type pulse sequence, and time node cycle pattern. The load diffusion in the geographical dimension is driven by the equipment distribution density gradient (such as the high load of the industrial park is transmitted to the surrounding residential areas), the user dimension behavior characteristics are constrained by the label tree hierarchy (such as the load fluctuation of the parent plant affects the child plant), and the time dimension equipment health decay modulates the time series characteristics (such as the load fluctuation of the aging line intensifies over time); the accurately extracted three-dimensional risk factors directly guide the scheduling, such as identifying "a certain industrial park with high load + equipment warning period in the next 3 days", automatically triggering the high-frequency collection of the electric meter in the area and linking the operation and maintenance to check the line joints.

[0095] In one specific implementation, the initial cube construction process involves: Geographic Dimension: Mapping the load parameters of grids A-01 to A-10 to an X-axis vector of [45, 62, ..., 78] kW. User Dimension: Arraying the behavioral pulse intensities of residential, commercial, and industrial users along the Y-axis (1.2, 2.3, 4.1). Temporal Dimension: Slicing the Z-axis to store device health trend vectors by month. Spatiotemporal Evolution Update: Load Diffusion: A density gradient of +5 units / km is applied from grid A-05 to A-06, driving a 2% daily load increase. Behavior Propagation: Limiting the "Industrial → Chemical" user node to only affect nodes in the same-label subtree, preventing any impact on residential users. Time Modulation: When the health of a transformer falls below 0.7, the load of the grid it resides in is automatically reduced by 10% (simulating a power curtailment policy). Map Generation: High-Load Region: Extracting grids on the X-axis with loads >80 kW. Anomalous Clusters: Detecting anomalous nodes on the Y-axis with residential user pulse intensities >2.0 (typically ≤1.5). Warning period: Locate the slice with the lowest health on the Z axis (week 4), and determine the maintenance window from 09:00 to 11:00 every day based on the load peak.

[0096] As an optional solution, meter reading tasks can be dynamically adjusted based on energy usage trends and network load, including:

[0097] Determine the adjustment factors of meter reading tasks;

[0098] Based on the 3D trend graph, the meter reading task chain is virtually simulated to generate a digital twin model;

[0099] Adjustment factors are a set of core parameters for optimizing meter reading tasks. These can include geographic clustering cycles (e.g., daily meter reading in industrial areas, weekly meter reading in residential areas), collection dimensions (e.g., harmonic data collection for industrial users), and task priorities (ordered by device health). The three-dimensional trend map is a three-dimensional predictive model composed of high-load areas (spatial), abnormal users (semantic), and warning periods (temporal), used to guide task scheduling.

[0100] Obtain network load status through the operator's network status interface;

[0101] Based on network load conditions, digital twin models, and adjustment factors, the task adjustment strategy is determined to dynamically adjust meter reading tasks.

[0102] Dynamically dispatch human resources, data resources, network resources and hardware resources based on the adjusted meter reading tasks.

[0103] For example, a city's power grid delineation and adjustment factors are as follows: Grouping cycle: Industrial grids read meters three times daily, residential grids read meters once weekly. Collection dimensions: Industrial users are now required to collect harmonic distortion rates, while residential users only require electricity consumption data. Priority rules: Grid tasks with equipment health levels below 60% are prioritized to the highest level. High-load areas (grids A-05 to A-08) predicted in the three-dimensional trend map are input into the Twin Engine to construct a virtual power grid scenario. The field strength for grid A-05 is set to 1.8 (for a predicted load density of 82 kW / km²), and for A-06 to 1.5 (for 65 kW / km²). A base station load heat map obtained through the operator interface shows that the load rate in area A-05 has reached 90% (threshold 75%). Meter reading tasks for A-05 are diverted to adjacent, lower-load grids A-04 and A-09. Harmonic collection is suspended for residential users in A-05, retaining only electricity consumption data.

[0104] Based on the embodiments provided in this application, a task adjustment mechanism based on three-dimensional trend maps and network loads is used to dynamically adapt meter reading tasks to the operating status of the power grid. Specifically, the digital twin model virtual simulation task chain is executed. For example, simulating the pressure of meter uploads during peak production periods in industrial areas, it is found that the base station is congested during a certain time period (such as 10:00-11:00), and the collection period of non-critical meters is adjusted to nighttime in advance; the operator network load is linked in real time. When the base station heat map finds that the load of the base stations around the substation exceeds the base station load threshold (such as 80%, 70%, etc.), the meters within its power supply range are located based on topology coding, non-urgent tasks (such as monthly collection for residents) are delayed, and real-time data upload for industrial users is prioritized; the geographical grouping and user-differentiated collection in the elements are adjusted, and a 30-minute collection cycle is set for key grid node grids (such as substation exits and large industrial access points), a 2-hour cycle is set for residential grids, and special dimension collection such as harmonics is added for industrial users to ensure the integrity of key data.

[0105] like Figure 2 As shown in the figure, as an optional solution, a virtual simulation of the meter reading task chain is performed based on a three-dimensional trend graph to generate a digital twin model, including:

[0106] S201: Input the three-dimensional trend map and the regional features and type features in the dynamic index system into the digital twin engine to build the basic data layer of the digital twin model;

[0107] Optionally, the load parameters of the geographic grid nodes in the three-dimensional trend map, the behavioral pulse sequence and time-series cycle pattern of the user type nodes, the real-time load density generated by the grid topology grid coding in the dynamic index system, the regional characteristics of the equipment distribution density, the load curve shape generated by the label tree hierarchical traversal, and the equipment configuration identification type characteristics are input into the digital twin engine to form a basic data layer;

[0108] S202: Encode the spatial field strength of the geographic grid nodes in the three-dimensional trend map into a dynamic phase vector, convert the behavioral pulse intensity of the user type node into a probability amplitude, and use the time window of the equipment health warning period as a probability modulation operation sequence to form the spatiotemporal constraint parameters of the digital twin model;

[0109] In this embodiment, the dynamic phase vector reflects the timing parameters of the physical conduction delay of the power grid, and the phase change rate is positively correlated with the field strength (field strength + 1 → phase growth rate 20%). Optionally, the spatial field strength of the geographic grid nodes in the three-dimensional trend map is encoded into a dynamic phase vector using a manifold learning network. The behavioral pulse intensity of the user type node is converted into a probability amplitude that follows the hierarchical constraints of the label tree. The time window of the equipment health warning period is used as a probabilistic modulation operation sequence based on fault tree analysis to construct spatiotemporal constraint parameters including spatial field strength gradient, behavioral pulse probability, and timing risk level.

[0110] In this embodiment, the conversion of spatial field strength at geographic grid nodes into phase vectors is based on the fact that the diffusion of electrical energy along the physical conduction paths of the power grid fluctuates, with higher field strengths corresponding to faster phase change rates (simulating power transmission delays). For example, when field strength decreases in a grid due to equipment failure, the frequency of phase vector updates decreases, reflecting the lower priority of meter reading tasks in that area.

[0111] For example, dynamic phase vector generation uses the following formula: phase change rate = field strength × 20° / min. For example, grid A-10 has a field strength of 1.2, resulting in a phase change rate of 24° / min. This simulates the transmission delay of electricity from the substation to the grid (0.5ms per kilometer). If a grid device failure causes a 30% drop in field strength, the phase change rate is reduced to 16.8° / min, and the task priority is automatically downgraded.

[0112] For example, the mapping of phase rate and grid voltage: the dynamic phase vector simulates the voltage phase lag phenomenon in long-distance transmission, the phase change rate = field strength × 20° / min, corresponding to a line delay of 0.5ms per kilometer (calculated according to the wave impedance propagation velocity formula in "Power System Steady-State Analysis").

[0113] Example: Grid A-10 is 10 km from the substation, with a field strength of 1.2, a phase rate of 24° / min, and a total delay of 10 × 0.5 = 5 ms, resulting in a phase deviation of ≤0.1° from the measured voltage. If a lightning strike on a branch causes the field strength to drop by 30%, the phase rate decreases to 16.8° / min, and the delay increases to 7 ms, triggering a task priority downgrade and line inspection.

[0114] The phase vector accurately reflects the physical state of the power grid, and task scheduling and line health are linked in real time.

[0115] S203, based on the topological structure of the physical conduction path of the power grid, a dynamic topological association channel is established between the geographic grid node and the user type node, so that the spatiotemporal attributes of the meter reading task and the physical characteristics of the power grid are bound to the digital twin model architecture;

[0116] Optionally, based on the topological structure of the physical conduction path of the power grid, the load conduction path between nodes is dynamically modeled according to the device distribution density gradient and the transmission impedance parameter, and a dynamic topological association channel is established between the geographic grid node and the user type node. This allows the spatiotemporal attributes of the meter reading task (such as grid location, user type label, task time window) and the physical characteristics of the power grid (such as conduction path impedance, load diffusion direction) to be tensor-bound through a causal neural field pulse network;

[0117] S204: Construct an energy potential well model for the meter reading task path in the manifold space. The order of task nodes is driven by the gravitational gradient between potential wells. The gravitational gradient between potential wells is dynamically calculated based on the device distribution density gradient to generate a task scheduling skeleton for the digital twin model.

[0118] In this embodiment, the energy well model is a virtual gravity model. High-device density areas create strong gravity wells, which prioritize meter reading task paths. In this energy well model, gravity gradient = device density gradient × network load factor. For example, if the density gradient from grid A-10 to A-11 is +5 devices / km and the load factor is 0.8, gravity gradient = 5 × 0.8 = 4. During task path planning, high-gravity gradient areas (such as A-11) are prioritized, and the task node dwell time is extended to 1.5 times the standard value.

[0119] Optionally, in the manifold space, low-dimensional manifold basis vectors are extracted by performing orthogonal decomposition on the dynamic feature cube (the basis vectors include geographic field intensity gradients, label tree hierarchy weights, and time series fluctuation patterns), and an energy potential well model for the meter reading task path is constructed. The order of task nodes is dynamically calculated based on the gravitational gradient between potential wells driven by the device distribution density gradient (the gravitational gradient quantifies the node load correlation using the cosine similarity algorithm), generating a task scheduling skeleton that includes task priorities and path dependencies.

[0120] In this embodiment, the gravity in the energy well model is not a physical force, but a virtual force constructed based on the device density gradient. High-density areas form a "strong gravitational potential well," forcing meter reading task paths to prioritize areas with dense device density (reducing the risk of missed detections). Density differences between adjacent grids are converted into a gravitational gradient, driving task nodes to automatically migrate to uncovered, high-density areas. The gravitational gradient formula is designed to be the product of the device density gradient and the network load heat map, ensuring that high-load areas are automatically avoided when resources are limited.

[0121] S205, deploying a virtualized meter cluster in the digital twin model. The simulation behavior of each meter node is regulated by the conduction path of the causal neural field pulse network. The interference of network load on the integrity of meter reading data is simulated through pulse propagation delay, forming an interactive digital twin simulation environment.

[0122] Optionally, a virtualized meter cluster based on meter identification is deployed in the digital twin model. The simulation behavior of each meter node is controlled by the physical conduction path of the power grid in the causal neural field pulse network. The data transmission delay / packet loss caused by network load is simulated through the pulse propagation delay algorithm, forming an interactive digital twin simulation environment and providing real-time feedback on the integrity of meter reading data under different network conditions.

[0123] S206, reversely inject the twin data stream generated by simulation into the geographic node field strength calculation module of the three-dimensional trend map, calibrate the phase parameters of the dynamic topology associated channel, and output a digital twin model including the grid residence time scheduling matrix, data storage partition strategy and pulse timing alignment parameters.

[0124] Optionally, the twin data stream generated by the simulation (including parameters such as grid residence time and data transmission delay) is reversely injected into the geographic node field strength calculation module of the three-dimensional trend map. By comparing the deviation between the virtual meter pulse signal and the actual grid conduction path, the phase parameters of the dynamic topology associated channel (such as load conduction delay coefficient and node coupling strength) are dynamically adjusted. The final output is a digital twin model including the grid residence time scheduling matrix (based on gravitational gradient optimization), data storage partitioning strategy (compressing non-critical data according to probability amplitude) and pulse timing alignment parameters (correcting transmission delay).

[0125] In this embodiment, the grid dwell time scheduling matrix plans the length of time meter readers spend in different geographic grids based on the device density gradient and task priority. For example, in industrial grids with high device density (e.g., grid topology code "GZ-110kV-3B"), the dwell time is set to 30 minutes per visit to verify high-frequency load data; in residential grids, the dwell time is shortened to 10 minutes per visit.

[0126] Data storage partitioning strategies include dividing storage areas by user type and data importance, and dynamically allocating storage resources. For example, critical information such as harmonic data and equipment health for industrial users can be stored in high-speed solid-state partitions, while basic electricity usage data for residential users can be compressed and stored in low-cost disk partitions.

[0127] Pulse timing alignment parameters are used to coordinate meter reading data transmission timing, avoiding network peaks and matching the grid load cycle. For example, aligning an industrial user's high-frequency data collection tasks (e.g., once per minute) to the nighttime off-peak period (02:00-05:00) reduces the impact of channel congestion on real-time data.

[0128] In some embodiments, when the twin data stream generated by the digital twin model is reversely injected into the trend map, the following parameters are calibrated first: High load density area boundary: shrink or expand the grid field strength range according to the actual coverage of the virtual task path. Abnormal user group judgment threshold: dynamically adjust the anomaly detection sensitivity based on the data error caused by the pulse propagation delay in the simulation environment.

[0129] In one specific implementation, the input data for grid A-10 in a three-dimensional trend map is a field strength of 1.2, an industrial user pulse strength of 3.5, and a warning period of [09:00-11:00]. The dynamic indexing system provides A-10's real-time load density of 58 kW / km² and the equipment label "Industry → Metallurgy → Arc Furnace." Dynamic phase vectors: The field strength of 1.2 is converted to a phase change rate of 24° / min (formula: rate = field strength × 20). Probability amplitude: The pulse strength of 3.5 is mapped to an amplitude of 0.85 (normalized to the range of 0-1). Time windowing: The warning period is converted into a probabilistic modulation sequence, reducing the task triggering probability by 30% between 09:00-11:00. Based on the grid transmission path, a bidirectional channel is established between grid A-10 and the "Metallurgy" user node, constraining the task path to follow the transmission route (e.g., A-10 → A-11 → A-12).

[0130] Potential well model construction: The device density gradient between A-10 and A-11 (+5 devices / km) was calculated, generating a gravitational gradient value of 0.15. In manifold space, task nodes automatically migrated from A-10 to A-11, with a minimum dwell time set to 15 minutes. Simulation environment verification: A virtual meter cluster simulated network latency at A-10. When the load exceeded 80%, the pulse propagation delay increased by 200ms, triggering the data retransmission mechanism. The field strength calculation module was reverse-calibrated, correcting the field strength at A-10 to 1.15.

[0131] Exemplarily, calibration trigger conditions include automatically triggering the calibration mechanism when the propagation delay of the virtual pulse signal deviates by more than 10% from the actual grid conduction path. For example, grid A-10 simulates a delay of 5ms, while the measured delay is 5.6ms (a 12% deviation), triggering calibration.

[0132] Parameter adjustment rules: Field strength correction: Adjust the field strength weight proportionally to the deviation. For example, if the deviation is 12%, the field strength is corrected from 1.2 to 1.2 × (1 - 0.12) = 1.056. Node coupling strength: Increase the coupling coefficient of adjacent grid nodes by 5% (for example, from 0.8 to 0.84) to compensate for conduction path errors. Phase rate synchronization: After calibration, the phase rate = corrected field strength × 20° / min to ensure matching with the line impedance.

[0133] Based on the embodiments provided in this application, the digital twin model construction process realizes the deep coupling of meter reading tasks and the physical characteristics of the power grid. Specifically, the basic data layer integrates the graph and index features, such as inputting the 220kV substation topology grid and the regional industrial user label tree into the engine to construct a virtual power grid topology, and the model nodes correspond to the actual line physical nodes. The field strength of the geographical node is converted into a dynamic phase vector (reflecting the voltage phase lag at the end of the long line), the user pulse is converted into a probability amplitude (the probability of high-load pulse when industrial equipment starts), and the equipment warning period is used as a time window constraint (forced high-frequency collection during the period when the transformer oil temperature exceeds the standard); the manifold space energy potential well model calculates the task node gravity based on the equipment distribution density (the grid with strong gravity at the substation outlet is dispatched first), and generates the task sequence along the "substation → main line → branch line" topology path to reduce invalid jumps across the line.

[0134] In a specific embodiment, Figure 3 The figure shows a flow chart of dynamic task adjustment and digital twin modeling, which specifically includes:

[0135] Input 3D trend maps and dynamic index features: 3D trend maps: generated by a causal neural field spike network, including high-load density areas (such as grid G05), abnormal user groups (such as a chemical plant), and equipment health warning periods (such as the warning period of transformer T-12); dynamic index features: including geographic grid codes (such as G05 is coded as "Grid-05") and user label tree hierarchies (such as "industrial user → chemical plant → production line");

[0136] Encoding dynamic phase vector: Encode the spatial field strength of the geographic grid node (such as G05 field strength = 0.8) into a dynamic phase vector (such as "Phase_G05 = 0.8") to quantify task priority;

[0137] Conversion probability amplitude: The behavior pulse intensity of the user type node (such as the pulse intensity of a chemical plant = 120) is converted into a probability amplitude (such as "Amp_Plant = 0.85") to represent the probability of task execution;

[0138] Construct dynamic topology association channels: Based on the physical transmission paths of the power grid (such as the impedance parameters of the trunk line), establish association channels between the geographic grid and user type nodes (such as G05 is bound to the chemical plant node);

[0139] Manifold space basis extraction: Basis vectors are extracted through a manifold learning network. Each basis vector carries the field intensity gradient (e.g., 0.8 for G05), the label tree level weight (e.g., chemical plant weight = 0.7), and the time window fluctuation parameter (e.g., warning period fluctuation = ±5%).

[0140] Generate an energy potential well model: In the manifold space, an energy potential well is generated based on the device distribution density gradient (e.g., G05 density gradient = 1.2). Task nodes are arranged according to the gravitational gradient between potential wells (tasks in high-density areas are executed first).

[0141] Deploy virtual meter clusters: Deploy virtual meters (e.g., a cluster of simulated chemical plant meters) within the digital twin model. Their behavior is controlled by a causal neural field spike network that mimics the conduction paths of a real power grid.

[0142] Simulating pulse propagation delay: Based on the network load (e.g., base station load = 80%), the pulse propagation delay time (e.g., delay = 200ms) is calculated to simulate a scenario where data acquisition integrity is disturbed.

[0143] Reverse calibration of field strength parameters: Compare simulation data (such as virtual meter pulse signals) with the actual grid conduction path. If the deviation exceeds a threshold (such as >10%), dynamically adjust the field strength parameters (such as correcting the G05 field strength from 0.8 to 0.75).

[0144] Output scheduling matrix and strategy: Generate grid residence time scheduling matrix (e.g., G05 residence time = 2 hours), data storage partition strategy (e.g., non-critical user data is compressed and stored in partition B), and pulse timing alignment parameters (e.g., high-frequency tasks are aligned to network off-peak hours).

[0145] As an optional solution, adjustment factors include meter reading cycles for geographical groups, differentiated collection dimensions for user types, task priority sorting in the time dimension, matching rules between human skills and task areas, sampling frequency and self-test cycle configuration of meter hardware, and priority mapping between data collection strategies and network channels. Network load conditions include regional base station load heat map data and communication channel congestion warning information.

[0146] As an optional solution, a task adjustment strategy is determined based on network load conditions, digital twin models, and adjustment factors to dynamically adjust meter reading tasks, including:

[0147] The signal attenuation gradient of the regional base station load heat map is mapped to the disturbance factor of the dynamic phase vector. The disturbance factor is calculated based on the product of the signal attenuation gradient and the grid topology coefficient.

[0148] The communication channel congestion warning information is mapped into the attenuation coefficient of the probability amplitude, and the attenuation coefficient is negatively correlated with the channel packet loss rate;

[0149] In this embodiment, the disturbance factor is the influence coefficient of network signal attenuation on the task phase, and is calculated as: attenuation gradient (dB / km) × topology coefficient (0.1-0.5).

[0150] In the manifold space, the geographical boundaries of the clustered meter reading periods are encoded as potential well location constraints, and the label tree hierarchy of the differentiated collection dimensions is converted into orthogonality constraints of the basis vectors.

[0151] In this embodiment, in the manifold space modeling, the basis vector is a set of basic vectors in the mathematical space that maps the differentiated collection dimensions of the label tree level to quantify the differences in collection requirements of different user types. Each basis vector corresponds to a level or category of the label tree (such as "industrial users", "residential users", "high-energy-consuming equipment", etc.), and its direction and amplitude reflect the collection priority, data type or transmission characteristics of the dimension. For example, when the label tree level is "industry → manufacturing → high-energy-consuming equipment", the corresponding basis vector can be expressed as Vindustry = [1,0,0] (representing the industrial user dimension) and Vhigh-energy = [0,1,0] (representing the high-energy-consuming equipment dimension). The basis vector of residential users is Vresidential = [0,0,1], which is orthogonal to the industrial dimension (no intersection), ensuring that the collection tasks of the two types of users are processed independently.

[0152] The orthogonality constraint means that different basis vectors are perpendicular to each other in the mathematical space (the inner product is zero), which means that the differentiated acquisition dimensions are independent of each other and do not interfere with each other. In this application, it is used to ensure that the data processing and task scheduling paths of different user types (such as industry / residential) or different acquisition dimensions under the same type (such as critical equipment / non-critical equipment) do not cause cross-interference. For example, industrial users need to collect "harmonic energy distribution" (corresponding to the label tree level "Industry → Equipment Configuration → Harmonic Monitoring"), and residential users only need to collect "basic electricity" (corresponding to "Residential → Conventional Collection").

[0153] Orthogonality is achieved by encoding the harmonic dimension of industrial users as a basis vector Vharmonic = [1, 0], and the energy dimension of residential users as Venergy = [0, 1]. Since these two dimensions are orthogonal (the inner product is 0), during task scheduling, industrial users' harmonic collection tasks do not trigger adjustments to residential users' energy collection parameters. When network load fluctuates, the transmission priority of Vharmonic can be adjusted independently (for example, by expanding bandwidth) without affecting the Venergy collection period.

[0154] In manifold space, the energy potential well model for task scheduling must satisfy the orthogonality of the basis vectors, meaning that task paths for different user types do not overlap in the vector space. For example, industrial user task nodes can only move within the subspace spanned by Vindustry, while residential user tasks are restricted to the Vresidential subspace to avoid cross-dimensional interference.

[0155] Potential well location constraints encode the boundaries of meter reading cycles for clustered geographic areas (e.g., grid topology divisions for industrial and residential areas) as spatial constraints on the energy wells in manifold space. This ensures that meter reading task chains are executed only within specific geographic boundaries, preventing inefficient cross-regional scheduling. For example, a city's power grid is topologically divided into industrial grids G01-G10 and residential grids R01-R20. The industrial cluster has a meter reading cycle of three times daily, and its geographic boundaries (G01-G10) are encoded as potential well location constraints. Task nodes can only move within the G01-G10 grids and are prohibited from scheduling to the residential grid R01. The residential cluster has a cycle of once a week, and its potential well location constraints are within the R01-R20 grids, preventing high-load industrial tasks from encroaching on low-load areas.

[0156] When the signal attenuation gradient exceeds the load threshold of the equipment health warning period, the dynamic phase synchronization of adjacent grids is triggered to migrate high-load tasks to the low-load period; the collection dimensions of non-critical equipment of industrial users (marked by equipment configuration identifiers) are stripped out according to the attenuation coefficient.

[0157] In this example, the signal attenuation gradient is the rate at which signal strength decreases with geographic distance (dB / km) in the regional base station load heat map, reflecting the degree of network congestion. For example, in an industrial park (grid code GZ-110kV-3B), due to the dense equipment density, the signal attenuation gradient reaches -5dB / km, requiring meter reading tasks to be adjusted to avoid areas with high attenuation.

[0158] High-load tasks are meter reading tasks that require frequent collection or transmission of large amounts of data (such as monitoring sudden load changes at the minute level for industrial users). For example, collecting transient characteristics once per minute for equipment in the "Industry → Chemical → Reactor" tag tree hierarchy is a high-load task.

[0159] Off-peak hours are periods of low network load (e.g., 2:00 AM to 5:00 AM at night) and are suitable for non-urgent meter reading tasks. For example, monthly data compression tasks for residential areas can be scheduled during the off-peak hours in the early morning to avoid impacting peak data collection for industrial users.

[0160] The attenuation coefficient is a parameter generated based on the packet loss rate when the communication channel is congested (negatively correlated with the packet loss rate). For example, if the packet loss rate on a channel reaches 15%, the attenuation coefficient is set to 0.85, triggering the industrial user to reduce the frequency of non-critical data collection from hourly to daily.

[0161] Non-critical equipment for industrial users: refers to equipment in non-core production links that has little impact on grid load fluctuations in industrial production. Its data collection dimensions can be dynamically reduced during network load peaks to prioritize data transmission for critical equipment.

[0162] For example, in the equipment tree at the "Industry → Manufacturing → Food Processing → Packaging Line" level, a backup packaging machine is considered non-critical equipment. When the communication channel packet loss rate exceeds 10% (attenuation coefficient less than 0.9), non-critical collection dimensions such as "Packaging Speed" and "Energy Consumption Breakdown" can be stripped out, retaining only basic power consumption data. This ensures complete upload of real-time load data for the main production line (such as sterilization equipment).

[0163] The collection dimensions of non-critical equipment of industrial users include data dimensions of equipment among industrial users that have a lower impact on grid security (such as real-time temperature monitoring of backup production lines).

[0164] Among them, the load threshold adopts a two-layer determination mechanism of "basic value + dynamic correction": first, the basic threshold is calculated based on the health status of the equipment, historical load safety peak and the complexity of the power grid topology, and then it is adjusted in real time through dynamic correction factors such as climate anomalies, time period sensitivity and equipment health warning level, so that the load threshold can not only reflect the inherent safety boundary of the equipment, but also adapt to the real-time environmental changes of the power grid operation.

[0165] For example, to determine the load threshold for a particular industrial park during high temperatures, the following calculations are used: Basic threshold calculation: Equipment health baseline value: Based on meter self-test data, battery life is 60% (medium health), set at 1.2 dB / km; Historical load safety peak value: The 80th percentile of the industrial load in the area over the past 30 days is taken as 2.0 dB / km; Topology complexity adjustment: Device density is 250 devices / km² (high density), with an adjustment factor of 0.8; Basic threshold = 1.2 × 2.0 × 0.8 = 1.92 dB / km. Dynamic correction factor adjustment: Climate anomaly correction: A high temperature warning of 38°C detected by the meteorological interface triggers a -15% correction (high temperatures can easily cause equipment failure); Equipment health warning correction: The three-dimensional trend map marks the area as "Equipment Health Warning Level II", triggering a -10% correction; The dynamically corrected threshold value = 1.92 × (1-15% - 10%) = 1.44 dB / km.

[0166] When the signal attenuation gradient in the area exceeds 1.44dB / km, high-load tasks are triggered to migrate to the nighttime off-peak period, and the collection dimensions of non-critical industrial equipment are stripped away to avoid data packet loss caused by network overload in high-temperature environments.

[0167] In some embodiments, meter reading tasks are dynamically scheduled to avoid network congestion based on the regional base station load heat map and channel congestion warning. For example, when an industrial park has a network traffic peak from 12:00 to 14:00 in the afternoon due to the peak production season, the meter reading task of the industrial meters in the area is postponed to 16:00-18:00, and the transmission volume of non-real-time data is compressed. At the same time, priority is given to ensuring that key equipment data is transmitted through high-bandwidth channels.

[0168] In one specific implementation, the perturbation factor is calculated as follows: the signal attenuation gradient of base station A-05 is -3 dB / km, and the topology coefficient is 0.3, resulting in a perturbation factor of -3 × 0.3 = -0.9. Attenuation coefficient generation is performed as follows: a 12% channel packet loss rate is used to generate a probability amplitude attenuation coefficient of 1 - 0.12 = 0.88. Constraint encoding: Potential well location: The industrial cluster boundaries (A-05 to A-08) are encoded as potential well coordinates (5, 8), and residential areas are encoded as (9, 12). Orthogonal basis: The industrial collection dimensions (harmonics and power) are mapped to the vector [1, 0], and the residential dimensions are mapped to [0, 1], ensuring data isolation. Dynamic adjustment: Phase synchronization triggering: When the perturbation factor of A-05 exceeds the threshold of -0.8, its task is migrated to the adjacent grid A-04 (perturbation factor -0.5). Dimension stripping: For industrial users with an attenuation coefficient < 0.9, only power collection is retained, removing harmonics and temperature data.

[0169] Based on the embodiments provided in this application, the quantitative mapping mechanism of the task adjustment strategy realizes the coordination of network status, business needs, and power grid topology. The base station signal attenuation gradient (such as the industrial park base station attenuates by 15dB due to obstruction) is mapped to a phase disturbance factor, which reduces the priority of the meter task in the area and triggers the temporary takeover of the meter task on the same topology line covered by the adjacent base station; the channel congestion warning is converted into the stripping of the collection dimension: when the packet loss rate of the channel in the industrial area exceeds the packet loss rate threshold (such as 20%), the non-critical equipment collection dimension is stripped according to the attenuation coefficient (the backup production line data is not collected temporarily), and only the key parameters of the main equipment and transformer are retained to ensure the safe data transmission of the power grid. The geographical boundaries of the group meter reading cycle are set according to the topological grid (such as continuous collection of the same 10kV line grid), avoiding the traditional division by cell, which causes the same line meters to belong to different task groups, thereby improving the efficiency of task execution.

[0170] As an optional solution, dynamic scheduling of human resources, data resources, network resources, and hardware resources includes:

[0171] The grid residence time scheduling matrix is ​​input into the skill matching engine. Based on the grid topology location code corresponding to the QR code metal identification piece, the matching degree between the meter reader skill vector and the geographical field intensity gradient is calculated to generate a dynamic scheduling sequence.

[0172] Optionally, the grid residence time scheduling matrix output by the digital twin model is input into a cross-domain mapping engine with grid field strength constraints. Based on the phase synchronization state of the dynamic topology-associated channel, the skill vector of the human resources and the field strength gradient of the geographic grid are tensor-fused, and the matching degree is calculated by the cosine similarity algorithm to generate a heat map of skill and field strength matching, thereby realizing dynamic binding of high-load density areas with personnel with fault diagnosis skills.

[0173] In some embodiments, high-load areas are predicted through three-dimensional trend maps, allowing for pre-scheduled data collection frequency and verification strategies. For example, if a commercial district is predicted to experience peak electricity consumption next week due to a promotional event, the following actions can be performed: A load transmission path for the grid in that area is generated using a causal neural field spike network model, increasing the meter collection frequency from once per hour to once every 30 minutes; Simulating task scheduling within the digital twin model to prioritize manual inspection tasks and dynamically scheduling meter readers based on their skill set (e.g., familiarity with the area's grid topology); and When network load exceeds a threshold, task migration is triggered to adjacent grids, delaying non-urgent meter reading tasks to low-demand periods (e.g., 2:00 AM to 5:00 AM).

[0174] In some embodiments of the present application, a meter reader skill vector is generated by constructing a multidimensional vector based on historical task data, including: Geographic field intensity gradient adaptability: This is used to calculate the operator's operational stability score based on the field intensity fluctuation range in the operator's historical task area. Equipment type familiarity: This is used to calculate the operator's success rate in resolving specific equipment failures based on the equipment maintenance records recorded on the QR code metal identification card.

[0175] The generation of dynamic scheduling sequences includes: performing a tensor product operation on the skill vector and the grid residence time scheduling matrix, outputting a matching matrix, and allowing task allocation only when the matching degree exceeds a certain threshold (such as >85%) to avoid human experience intervention.

[0176] For example, a meter reader's skill vector includes: regional familiarity (historical task success rate), equipment type experience (number of times a specific device has been handled), and emergency response score (mean time to recovery). For example, Person A's skill vector is [0.9, 0.8, 0.7], indicating 90% familiarity, 80% experience, and 70% response score.

[0177] Dynamic scheduling rules: Matching degree = cosine similarity between the skill vector and the grid field intensity gradient. For grid A-05 with a field intensity gradient of 0.15, the matching degree for person A is 0.9 × 0.15 + 0.8 × 0.1 + 0.7 × 0.05 = 0.29 (threshold 0.25). When the grid device health falls below 60%, the matching degree threshold is automatically raised to 0.3, allowing only highly skilled personnel to perform the task.

[0178] According to the pulse timing alignment parameters, the probability amplitude branch of non-urgent tasks is activated during the low-load period, and the non-critical user time window data in the type feature is compressed and stored;

[0179] Optionally, according to the pulse timing alignment parameters and the data storage partition strategy, the probability amplitude branch of non-urgent tasks is activated during the low network load period, and the corresponding data storage partition is marked as delayed write mode; the storage partition data is dynamically compressed and decompressed based on the probability amplitude distribution, and the time window detail data of non-critical users is stripped out when the channel is congested, retaining the main component characteristics of the load curve morphology.

[0180] When high-frequency demand from industrial users is detected (triggered by a sudden change in the intensity of behavioral pulses), a bandwidth expansion request pulse is generated based on the disturbance factor. The bandwidth increment is calculated based on the product of the pulse intensity and the operator's configuration coefficient.

[0181] In this embodiment, the bandwidth expansion request pulse is a network resource scheduling signal generated according to the intensity of industrial user demand. The pulse width represents the demand duration, and the amplitude represents the bandwidth increment.

[0182] In some embodiments of the present application, when a three-dimensional trend map marks an area as having a high load density or an abnormal user group, the sampling frequency and self-test period of the corresponding meter are dynamically adjusted based on the identification of the meter, where the sampling frequency of industrial user meters is increased to 3 times a day, and the sampling frequency of residential users is maintained at 1 time a day.

[0183] Optionally, during network load trough periods, temporarily expanded bandwidth resources are automatically released through nonlinear mapping between pulse signal strength and bandwidth requirements.

[0184] In this embodiment, bandwidth expansion request pulse generation involves generating a strong pulse signal from the causal neural field pulse network when the harmonic energy distribution characteristics triggered by high-frequency demand from industrial users exceed a certain threshold. The pulse intensity is normalized using operator-configured coefficients (e.g., maximum base station bandwidth divided by current load) to generate a bandwidth increment. This increment is distributed in real time to the corresponding regional base stations via the SDN controller, dynamically adjusting the communication priority of the meter cluster.

[0185] When the network interrupts the propagation of abnormal pulses, a dynamic topology association token is generated based on the meter identification. The token priority is dynamically adjusted according to the risk level of the equipment health warning period to ensure that key equipment data is reissued according to the calibrated field strength gradient.

[0186] In one specific implementation, resources such as hardware collection frequency and human verification priority are scheduled differently based on user type and usage trends. For example, in a scenario where electricity consumption in a residential area is predicted to surge in winter, hardware resources are used to temporarily increase the sampling frequency of residential meters from once a day to twice a day. Human resources are used to automatically flag meters with a sudden increase in electricity consumption exceeding 30% and assign them to on-site verification tasks. Network resources are used to transmit non-urgent data using the LoRa protocol during off-peak hours at night. Data resources are used to store real-time electricity usage data in a local cache and archive historical data to cloud-based cold storage.

[0187] Optionally, when a network outage causes meter reading data loss, the dynamic topology-associated token generated based on the meter ID contains the following metadata: a priority identifier, dynamically set by the risk level of the device during the health warning period (tokens for high-risk devices are prioritized for retransmission); a topology path identifier, which selects the optimal retransmission path based on the grid topology location code (e.g., avoiding faulty lines or relaying through adjacent grids). During dynamic scheduling, a skills matching engine matches the token's topology path identifier with the meter reader's skill vector (e.g., familiarity with the equipment types in a particular area), ensuring that high-priority tasks are performed by individuals with the appropriate skills.

[0188] When the network is interrupted, a dynamic topology association token is generated through the identification of the meter. The token priority is dynamically adjusted by the risk level of the equipment health warning period through the fault tree analysis model; during the switching process of the backup communication channel, based on the probability amplitude distribution priority of the token, the data retransmission sequence during the interruption period is dynamically reconstructed according to the grid topology field strength gradient order to generate a cross-operator retransmission token chain.

[0189] In one specific implementation, human resource scheduling involves: Skill matching: Personnel A's skill vector is [0.9, 0.8, 0.7] (familiar with area A-05, skilled in handling industrial equipment, and quick to respond), which matches the A-05 field intensity gradient of 0.15 by 92%, and is assigned to this area. Shift scheduling: Based on the dwell time matrix, daily meter reading hours for A-05 are set to 08:00, 14:00, and 20:00 (avoiding the early warning period of 09:00-11:00). Data resource optimization: Pulse timing alignment: Non-urgent tasks (such as backing up historical residential data) are activated during the low-load period of 02:00-05:00 in the morning, using compressed storage (60% compression ratio). Network resource expansion: High-frequency demand (pulse intensity 4.2) is detected at a steel plant, and the operator configuration coefficient is 0.2, resulting in a bandwidth increase of 4.2 × 0.2 = 0.84 Gbps. The bandwidth in this area is temporarily increased. Fault recovery processing: When A-07 grid communication is interrupted, a dynamic token 0x01|A-07→A-06|E-12-5B is generated, and data is preferentially transmitted through the A-06 relay, reducing the latency from 8 hours to 45 minutes.

[0190] Based on the embodiments provided in this application, a multi-resource dynamic scheduling mechanism realizes topology-aware resource collaboration. Specifically, based on the grid residence time and the skill matching of meter readers (meter readers familiar with a certain substation area are assigned to that grid first), a scheduling sequence along the line path is generated to reduce cross-line movement; during low-load periods of the power grid, the storage of non-critical user data is compressed (for example, the sampling frequency of residential meters is reduced from 15 minutes / time to 1 hour / time), and temporary bandwidth expansion is applied before industrial peaks to ensure the upload of critical load data; hardware resource scheduling is based on power grid risks: when the network is interrupted, a topology association token is generated, and data is retransmitted according to the priority of "substation outlet meter → large industrial user → residential meter", shortening the recovery time of critical data and providing timely support for power grid scheduling.

[0191] It's important to note that data retransmission tokens are used to resend data when network anomalies occur. They contain device identification, priority, and topology paths. Dynamic topology association tokens are generated when the network is down or at high risk, and their priority is dynamically adjusted based on device health risk. Both tokens are based on meter identification and correspond to passive recovery and active disaster recovery scenarios, respectively, forming a two-tiered fault tolerance mechanism.

[0192] According to another aspect of the embodiment of the present application, a meter reading data management and resource scheduling optimization system based on AI is provided, such as Figure 4 As shown in , the system includes:

[0193] The meter reading data acquisition module 401 is used to obtain meter reading data of a meter. The meter has a corresponding metal identification piece with a QR code. The QR code on the metal identification piece stores the identification of the meter and the grid topology location code.

[0194] The meter reading data management module 402 is used to manage meter reading data by geographical area and user type to form a multi-level data feature system. User types include residential type and industrial type.

[0195] The usage trend determination module 403 is used to use an AI model to perform trend modeling on the multi-level data feature system to determine the energy usage trend in the future period;

[0196] Resource scheduling optimization module 404 is used to dynamically adjust meter reading tasks based on energy usage trends and network load conditions to dynamically schedule human resources, data resources, network resources, and hardware resources. Among them, the meter identifier is used to generate a data retransmission token, and the grid topology location code is used to bind the relationship between the meter reading task and the physical transmission path of the grid.

[0197] By predicting future energy usage trends, managers can develop reasonable resource allocation plans, optimize scheduling strategies, and achieve more efficient management.

[0198] In some embodiments of the present application, the AI-based meter reading data management and resource scheduling optimization system also includes: a meter reading data approval module, an IoT data dispatch approval module, a new meter reading trial application module, an IoT achievement data display module, a manual performance appraisal module, an account authority setting module, and a system setting module.

[0199] In a specific embodiment, the meter reading data approval module includes the meter reading name, submitter, submission date, and application field. If the user clicks an "Approve" icon, the decision is approved; if the user clicks a "Disapprove" icon, the decision is rejected and the user's reason for rejection is received.

[0200] The meter reading data management module includes ranking, meter reading data name, number, date, and related operations and applications. In response to the user clicking the add icon, it receives the meter reading data name, price, and date entered by the user; in response to the user deleting a meter reading data entry, it deletes the meter reading data entry.

[0201] The IoT data dispatch approval module includes the name, applicant, start date, end date, and related operations and applications. In response to the user clicking the approve icon, the user is prompted to approve the application; in response to the user clicking the disapprove icon, the application information is rejected and the user's incorrect information is received.

[0202] The new meter reading trial application module includes receiving the meter reading name, meter reading number, meter reading data chart and applicant information filled in by the user, and jumping to the waiting for approval section.

[0203] The IoT Achievement Data Display module includes a chart showing the occupational distribution of system personnel. The Manual Performance Appraisal module includes a bar chart showing manual performance appraisal data, including daily performance figures for the week, performance trends over the past seven days, project completion numbers over the past twelve months, and project completion trends over the past twelve months. The module also includes relevant data such as performance, additional commission performance, name, date, and more. The System Settings module contains three subsections: Personal Data Modification, Password Modification, and Secure Logout.

[0204] According to another aspect of an embodiment of the present application, an electronic device for implementing the aforementioned AI-based meter reading data management and resource scheduling optimization method is also provided. The electronic device can be a terminal device or a server. This embodiment is described using the electronic device as a server. The electronic device includes a memory and a processor. The memory stores a computer program. The processor is configured to execute the steps of any of the aforementioned method embodiments using the computer program.

[0205] Optionally, in this embodiment, the electronic device may be located in at least one network device among a plurality of network devices of a computer network.

[0206] Alternatively, those skilled in the art will appreciate that the electronic device may also be a smart phone, tablet computer, PDA, mobile Internet device, PAD or other terminal device.

[0207] In other embodiments, the terminal device or server may be a node in a distributed system, wherein the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting multiple nodes through network communication. The nodes may form a peer-to-peer (P2P) network, and any computing device, such as a server, terminal, or other electronic device, may become a node in the blockchain system by joining the peer-to-peer network.

[0208] Optionally, in this embodiment, the embodiments to be implemented by the above-mentioned various unit modules can refer to the above-mentioned various method embodiments, which will not be repeated here.

[0209] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0210] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An AI-based meter reading data management and resource scheduling optimization method, characterized in that: include: Obtaining meter reading data of a meter, wherein the meter has a corresponding metal identification piece with a QR code, and the QR code on the metal identification piece stores an identification of the meter and a grid topology location code; Construct a dynamic index system based on geographic region, user type, and time dimensions. The geographic region dimension is divided into grid codes according to the power grid topology, and the user type dimension is divided according to the label tree hierarchy. The meter reading data is hierarchically reorganized based on the dynamic index system; in the geographical area dimension, the meter reading data is spatially clustered according to grid codes to generate regional features including real-time load density and device distribution density; in the user type dimension, user features are matched through label tree traversal to generate type features including load curve shape and device configuration identification; The reorganized meter reading data is multi-scale associated with the time dimension to generate a multi-level data feature system including a basic data layer, a dimensional feature layer, and a cross-dimensional fusion layer. User types include residential and industrial types. Mapping the features of the basic data layer, the features of the dimensional feature layer, and the features of the cross-dimensional fusion layer to nodes of a causal neural field spike network model; The spatial field strength of geographic grid nodes is constructed based on grid coding and carries load parameters. The behavioral pulse intensity of user type nodes is generated based on the load curve shape and device configuration identification of the label tree. The periodic pattern of time nodes is modulated by the decay rate of healthy devices and load fluctuation characteristics. The causal neural field pulse network model is trained based on historical power grid topology and load data, with the optimization goal of minimizing the simulation deviation of the power grid conduction path. The training process of the causal neural field pulse network model includes: when backpropagating to adjust the field strength weight, if the actual load diffusion speed of a line is 20% faster than the model prediction value, then the field strength coupling coefficient between the line node and the downstream node is increased by 10%; A physical power grid conduction path is constructed in the causal neural field pulse network model; wherein the load conduction path between geographic grid nodes is dynamically modeled based on the device distribution density gradient and transmission impedance parameters to simulate the diffusion process of electric energy along the power grid topology; the load impact between user type nodes is constrained by the hierarchical relationship of the label tree; Mapping the geographic grid node load parameters, user type node behavior pulse sequence, and time node cycle pattern output by the causal neural field pulse network model into feature tensors of geographic area dimension, user type dimension, and time dimension, respectively, to form an initial feature cube; Based on the physical conduction path of the power grid, a spatiotemporal evolution operator is designed to update the initial feature cube to generate a dynamic feature cube, wherein the load diffusion feature matrix in the geographical area dimension is driven by the device distribution density gradient; the behavior feature propagation matrix in the user type dimension is constrained by the label tree hierarchy; and the time series feature modulation vector in the time dimension is integrated with the healthy device attenuation rate parameter. Extracting high-load density areas, abnormal user groups, and equipment health warning periods from the dynamic feature cube to generate a three-dimensional trend map to determine energy usage trends in future cycles; Determine the adjustment factors of meter reading tasks; Inputting the three-dimensional trend map and the regional features and type features in the dynamic index system into the digital twin engine to construct a basic data layer of the digital twin model; The spatial field intensity of the geographic grid nodes in the three-dimensional trend map is encoded into a dynamic phase vector through a manifold learning network, the behavioral pulse intensity of the user type node is converted into a probability amplitude, and the time window of the equipment health warning period is used as a probability modulation operation sequence to form a spatiotemporal constraint parameter including the spatial field intensity gradient, the behavioral pulse probability and the time series risk level; Based on the topological structure of the physical conduction paths of the power grid, load conduction paths between nodes are dynamically modeled according to the device distribution density gradient and transmission impedance parameters. Dynamic topological association channels are established between geographic grid nodes and user type nodes. The spatiotemporal attributes of the meter reading task and the physical characteristics of the power grid are bound to the digital twin model architecture through a causal neural field pulse network. The spatiotemporal attributes include grid location, user type label, and task time window; the physical characteristics of the power grid include conduction path impedance and load diffusion direction. In the manifold space, the low-dimensional manifold basis vectors are extracted by performing orthogonal decomposition on the dynamic feature cube to construct an energy potential well model for the meter reading task path. The order of task nodes is driven by the gravitational gradient between potential wells, which is dynamically calculated based on the device distribution density gradient. This generates a task scheduling skeleton for the digital twin model that includes task priorities and path dependencies. A virtualized meter cluster is deployed in the digital twin model. The simulation behavior of each meter node is regulated by the conduction path of the causal neural field pulse network. The interference of network load on the integrity of meter reading data is simulated through pulse propagation delay, forming an interactive digital twin simulation environment. The simulated twin data stream is reverse-injected into the geographic node field strength calculation module of the three-dimensional trend map. By comparing the deviation between the virtual meter pulse signal and the actual grid conduction path, the phase parameters of the dynamic topology-related channels are calibrated. The output is a digital twin model including the grid residence time scheduling matrix, data storage partitioning strategy, and pulse timing alignment parameters. The grid residence time scheduling matrix is ​​used to indicate the length of time meter readers stay in different geographic grids based on the device distribution density gradient and task priority. Obtain network load status through the operator's network status interface; Mapping the signal attenuation gradient of the regional base station load heat map into a disturbance factor of the dynamic phase vector, wherein the disturbance factor is calculated based on the product of the signal attenuation gradient and the grid topology coefficient; Mapping the communication channel congestion warning information to an attenuation coefficient of the probability amplitude, where the attenuation coefficient is negatively correlated with the channel packet loss rate; In the manifold space, the geographical boundaries of the clustered meter reading periods are encoded as spatial position constraints of the energy potential wells, and the label tree hierarchy of the differentiated collection dimensions is converted into orthogonality constraints of the basis vectors; When the signal attenuation gradient exceeds the load threshold of the equipment health warning period, dynamic phase synchronization of adjacent grids is triggered to migrate high-load tasks to low-load periods. The collection dimensions of non-critical industrial users' equipment are stripped away based on the attenuation coefficient to dynamically schedule human resources, data resources, network resources, and hardware resources. The meter identifier is used to generate a data retransmission token, and the grid topology location code is used to bind the relationship between meter reading tasks and the physical conduction path of the grid. The dynamic scheduling of human resources, data resources, network resources and hardware resources includes: The grid residence time scheduling matrix is ​​input into the skill matching engine. Based on the grid topology location code corresponding to the QR code metal identification piece, the matching degree between the meter reader skill vector and the geographical field intensity gradient is calculated to generate a dynamic scheduling sequence. According to the pulse timing alignment parameters, the probability amplitude branch of the non-urgent task is activated during the low-load period, and the corresponding data storage partition is marked as a delayed write mode; based on the probability amplitude distribution, the storage partition data is dynamically compressed and decompressed, and the non-critical user time window data in the type feature is compressed and stored; When it is detected that the harmonic energy distribution characteristics triggered by high-frequency demand from industrial users exceed a certain threshold, a bandwidth expansion request pulse is generated based on the disturbance factor. The causal neural field pulse network generates a strong pulse signal. The pulse intensity is normalized by the operator's configuration coefficient to generate a bandwidth increment value. The bandwidth increment value is sent to the corresponding regional base station in real time through the SDN controller to dynamically adjust the communication priority of the meter cluster. The bandwidth increment is calculated based on the product of the pulse strength and the operator configuration coefficient; When the network interrupts the propagation of abnormal pulses, a dynamic topology association token is generated based on the identification of the meter, and the token priority is dynamically adjusted by the risk level of the equipment health warning period to ensure that key equipment data is reissued according to the calibrated field strength gradient.

2. The AI-based meter reading data management and resource scheduling optimization method according to claim 1 is characterized in that: The reorganized meter reading data is multi-scale associated with the time dimension to generate a multi-layered data feature system consisting of a basic data layer, a dimensional feature layer, and a cross-dimensional fusion layer, including: Based on the time dimension, the reorganized meter reading data is calibrated into four time slices: minutes, hours, days, and months. Each time slice is injected with climate anomaly factor codes and device event flags to generate time series data blocks with spatiotemporal attributes. Differentiated feature coupling operations are performed for different time slices. At the minute level, the frequency of load mutations is extracted to form transient features. At the hour level, harmonic energy distribution features are generated through spectrum analysis. At the daily level, the load peak and valley attenuation rates are calculated. At the monthly level, the equipment health trend vector is generated, forming a time series feature layer that is linked to the dynamic indexing system. Performing a spatial convolution operation on the harmonic energy distribution features in the time series feature layer and the regional features, performing behavioral correlation mapping on the load peak-valley attenuation rate and the load curve shape of the user type dimension, and correcting the device distribution density using the device health trend vector to generate a three-dimensional fusion matrix; Based on the three-dimensional fusion matrix, a multi-level data feature system is constructed, including a basic data layer, a dimensional feature layer and a cross-dimensional fusion layer.

3. The AI-based meter reading data management and resource scheduling optimization method according to claim 1 is characterized in that: The adjustment factors include meter reading cycles for different geographical areas, differentiated collection dimensions for user types, task priority sorting in the time dimension, matching rules between human skills and task areas, sampling frequency and self-test cycle configuration of meter hardware, and data collection strategy and network channel priority mapping; The network load condition includes the base station load heat map data in the area and the communication channel congestion warning information.

4. An AI-based meter reading data management and resource scheduling optimization system, characterized in that: include: A meter reading data acquisition module is used to obtain meter reading data of a meter, wherein the meter has a corresponding metal identification piece with a QR code, and the QR code on the metal identification piece stores the identification of the meter and the grid topology location code; The meter reading data management module is used to construct a dynamic index system based on the geographical area dimension, the user type dimension, and the time dimension; wherein the geographical area dimension is divided into grid codes according to the power grid topology, and the user type dimension is divided into label tree levels; the meter reading data is hierarchically reorganized based on the dynamic index system; wherein, in the geographical area dimension, the meter reading data is spatially clustered according to the grid codes to generate regional features including real-time load density and equipment distribution density; in the user type dimension, user features are matched by label tree traversal to generate type features including load curve morphology and equipment configuration identification; the reorganized meter reading data is multi-scale associated with the time dimension to generate a multi-layered data feature system including a basic data layer, a dimensional feature layer, and a cross-dimensional fusion layer, where user types include residential type and industrial type; The usage trend determination module is used to map the characteristics of the basic data layer, the characteristics of the dimensional feature layer and the characteristics of the cross-dimensional fusion layer to the nodes of the causal neural field pulse network model; wherein, the spatial field strength of the geographic grid node is constructed based on the grid code and carries the load parameter, the behavioral pulse intensity of the user type node is generated based on the load curve shape and equipment configuration identification of the label tree, and the periodic pattern of the time node is modulated by the attenuation rate of the healthy equipment and the load fluctuation characteristics; wherein, the causal neural field pulse network model is trained based on historical power grid topology data and load data, and the optimization goal is to minimize the simulation deviation of the power grid conduction path; wherein, the training process of the causal neural field pulse network model includes: when backpropagation adjusts the field strength weight, if the actual load diffusion speed of a line is 20% faster than the model prediction value, then the field strength coupling coefficient of the line node and the downstream node is increased by 10% ; Construct a physical conduction path of the power grid in the causal neural field pulse network model; wherein, the load conduction path between geographic grid nodes is dynamically modeled according to the equipment distribution density gradient and the transmission impedance parameter to simulate the diffusion process of electric energy along the power grid topology; the load impact between user type nodes is constrained by the hierarchical relationship of the label tree; the geographic grid node load parameters, user type node behavior pulse sequence and time node periodic pattern output by the causal neural field pulse network model are mapped into feature tensors of geographic area dimension, user type dimension and time dimension respectively to form an initial feature cube; based on the physical conduction path of the power grid, a spatiotemporal evolution operator is designed to update the initial feature cube to generate a dynamic feature cube, wherein the load diffusion feature matrix of the geographic area dimension is driven by the equipment distribution density gradient; the behavior feature propagation matrix of the user type dimension is constrained by the label tree hierarchy; the time series feature modulation vector of the time dimension is fused with the healthy equipment attenuation rate parameter; high load density areas, abnormal user groups and equipment health warning periods are extracted from the dynamic feature cube to generate a three-dimensional trend map to determine the energy consumption trend in the future cycle; A resource scheduling optimization module is used to determine the adjustment factors of the meter reading task; the regional characteristics and type characteristics in the three-dimensional trend map and the dynamic index system are input into the digital twin engine to construct the basic data layer of the digital twin model; the spatial field strength of the geographic grid node in the three-dimensional trend map is encoded into a dynamic phase vector through a manifold learning network, the behavioral pulse intensity of the user type node is converted into a probability amplitude, and the time window of the equipment health warning period is used as a probability modulation operation sequence to form a spatiotemporal constraint parameter including the spatial field strength gradient, behavioral pulse probability and time series risk level; based on the topological structure of the physical conduction path of the power grid, the load conduction path between nodes is dynamically modeled according to the equipment distribution density gradient and the transmission impedance parameter, and a dynamic topological association channel is established between the geographic grid node and the user type node, so that the spatiotemporal attributes of the meter reading task and the physical characteristics of the power grid are bound to the digital twin model architecture through the causal neural field pulse network; wherein, the spatiotemporal attributes include grid location, user type label and task time window; the physical characteristics of the power grid include the conduction path Impedance and load diffusion direction; in the manifold space, the low-dimensional manifold basis vectors are extracted by orthogonal decomposition of the dynamic feature cube to construct an energy potential well model of the meter reading task path, and the arrangement order of the task nodes is driven by the gravitational gradient between the potential wells. The gravitational gradient between the potential wells is dynamically calculated according to the device distribution density gradient to generate a task scheduling skeleton of the digital twin model containing task priority and path dependency; a virtualized meter cluster is deployed in the digital twin model, and the simulation behavior of each meter node is regulated by the conduction path of the causal neural field pulse network. The interference of the network load on the integrity of the meter reading data is simulated by the pulse propagation delay to form an interactive digital twin simulation environment; the twin data stream generated by the simulation is reversely injected into the geographic node field strength calculation module of the three-dimensional trend map, and the phase parameters of the dynamic topology associated channel are calibrated by comparing the deviation between the virtual meter pulse signal and the actual power grid conduction path, and the digital twin model including the grid residence time scheduling matrix, data storage partition strategy and pulse timing alignment parameters is output; Among them, the grid residence time scheduling matrix is ​​used to indicate the length of time meter readers stay in different geographical grids based on the equipment distribution density gradient and task priority; the network load situation is obtained through the operator's network status interface; the signal attenuation gradient of the regional base station load heat map is mapped to the disturbance factor of the dynamic phase vector, and the disturbance factor is calculated based on the product of the signal attenuation gradient and the grid topology coefficient; the communication channel congestion warning information is mapped to the attenuation coefficient of the probability amplitude, and the attenuation coefficient is negatively correlated with the channel packet loss rate; in the manifold space, the geographical boundaries of the group meter reading cycle are encoded as the spatial position constraints of the energy potential well, and the label tree hierarchy of the differentiated acquisition dimension is converted into the orthogonality constraint of the basis vector; when the signal attenuation gradient exceeds the load threshold of the equipment health warning period, the dynamic phase synchronization of the adjacent grids is triggered, and the high-load task is migrated to the low-load period;The acquisition dimensions of non-critical equipment of industrial users are stripped away based on the attenuation coefficient to dynamically schedule human resources, data resources, network resources, and hardware resources. The meter identifier is used to generate a data retransmission token, and the grid topology location code is used to bind the relationship between the meter reading task and the physical conduction path of the grid. Among them, the dynamic scheduling of human resources, data resources, network resources and hardware resources includes: inputting the grid residence time scheduling matrix into the skill matching engine, calculating the matching degree between the meter reader skill vector and the geographical field intensity gradient based on the grid topology position code corresponding to the two-dimensional code metal identification piece, and generating a dynamic scheduling sequence; according to the pulse timing alignment parameters, activating the probability amplitude branch of non-emergency tasks during low-load periods, and marking the corresponding data storage partition as delayed write mode; dynamically compressing and decompressing the storage partition data based on the probability amplitude distribution, and compressing and storing the non-critical user time window data in the type characteristics; when a high-frequency demand trigger of an industrial user is detected, When the harmonic energy distribution characteristics exceed a certain threshold, a bandwidth expansion request pulse is generated based on the disturbance factor, and the causal neural field pulse network generates a strong pulse signal. After the pulse intensity is normalized by the operator configuration coefficient, a bandwidth increment value is generated; the bandwidth increment value is sent to the corresponding regional base station in real time through the SDN controller to dynamically adjust the communication priority of the meter cluster; the bandwidth increment is calculated based on the product of the pulse intensity and the operator configuration coefficient; when the network interrupts the propagation of abnormal pulses, a dynamic topology association token is generated based on the identification of the meter, and the token priority is dynamically adjusted by the risk level of the equipment health warning period to ensure that the key equipment data is reissued according to the calibrated field strength gradient.

Images