Big data full life cycle state monitoring method and running self-feedback electric energy meter

Abstract

This invention discloses a big data full lifecycle status monitoring method and a self-feedback energy meter, comprising the following steps: S1, generating a set of energy meter twins based on energy meter operating data; S2, extracting state parameter vectors, spatial coordinates, and layer attribute representation vectors to generate nested state embedding tensors; S3, constructing a state inference tensor field; S4, calculating the cumulative score value of abnormal trends and marking states requiring feedback; S5, combining trend direction, change amplitude, and neighborhood state density to generate a set of feedback-induced actions; S6, adjusting the energy meter's data reading cycle, communication frequency, and state self-marking to generate feedback execution records; S7, reconstructing the state inference tensor field to form a closed loop of status monitoring and autonomous feedback. This invention achieves accurate monitoring and rapid feedback of energy meter status, improving system stability and intelligent operation management.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and in particular to a big data-based full lifecycle condition monitoring method and an operating self-feedback energy meter. Background Technology

[0002] Against the backdrop of the continuous deepening of digital management of power distribution terminals, electricity meters, as key sensing devices, undertake the important functions of energy consumption data reading and status monitoring. In the existing technology, electricity meters generally adopt a fixed period reading and uniform frequency reporting strategy, which cannot dynamically adjust the sensing intensity and communication behavior according to the operating status. These shortcomings lead to low resource utilization efficiency and delayed feedback response in application scenarios with varying stability, making it difficult to meet the needs of refined management.

[0003] Current methods combine GIS maps to construct spatial information models of electricity meters, enabling geographic location visualization. However, they lack the ability to collaboratively model spatial attributes and operational status, cannot establish trend reasoning mechanisms based on spatial adjacency relationships and differences in state parameters, and lack the ability to fuse high-dimensional information. Most existing studies focus on static anomaly judgment of individual electricity meter states, ignoring the interaction between neighboring electricity meters during state evolution. This makes it difficult to accurately identify and warn of local state change trends.

[0004] Furthermore, current methods generally lack anomaly scoring mechanisms based on trend evolution rates. Current threshold-based comparisons are insufficient to handle minor disturbances and potential abnormal growth paths under complex operating conditions. The granularity of state judgment is inadequate, and the feedback control mechanism of the electricity meter also suffers from a single structure. This makes it impossible to dynamically generate customized feedback strategies based on the trend of equipment state changes and the surrounding environment, and also makes it impossible to achieve intelligent adjustment and closed-loop control of the electricity meter itself.

[0005] While existing digital twin methods have made some progress in device modeling, their application in the field of electricity meters is mostly limited to static data mapping. They lack the ability to fuse state parameters and spatial features for modeling and have not yet formed a dynamic reasoning and feedback path driven by state evolution trends. To address these issues, the field needs a full lifecycle monitoring method that integrates spatial location, state parameters, and trend evolution modeling to construct electricity meters with autonomous feedback capabilities, thereby improving the depth of operational perception and the level of intelligent response.

[0006] Therefore, how to provide a big data full life cycle status monitoring method and an operating self-feedback energy meter is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] One objective of this invention is to propose a big data full lifecycle status monitoring method and an operating self-feedback energy meter. This invention utilizes location-aware modeling, trend reasoning scoring, and dynamic feedback generation mechanisms to construct a digital twin system that supports closed-loop status updates, possessing the advantages of precise perception, timely response, and intelligent feedback.

[0008] A method for monitoring the entire lifecycle status of big data according to an embodiment of the present invention includes the following steps:

[0009] S1. Based on the data of each electricity meter, construct a set of electricity meter twins;

[0010] S2. Extract the state parameter vector, spatial coordinates and layer attribute representation vector of each electricity meter twin, input them into the modeling structure for fusing state and spatial features, introduce a position-aware attention mechanism, encode the spatial coordinates, participate in feature fusion calculation, and generate a nested state embedding tensor.

[0011] S3. Output the state inference tensor field based on the nested state embedding tensor, the difference in state parameter vectors between adjacent energy meter twins, and the spatial distance.

[0012] S4. Extract the continuous state change rate of each electricity meter twin from the state reasoning tensor field, calculate the cumulative score of abnormal trends, and automatically mark the electricity meter twin as a state that needs feedback when the cumulative score of abnormal trends exceeds the preset threshold.

[0013] S5. Generate a set of feedback-induced actions based on the trend direction, change amplitude, and adjacent state parameter vector density of the energy meter twin in the state inference tensor field that requires feedback.

[0014] S6. The energy meter automatically performs data reading cycle adjustment, communication frequency adjustment and operation status self-marking operations based on the feedback induced action set, records the feedback execution data, and generates feedback execution records.

[0015] S7. Update the feedback execution record and the state parameter vector after execution to the electricity meter twin, update the nested state embedding tensor, reconstruct the state inference tensor field, and realize the closed loop of electricity meter state monitoring and autonomous feedback.

[0016] Optionally, S1 specifically includes: calculating a stability score based on the fluctuation of operating data and communication stability of each energy meter in the most recent continuous period. The stability score is calculated by weighting voltage variance, current fluctuation rate, active power offset rate, and number of communication interruptions. Based on the stability score, the energy meters are divided into high stability level, medium stability level, and low stability level: high stability level energy meters are those with a stability score below a first threshold, and the sampling period is set to 6 hours; medium stability level energy meters are those with a stability score between the first threshold and the second threshold, and the sampling period is set to 1 hour; low stability level energy meters are those with a stability score above the second threshold, and the sampling period is set to 10 minutes. The constructed energy meter twin includes a state parameter vector, spatial coordinates, and layer attribute representation vector.

[0017] Optionally, S2 specifically includes:

[0018] S21. Extract the state parameter vector, spatial coordinates, layer attribute representation vector, and stability level identifier for each energy meter twin. The state parameter vector includes voltage, current, active power, reactive power, power factor, and sampling time identifier. The spatial coordinates are the longitude and latitude of the energy meter. The layer attribute representation vector includes power supply path level, physical installation type, and topology identifier. The stability level identifier indicates whether the energy meter belongs to a high stability level, a medium stability level, or a low stability level.

[0019] S22. Based on spatial coordinates, position coding is performed. An improved multi-scale sine and cosine function is used to encode longitude and latitude respectively to generate a position coding vector. The position coding vector is concatenated with the layer attribute representation vector to form a spatial structure representation vector.

[0020] S23. The state parameter vector, spatial structure representation vector and stability level identifier are concatenated into a set of fusion input vectors and input to the position-aware attention mechanism. The position-aware attention mechanism generates attention weights by calculating the spatial correlation between the fusion input vectors.

[0021] S24. Based on attention weights, perform weighted fusion of the fused input vector set to generate a nested state embedding tensor for each electricity meter twin.

[0022] Optionally, the process of encoding longitude and latitude using an improved multi-scale sine and cosine function in step S22 includes:

[0023] S221. Let the longitude of each electricity meter twin be x, the latitude be y, the coding dimension be d, the coding sequence number of each dimension be i, and the perturbation parameter sequence corresponding to the i-th dimension be δ. i The frequency disturbance parameter is φ i The phase perturbation parameter is ψ iThe tensor weighting coefficient corresponding to the stability level is ω. s Calculate the longitude code value e i With latitude code value l i The nested expression is:

[0024]

[0025] Where, ω s The weighting coefficients corresponding to the stability levels of the electricity meter are 1.0 for high stability level, 1.3 for medium stability level, and 1.7 for low stability level.

[0026] S222, Encode the longitude values ​​e1, e2, ..., e along all dimensions. d With latitude coding values ​​l1, l2, ..., l d Concatenate them into a position encoding vector of length 2d.

[0027] Optionally, the location-aware attention mechanism in step S23 specifically includes the following processing steps: Based on the state parameter vector and the location encoding vector, normalization operations are performed to form a state input sequence and a location input sequence, respectively. The location input sequence is split according to the longitude and latitude dimensions to generate a sensitive guidance sequence for controlling the direction of attention distribution. Combining the Euclidean distance of the spatial coordinates between the energy meter twins, the grouping differences of the stability levels, and the change amplitude of the state parameter vector, a multi-factor guidance matrix for measuring the degree of inter-correlation is constructed as the location guidance weight in the attention mechanism. During the attention calculation process, the weight distribution relationship between the state input sequences is jointly adjusted according to the sensitive guidance sequence and the location guidance weight. The state input sequences are dynamically weighted before fusion based on the weight adjustment coefficients corresponding to the high stability, medium stability, and low stability levels.

[0028] Optionally, S4 specifically includes:

[0029] S41. Extract the state parameter vector of each energy meter twin in the state reasoning tensor field at continuous sampling time steps, and construct the time-series state trajectory.

[0030] S42. Perform time difference operation on each state trajectory to obtain the state change rate between each time step, and mark the time step with the change rate greater than the fluctuation threshold corresponding to the stability level as an abnormal fluctuation point.

[0031] S43. For each electricity meter twin, count the number of consecutive abnormal fluctuation points within a fixed sampling time window, and combine the change rate value with the consistency coefficient of the change direction to generate a trend disturbance score.

[0032] S44. Perform time window integration processing based on the trend disturbance score and the duration of abnormal fluctuation points to form an abnormal trend cumulative score;

[0033] S45. Compare the cumulative score of abnormal trends with the preset threshold for stability level classification. If the score exceeds the upper limit threshold corresponding to the current stability level of the electricity meter twin, the electricity meter twin is automatically marked as requiring feedback.

[0034] Optionally, S5 specifically includes:

[0035] In the energy meter twin marked as requiring feedback, the state parameter vectors of the energy meter twin within three consecutive time steps in the state inference tensor field are extracted. The state evolution trend vector is formed based on the direction of change of the state parameter vector. The density distribution of the state parameter vectors of the energy meter twin within the third order of the spatial neighborhood is calculated. The density gradient changes in each direction are statistically analyzed, and a density direction vector is constructed. The cosine value of the angle between the state evolution trend vector and the density direction vector is used to measure the degree of trend consistency. The degree of consistency, the mean of the change amplitude, and the density intensity are used as joint feature inputs for rule mapping to determine the optimal feedback direction and feedback action type of the energy meter twin in the current state.

[0036] The strategy for adjustment based on stability level feedback includes the following: When the stability level is high, the data reading cycle adjustment step size is 10% of the current cycle, the communication frequency adjustment factor is the base frequency minus 20 seconds, and the status self-marking type is status maintenance; when the stability level is medium, the data reading cycle adjustment step size is 30% of the current cycle, the communication frequency adjustment factor is the base frequency minus 10 seconds, and the status self-marking type is dynamic observation; when the stability level is low, the data reading cycle adjustment step size is 50% of the current cycle, the communication frequency adjustment factor is not adjusted to the base frequency, and the status self-marking type is high-frequency monitoring.

[0037] An electricity meter according to an embodiment of the present invention includes the following modules:

[0038] The data reading module is used to read raw measurement data and construct the energy meter state parameter vector to generate a set of energy meter twins;

[0039] The spatial fusion module is used to extract state parameter vectors, spatial coordinates, and layer attribute representation vectors, and to perform fusion feature modeling.

[0040] The position encoding module is used to perform position-aware encoding of spatial coordinates and generate nested state embedding tensors;

[0041] The state reasoning module is used to calculate the state difference and spatial distance, and output the state reasoning tensor field.

[0042] The anomaly detection module is used to extract the rate of change of continuous states, calculate the cumulative score of abnormal trends, and mark the states that need to be fed back.

[0043] The feedback generation module is used to generate a set of feedback-induced actions based on the trend direction, change amplitude, and density.

[0044] The self-feedback execution module is used to perform sampling period adjustment, communication frequency adjustment and status self-marking operations, and generate feedback execution records.

[0045] The state update module is used to update the state parameter vector and nested state embedding tensor of the electricity meter twin, and reconstruct the state inference tensor field.

[0046] The beneficial effects of this invention are:

[0047] Improving the accuracy and efficiency of electricity meter status monitoring: This invention constructs a nested state embedding tensor and a state inference tensor field to achieve joint modeling of electricity meters in the dimensions of spatial correlation and state evolution. It can accurately extract the rate of continuous state change and perform trend identification, thereby improving the accuracy of anomaly detection and the foresight of response.

[0048] Realizing active feedback and dynamic control of electricity meters: This invention generates a differentiated set of feedback-induced actions through joint analysis based on trend direction, change amplitude and neighborhood state density, enabling electricity meters to adaptively adjust sampling period, communication frequency and state self-labeling capabilities, and complete a closed-loop control process from state recognition to feedback execution.

[0049] Improving resource utilization efficiency by constructing a hierarchical sensing mechanism: This invention divides the stability level of electricity meters and sets differentiated data reading strategies to achieve low-frequency sensing for high-stability level devices, medium-frequency sensing for medium-stability level devices, and high-frequency sensing for low-stability level devices, thereby optimizing the allocation of sensing resources and reducing communication and computing overhead. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This invention presents a big data full lifecycle status monitoring method and an overall flowchart of a self-feedback energy meter.

[0052] Figure 2 This is a flowchart illustrating the feedback-induced action set generation and execution process of a big data full lifecycle status monitoring method proposed in this invention. Detailed Implementation

[0053] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0054] refer to Figure 1 and Figure 2 A method for monitoring the entire lifecycle status of big data includes the following steps:

[0055] S1. Based on the data of each electricity meter, construct a set of electricity meter twins;

[0056] S2. Extract the state parameter vector, spatial coordinates and layer attribute representation vector of each electricity meter twin, input them into the modeling structure for fusing state and spatial features, introduce a position-aware attention mechanism, encode the spatial coordinates, participate in feature fusion calculation, and generate a nested state embedding tensor.

[0057] S3. Output the state inference tensor field based on the nested state embedding tensor, the difference in state parameter vectors between adjacent energy meter twins, and the spatial distance. Specifically, in the construction of the state inference tensor field, firstly, the current state expression is extracted based on the nested state embedding tensor of each energy meter twin. Then, the set of adjacent twins is selected and the element-wise difference between the state parameter vectors is calculated to represent the local state change. Subsequently, the spatial distance is calculated by combining the spatial coordinate information between the energy meter twins. A Gaussian decay weight function is introduced to perform position-sensitive weighting on the state difference to enhance the influence of the nearest neighbor change. Finally, the weighted state difference is superimposed and mapped in the spatial adjacency structure to form a state inference tensor field with a continuous state evolution trend.

[0058] S4. Extract the continuous state change rate of each electricity meter twin from the state reasoning tensor field, calculate the cumulative score of abnormal trends, and automatically mark the electricity meter twin as a state that needs feedback when the cumulative score of abnormal trends exceeds the preset threshold.

[0059] S5. Generate a set of feedback-induced actions based on the trend direction, change amplitude, and adjacent state parameter vector density of the energy meter twin in the state inference tensor field that requires feedback.

[0060] S6. The energy meter automatically performs data reading cycle adjustment, communication frequency adjustment and operation status self-marking operations based on the feedback induced action set, records the feedback execution data, and generates feedback execution records.

[0061] S7. Update the feedback execution record and the state parameter vector after execution to the electricity meter twin, update the nested state embedding tensor, and reconstruct the state inference tensor field to realize the closed loop of electricity meter state monitoring and autonomous feedback. Specifically, first, write the feedback execution record and the updated state parameter vector generated after each electricity meter performs the feedback induced action into the corresponding electricity meter twin to form a dynamically evolving state entity; then, based on the latest twin set, re-extract the state parameter vector, spatial coordinates, and layer attribute representation vector, and input them into the fusion modeling structure, and update the nested state embedding tensor through a position-aware attention mechanism; on this basis, combine the state difference and spatial distance between adjacent twins to regenerate the state inference tensor field, realize continuous monitoring and real-time adjustment of the overall state evolution trend, and thus form a closed-loop control process of self-updating, self-sensing, and self-feedback.

[0062] This invention constructs a set of electricity meter twins with state parameter vectors as the core, and integrates position-aware attention modeling, state reasoning tensor field construction and feedback induction mechanism to realize closed-loop control of the electricity meter from perception and judgment to feedback. Compared with the traditional mode of only monitoring or periodic reporting, this invention has the advantages of high accuracy of state evolution monitoring, flexible feedback triggering mechanism and strong structural response timeliness, and can adapt to remote intelligent management of various types of electricity meters under different operating environments.

[0063] In this embodiment, S1 specifically includes: calculating a stability score based on the fluctuation of operating data and communication stability of each energy meter in the most recent continuous period. The stability score is obtained by weighting voltage variance, current fluctuation rate, active power offset rate, and number of communication interruptions. Based on the stability score, energy meters are classified into high stability level, medium stability level, and low stability level: high stability level energy meters are those with a stability score below a first threshold, and their data reading period is set to 6 hours; medium stability level energy meters are those with a stability score between the first and second thresholds, and their data reading period is set to... The data reading cycle is set to 1 hour; low-stability level energy meters are those with a stability score higher than the second threshold, and the data reading cycle is set to 10 minutes. The constructed energy meter twin includes a state parameter vector, spatial coordinates, and layer attribute representation vector. The data reading level and data reading cycle together constitute the state configuration parameter, which is embedded as an additional field in the energy meter state parameter vector. It is used to participate in the state feature fusion calculation in S2, to dynamically control the calculation window of the state difference in S3, to form a level-sensitive early warning by combining the cumulative score of abnormal trends in S4, and to guide the execution frequency and feedback priority of the set of feedback-induced actions in S6.

[0064] This invention addresses the differences in the stability of electricity meters by proposing a tiered data reading strategy based on stability scores. It constructs a set of multi-sampling frequency twins for high, medium, and low stability levels. This strategy improves data utilization efficiency, reduces redundant communication burden, and ensures timely capture of critical state changes. This mechanism has on-demand scheduling capabilities, enhances the precision and flexibility of electricity meter status monitoring, and overcomes the problems of extensive and delayed feedback responses in existing technologies.

[0065] In this embodiment, S2 specifically includes:

[0066] S21. Extract the state parameter vector, spatial coordinates, layer attribute representation vector, and stability level identifier for each energy meter twin. The state parameter vector includes voltage, current, active power, reactive power, power factor, and sampling time identifier. The spatial coordinates are the longitude and latitude of the energy meter. The layer attribute representation vector includes power supply path level, physical installation type, and topology identifier. The stability level identifier indicates whether the energy meter belongs to a high stability level, a medium stability level, or a low stability level.

[0067] S22. Based on spatial coordinates, position coding is performed. An improved multi-scale sine and cosine function is used to encode longitude and latitude respectively to generate a position coding vector. The position coding vector is concatenated with the layer attribute representation vector to form a spatial structure representation vector.

[0068] S23. The state parameter vector, spatial structure representation vector and stability level identifier are concatenated into a set of fusion input vectors and input to the location-aware attention mechanism. The location-aware attention mechanism generates attention weights by calculating the spatial correlation between the fusion input vectors, which are used to enhance the state linkage representation capability between spatially adjacent energy meters. At the same time, the feature fusion intensity is dynamically adjusted based on the stability level difference.

[0069] S24. Based on attention weights, perform weighted fusion of the fused input vector set to generate a nested state embedding tensor for each electricity meter twin.

[0070] This invention introduces a bidirectional embedding mechanism of layer attribute representation and spatial location encoding into the fusion modeling structure. It combines a position-aware attention mechanism to enhance the coupling and expression capabilities of state and spatial features, and constructs a nested state embedding tensor as the core intermediate representation. Compared with existing isolated modeling methods, this method can improve spatial awareness and fine-grained state expression, making subsequent state reasoning more context-adaptive.

[0071] In this embodiment, the process of encoding longitude and latitude using an improved multi-scale sine and cosine function in step S22 includes:

[0072] S221. Let the longitude of each electricity meter twin be x, the latitude be y, the coding dimension be d, the coding sequence number of each dimension be i, and the perturbation parameter sequence corresponding to the i-th dimension be δ. i The frequency disturbance parameter is φ i The phase perturbation parameter is ψ i The tensor weighting coefficient corresponding to the stability level is ω. s Calculate the longitude code value e i With latitude code value l i The nested expression is:

[0073]

[0074] Where, ω s The weighting coefficients corresponding to the stability levels of the electricity meter are 1.0 for high stability level, 1.3 for medium stability level, and 1.7 for low stability level.

[0075] S222, Encode the longitude values ​​e1, e2, ..., e along all dimensions. d With latitude coding values ​​l1, l2, ..., l d Concatenate them into a position encoding vector of length 2d.

[0076] The location encoding formula constructed in this invention introduces dimensional perturbation parameters, frequency perturbation parameters, phase perturbation parameters, and tensor weighting coefficients corresponding to stability levels, thereby improving and optimizing the traditional sine and cosine encoding method. By dynamically weighting longitude and latitude at different scales and superimposing modulation signals driven by layer categories, the generated location encoding vector has the characteristics of strong direction resolution, high layer correlation, and fast response to local changes. Compared with the existing fixed-period encoding method, this encoding method enhances the structural distinguishability of spatial representation and the fusion efficiency of state embedding, providing a more distinguishable and stable spatial structure representation for subsequent state-space coupling modeling, and improving the spatial adaptability and state modeling robustness of the electricity meter twin in multiple scenarios.

[0077] In this embodiment, the location-aware attention mechanism in step S23 specifically includes the following processing steps: Based on the state parameter vector and the location encoding vector, normalization operations are performed to form a state input sequence and a location input sequence, respectively. The location input sequence is split according to the longitude and latitude dimensions to generate a sensitive guidance sequence for controlling the direction of attention distribution. Combining the Euclidean distance of the spatial coordinates between the energy meter twins, the grouping differences of the stability levels, and the change amplitude of the state parameter vector, a multi-factor guidance matrix for measuring the degree of inter-correlation is constructed as the location guidance weight in the attention mechanism. During the attention calculation process, the weight distribution relationship between the state input sequences is jointly adjusted according to the sensitive guidance sequence and the location guidance weight. The state input sequences are dynamically weighted before fusion according to the weight adjustment coefficients corresponding to the high stability, medium stability, and low stability levels.

[0078] The position-aware attention mechanism designed in this invention integrates a weighted strategy based on spatial position guidance, a stability level sensitivity adjustment factor, and an attention gating layer control structure to generate a fused feature vector to enhance the accuracy of state inference. This mechanism can solve the technical difficulties of insufficient spatial distribution sensitivity and poor response to data reading stability in existing attention mechanisms. It can realize the dynamic coupling expression of position and stability in state modeling and improve the model's generalization ability to heterogeneous states.

[0079] In this embodiment, S4 specifically includes:

[0080] S41. Extract the state parameter vector of each energy meter twin in the state reasoning tensor field at continuous sampling time steps, and construct the time-series state trajectory.

[0081] S42. Perform time difference operation on each state trajectory to obtain the state change rate between each time step, and mark the time step with the change rate greater than the fluctuation threshold corresponding to the stability level as an abnormal fluctuation point.

[0082] S43. For each electricity meter twin, count the number of consecutive abnormal fluctuation points within a fixed sampling time window, and combine the change rate value with the consistency coefficient of the change direction to generate a trend disturbance score.

[0083] S44. Perform time window integration processing based on the trend disturbance score and the duration of abnormal fluctuation points to form an abnormal trend cumulative score;

[0084] S45. Compare the cumulative score of abnormal trends with the preset threshold for stability level classification. If the score exceeds the upper limit threshold corresponding to the current stability level of the electricity meter twin, the electricity meter twin is automatically marked as requiring feedback.

[0085] This invention constructs a time series of state change rates and, based on dynamic trend extraction, anomaly cumulative score calculation, and stability weighted evaluation, accurately marks the abnormal trends of electricity meter states. Compared with static threshold discrimination methods, this invention achieves trend sensitivity, steady-state response, and false alarm suppression in the anomaly identification process, which can improve the accuracy of anomaly detection and the reliability of the triggering mechanism, and also enhance the practicality of state feedback of digital twin models.

[0086] In this embodiment, S5 specifically includes:

[0087] In the energy meter twin marked as requiring feedback, the state parameter vectors of the energy meter twin within three consecutive time steps in the state inference tensor field are extracted. The state evolution trend vector is formed based on the direction of change of the state parameter vector. The density distribution of the state parameter vectors of the energy meter twin within the third order of the spatial neighborhood is calculated. The density gradient changes in each direction are statistically analyzed, and a density direction vector is constructed. The cosine value of the angle between the state evolution trend vector and the density direction vector is used to measure the degree of trend consistency. The degree of consistency, the mean of the change amplitude, and the density intensity are used as joint feature inputs for rule mapping to determine the optimal feedback direction and feedback action type of the energy meter twin in the current state.

[0088] The strategy for adjustment based on stability level feedback includes the following: When the stability level is high, the data reading cycle adjustment step size is 10% of the current cycle, the communication frequency adjustment factor is the base frequency minus 20 seconds, and the status self-marking type is status maintenance; when the stability level is medium, the data reading cycle adjustment step size is 30% of the current cycle, the communication frequency adjustment factor is the base frequency minus 10 seconds, and the status self-marking type is dynamic observation; when the stability level is low, the data reading cycle adjustment step size is 50% of the current cycle, the communication frequency adjustment factor is not adjusted to the base frequency, and the status self-marking type is high-frequency monitoring.

[0089] This invention constructs a set of feedback-induced actions through joint analysis of trend direction, change magnitude, and neighborhood density. These actions are further refined into fine-tuning of data reading cycles, adjustment of communication frequency, and self-marking operation strategies for operating status. This feedback mechanism achieves a closed-loop linkage of state triggering, strategy generation, and operation execution. Compared with manually setting feedback rules, it has dynamic adaptive capabilities and precise strategy alignment characteristics. This method enhances the intelligence level of energy meter operation feedback.

[0090] An electricity meter according to an embodiment of the present invention includes the following modules:

[0091] The data reading module is used to read raw measurement data and construct the energy meter state parameter vector to generate a set of energy meter twins;

[0092] The spatial fusion module is used to extract state parameter vectors, spatial coordinates, and layer attribute representation vectors, and to perform fusion feature modeling.

[0093] The position encoding module is used to perform position-aware encoding of spatial coordinates and generate nested state embedding tensors;

[0094] The state reasoning module is used to calculate the state difference and spatial distance, and output the state reasoning tensor field.

[0095] The anomaly detection module is used to extract the rate of change of continuous states, calculate the cumulative score of abnormal trends, and mark the states that need to be fed back.

[0096] The feedback generation module is used to generate a set of feedback-induced actions based on the trend direction, change amplitude, and density.

[0097] The self-feedback execution module is used to perform sampling period adjustment, communication frequency adjustment and status self-marking operations, and generate feedback execution records.

[0098] The state update module is used to update the state parameter vector and nested state embedding tensor of the electricity meter twin, and reconstruct the state inference tensor field.

[0099] The proposed electricity meter structure includes a data reading unit, a location encoding unit, an embedded modeling unit, a trend recognition unit, a feedback execution unit, and a state update unit. The functions of each unit correspond one-to-one with the logic of the claims, forming a closed-loop feedback platform that integrates hardware and software. This structure effectively supports the execution of the entire process of digital twin, spatial modeling, and feedback control, and can realize a modular implementation path from state monitoring to self-feedback, thereby improving the overall system's deployability and engineering implementation capabilities.

[0100] Example 1:

[0101] To verify the feasibility of this invention in practice, it was applied to a large-scale smart meter status monitoring and operation feedback task deployed in a regional power distribution network. The region included different types of residential buildings, office buildings, and small industrial users, with a total of more than 6,000 meters. These meters had diverse operating states, communication frequencies, and spatial distribution characteristics, making it a typical scenario for testing the performance of this invention.

[0102] In practical applications, the real-time operating data of all electricity meters is first collected and read centrally. The data includes parameters such as active power, reactive power, voltage, current, and equipment temperature sampled every 15 minutes, and corresponding state parameter vectors are generated. The latitude and longitude coordinates and layer attributes of the electricity meters are obtained through GIS information and fused with the state parameter vectors to construct a set of electricity meter twins. Combining the location-aware attention mechanism and spatial structure coding strategy proposed in this invention, the spatial coordinates are improved with multi-scale sine and cosine coding. The location code and layer attribute representation vector are concatenated to form a spatial structure representation vector.

[0103] The spatial structure representation vector and the state parameter vector are input into the nested state modeling structure to generate a multidimensional nested state embedding tensor. By comparing the state parameter differences and spatial distance relationships between adjacent energy meter twins, a state inference tensor field is constructed. The continuous state change rate is extracted from this field, and the cumulative score of abnormal trends is calculated. Energy meter twins with scores exceeding a preset threshold are automatically identified as states requiring feedback. The set of feedback-induced actions is generated by combining the trend direction, change amplitude, and density of neighboring state parameters.

[0104] During the feedback induction process, the system performs differentiated operations based on stability levels. Meters with high stability levels primarily perform periodic state self-marking; meters with medium stability levels trigger sampling period compression and communication interval adjustment; and meters with low stability levels perform continuous high-frequency communication within a certain time window and record state fluctuations in real time. All feedback execution results are recorded as feedback execution records, which, combined with newly read state data, are updated to the corresponding twin. The nested state tensor and state inference tensor field are then reconstructed, forming a monitoring-feedback closed loop.

[0105] Two months of continuous monitoring and comparison revealed that this invention can effectively improve the accuracy and response speed of abnormal trend identification. Traditional methods have an average delay of 2.1 hours in detecting severe fluctuations, while the average delay of this invention is 17 minutes. The false alarm rate in traditional strategies is approximately 6.3%, while this invention reduces the false alarm rate to 2.1% with the same amount of data. In feedback closed-loop testing, the average stability score after feedback increased from 0.78 to 0.94, indicating that the feedback strategy has good intervention and control capabilities in actual operation.

[0106] In the actual deployment area, the average recovery time of the electricity meter status after feedback was reduced by 36%. For some low-stability edge nodes, the stability score improved by more than 0.2 due to the introduction of a high-frequency self-feedback mechanism. Simultaneously, the system's response capability to spatially clustered anomalies was significantly improved. The nested state embedding model can accurately identify the trend diffusion path in local areas, providing early warnings in the inference tensor field and ensuring that the risk range is controllable. The table below summarizes the comparison data of the electricity meter status before and after feedback:

[0107] Table 1: Comparison of Energy Meter Stability Scores Before and After Feedback

[0108]

[0109]

[0110] As can be seen from the "Comparison Data Table of Energy Meter State Stability Scores Before and After Feedback," the present invention significantly improves the state stability of energy meters before and after feedback. Firstly, in terms of the magnitude of the score improvement, all seven energy meters showed an increase in stability score after feedback, with an average improvement of 0.17. The improvement was particularly pronounced for low-stability energy meters; C-3112's score increased from 0.53 to 0.81, an improvement of 0.28; E-4578 increased from 0.49 to 0.78, an improvement of 0.29, demonstrating the strong intervention capability of the present invention when dealing with energy meters with poor stability. Medium-stability meters B-2093, D-1548, and G-3982 improved by 0.16, 0.18, and 0.15 respectively, demonstrating the effectiveness of the medium-amplitude feedback strategy for state adjustment. High-stability devices such as A-1035 and F-6671 improved by 0.04 and 0.05 respectively, with relatively smaller improvements, but demonstrating the present invention's ability to fine-tune state accuracy.

[0111] In terms of the reduction in state recovery time, the average reduction reached 32.6%, with the most significant improvement observed in low-stability devices. E-4578 saw a 52% reduction in recovery time, and C-3112 a 48% reduction, indicating that the high-frequency communication and state resampling strategy provides a fast response and tight feedback rhythm when dealing with unstable states. Medium-stability devices G-3982 saw a 36% reduction, B-2093 a 34% reduction, and D-1548 a 31% reduction, demonstrating the ability of the medium-frequency control mechanism to balance efficiency and resource consumption. High-stability devices A-1035 and F-6671 saw reductions of 15% and 12% respectively, indicating that even under initially good conditions, the state self-marking mechanism can adjust and optimize for small fluctuations.

[0112] In terms of the magnitude of the decrease in anomaly scores, the decrease was most significant for low-stability energy meters, with E-4578 decreasing by 0.29 and C-3112 decreasing by 0.28, indicating that the state fluctuations of high-risk equipment were effectively suppressed. The decreases for medium-stability equipment D-1548, B-2093, and G-3982 were 0.18, 0.16, and 0.15, respectively, showing concentrated and stable improvement in scores. High-stability equipment A-1035 and F-6671 also achieved decreases of 0.04 and 0.05, respectively, proving that the state awareness mechanism still has optimization value for high-stability energy meters.

[0113] Overall, all three types of stability-level energy meters benefit from the nested state modeling, state inference tensor field, and feedback-induced mechanism combination strategy proposed in this invention. Low-stability-level energy meters improve operational stability and response sensitivity through high-frequency communication and state resampling strategies. Medium-stability-level energy meters achieve steady-state transition through sampling period compression and communication interval optimization. High-stability-level energy meters maintain state accuracy and avoid potential disturbance accumulation through a lightweight state self-marking mechanism. This invention demonstrates good consistency and control capabilities in terms of the matching of hierarchical feedback strategies, the effectiveness of feedback action execution, and the continuity of state adjustment.

[0114] In summary, the feedback method based on constructing a state reasoning tensor field using nested state embedding tensors and combining it with a trend scoring mechanism has the characteristics of accurate identification, timely response, and effective regulation. It can realize continuous state monitoring and closed-loop self-feedback control of large-scale electricity meter twins, while improving the operational stability and data reliability of power sensing terminals.

[0115] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A big data full life cycle state monitoring method, characterized in that, The method comprises the following steps: S1, based on the data of each electric energy meter, a set of electric energy meter twins is constructed; S2, the state parameter vector, spatial coordinates and layer attribute representation vector of each electric energy meter twin are extracted and input into a modeling structure for fusing state and spatial features, a position-aware attention mechanism is introduced, the spatial coordinates are position-encoded, participate in feature fusion calculation, and a nested state embedding tensor is generated; S3, based on the nested state embedding tensor, the difference between the state parameter vectors of adjacent electric energy meter twins and the spatial distance, an output state reasoning tensor field is generated; S4, the continuous state change rate of each electric energy meter twin is extracted from the state reasoning tensor field, the abnormal trend cumulative score value is calculated, and when the abnormal trend cumulative score value exceeds the preset threshold value, the electric energy meter twin is automatically marked as a state that needs feedback; S5, according to the trend direction, change amplitude and adjacent state parameter vector density of the electric energy meter twin in the state reasoning tensor field, a feedback induction action set is generated; S6, the electric energy meter automatically executes data reading period adjustment, communication frequency adjustment and running state self-marking operation according to the feedback induction action set, records the feedback execution data, and generates a feedback execution record; S7, the feedback execution record and the state parameter vector after execution are updated to the electric energy meter twin, the nested state embedding tensor is updated, the state reasoning tensor field is reconstructed, and the electric energy meter state monitoring and autonomous feedback closed loop is realized.

2. The big data full life cycle state monitoring method of claim 1, wherein, The S1 specifically comprises: according to the running data volatility and communication stability of each electric energy meter in the latest continuous period, a stability score value is calculated, the stability score value is calculated by weighting the voltage variance, current fluctuation rate, active power offset rate and communication interruption times, and the electric energy meter is divided into high stability level, medium stability level and low stability level according to the stability score value: the high stability level electric energy meter is the electric energy meter with the stability score value lower than the first threshold value, the data reading period is set to 6 hours; the medium stability level electric energy meter is the electric energy meter with the stability score value between the first threshold value and the second threshold value, the data reading period is set to 1 hour; the low stability level electric energy meter is the electric energy meter with the stability score value higher than the second threshold value, the data reading period is set to 10 minutes, and the constructed electric energy meter twin includes the state parameter vector, the spatial coordinates and the layer attribute representation vector.

3. The big data full life cycle state monitoring method of claim 2, wherein, The S2 specifically comprises: S21, the state parameter vector, spatial coordinates, layer attribute representation vector and stability level identifier of each electric energy meter twin are extracted, the state parameter vector includes voltage, current, active power, reactive power, power factor and sampling time identifier, the spatial coordinates are the longitude and latitude of the electric energy meter, the layer attribute representation vector includes the supply path level, the physical installation type and the topology structure identifier, and the stability level identifier indicates that the electric energy meter belongs to the high stability level, the medium stability level or the low stability level; S22, position encoding is performed based on the spatial coordinates, an improved multi-scale sine and cosine function is used to encode the longitude and latitude respectively, a position encoding vector is generated, and the position encoding vector and the layer attribute representation vector are spliced to form a spatial structure representation vector; S23, splice the state parameter vector, the spatial structure representation vector, and the stability level identifier into a fusion input vector set, and input the fusion input vector set to a position-aware attention mechanism, and the position-aware attention mechanism generates attention weights by calculating the spatial correlation between the fusion input vectors; S24, weight and fuse the fusion input vector set based on the attention weights to generate a nested state embedding tensor of each electric energy meter twin.

4. The big data full life cycle state monitoring method of claim 3, wherein, The step S22 includes the following processes: S221、set the longitude of each electric energy meter twin as x, the latitude as y, the coding dimension as d, the coding serial number of each dimension as i, the disturbance parameter sequence corresponding to the i-th dimension as δ i , the frequency disturbance parameter as φ i , the phase disturbance parameter as ψ i , the tensor weighting coefficient corresponding to the stability level as ω s , calculate the longitude coding value e i and the latitude coding value l i The combination nested expression of is: wherein ω s is the weighting coefficient corresponding to the stability level of the electric energy meter, the high stability level is 1.0, the medium stability level is 1.3, and the low stability level is 1.7; S222、the longitude encoding values e1, e2,..., e d with the latitude encoding values l1, l2,..., l d are concatenated into a position encoding vector of length 2d.

5. The big data full life cycle state monitoring method of claim 4, wherein, The position-aware attention mechanism in the step S23 includes the following processes: based on the state parameter vector and the position encoding vector, perform normalization operations to form a state input sequence and a position input sequence, respectively, split the position input sequence according to the longitude and latitude dimensions to generate a sensitive guide sequence for controlling the attention distribution direction, construct a multi-factor guide matrix for measuring the degree of correlation based on the Euclidean distance of the spatial coordinates between the electric energy meter twins, the grouping difference of the stability level, and the change amplitude of the state parameter vector, and use the multi-factor guide matrix as the position guide weight in the attention mechanism, and in the attention calculation process, jointly adjust the weight distribution relationship between the state input sequences according to the sensitive guide sequence and the position guide weight, and dynamically weight the state input sequences before fusion according to the weight adjustment coefficients corresponding to the high stability, medium stability, and low stability levels.

6. The big data full life cycle state monitoring method of claim 5, wherein, The S4 specifically includes: S41, extract the state parameter vector of each electric energy meter twin in the state reasoning tensor field at each continuous sampling time step, and construct a time sequence state trajectory; S42, perform a time difference operation on each state trajectory to obtain the state change rate between each time step, and mark the time steps with a change rate greater than the fluctuation threshold corresponding to the stability level as abnormal fluctuation points; S43, for each electric energy meter twin, count the number of continuous abnormal fluctuation points within a fixed sampling time window, and combine the change rate value and the change direction consistency coefficient to accumulate a trend disturbance score; S44, perform time window integration processing on the trend disturbance score and the duration of the abnormal fluctuation point to form an abnormal trend cumulative score value; S45, compare the abnormal trend cumulative score value with the threshold value of the stability level classification, and if the score value exceeds the upper threshold value corresponding to the stability level to which the electric energy meter twin belongs, the electric energy meter twin is automatically marked as a state that needs to be fed back.

7. The big data full life cycle state monitoring method of claim 6, wherein, The S5 specifically includes: In the twin of the electric energy meter marked as the state requiring feedback, the state parameter vectors of the electric energy meter twin at three consecutive time steps in the state reasoning tensor field are extracted, the state evolution trend vector is formed according to the change direction of the state parameter vector, the state parameter vector density distribution of the adjacent electric energy meter twin within three orders in the spatial neighborhood of the electric energy meter twin is calculated, the density gradient change in each direction is counted, the density direction vector is constructed, and the consistency degree between the cosine value of the included angle of the state evolution trend vector and the density direction vector is used to measure the trend consistency degree, and the consistency degree, the average change amplitude and the density intensity are used as the joint features input, the rule mapping is performed, and the optimal feedback direction and the feedback action type of the electric energy meter twin in the current state are judged; The strategy for feedback regulation according to the stability level includes the following contents: when the stability level is high, the data reading period adjustment step is ten percent of the current period, the communication frequency adjustment factor is the basic frequency minus twenty seconds, and the state self-labeling type is state retention; when the stability level is medium, the data reading period adjustment step is thirty percent of the current period, the communication frequency adjustment factor is the basic frequency minus ten seconds, and the state self-labeling type is dynamic observation; when the stability level is low, the data reading period adjustment step is fifty percent of the current period, the communication frequency adjustment factor is not adjusted, and the state self-labeling type is high-frequency monitoring.

8. An electric energy meter, used for implementing the big data full life cycle state monitoring method of any one of claims 1 to 7, characterized in that, The electric energy meter comprises: a data reading module for reading original measurement data and constructing an electric energy meter state parameter vector to generate an electric energy meter twin set; a spatial fusion module for extracting state parameter vectors, spatial coordinates and layer attribute representation vectors, and performing fusion feature modeling; a position coding module for position-aware coding of spatial coordinates to generate a nested state embedding tensor; a state reasoning module for calculating state difference and spatial distance to output a state reasoning tensor field; an anomaly detection module for extracting continuous state change rate, calculating abnormal trend cumulative score value and marking the state requiring feedback; a feedback generation module for generating a feedback inducing action set according to the trend direction, change amplitude and density; a self-feedback execution module for performing sampling period adjustment, communication frequency adjustment and state self-labeling operation to generate a feedback execution record; a state updating module for updating the state parameter vector and the nested state embedding tensor of the electric energy meter twin to reconstruct the state reasoning tensor field.

Images