Data acquisition and transmission method for intelligent electric energy meter

Abstract

The invention relates to the technical field of data acquisition and transmission, and discloses a data acquisition and transmission method for an intelligent electric energy meter, and the method comprises the steps: carrying out the analysis of measurement data and operation state data, obtaining a measurement feature vector and an operation feature vector, inputting the measurement feature vector and the operation feature vector into a power utilization scene recognition model, and obtaining a power utilization scene recognition model; generating a real-time acquisition label, a periodic acquisition label and an event triggering acquisition label; inputting a data transmission decision model, and generating a transmission priority parameter, a data integrity weight and an abnormal data identifier; and calculating a comprehensive transmission load evaluation value, and adaptively configuring a data acquisition frequency, a data packaging mode and a transmission scheduling strategy of single communication between the intelligent electric energy meter and the data concentrator or the cloud management platform according to the comprehensive transmission load evaluation value. According to the invention, intelligent and refined management and control of data acquisition and transmission processes can be realized through deep cooperation of multi-dimensional power utilization scene identification and dynamic transmission decision.

Description

Technical Field

[0001] This invention relates to the field of data acquisition and transmission technology, and more specifically, to a data acquisition and transmission method for smart meters. Background Technology

[0002] Traditional smart meter data acquisition and transmission methods often employ fixed data acquisition frequency settings, making it difficult to dynamically adjust based on actual electricity consumption and data importance. This leads to transmission congestion during peak hours due to excessive data volume, while data redundancy and resource waste occur during off-peak hours. Furthermore, data packetization methods often lack differentiated processing for different data types (such as real-time electricity consumption, instantaneous voltage and current values, and fault alarm information), resulting in the mixing and synchronization of important and non-important data, potentially causing delays or loss of critical information. In addition, transmission scheduling strategies are relatively simplistic, failing to fully consider network load conditions, data priorities, and varying user needs for data timeliness. This makes it difficult to effectively guarantee the reliability and efficiency of data transmission, especially in complex power network environments, where issues such as unstable data transmission, high packet loss rates, and slow response times can easily arise, affecting the smart grid's accurate monitoring and efficient management of user electricity data.

[0003] Therefore, it is necessary to design a data acquisition and transmission method for smart meters to solve the problems existing in the current technology. Summary of the Invention

[0004] In view of this, the present invention proposes a data acquisition and transmission method for smart meters, aiming to solve the problems existing in the current technology of smart meter data acquisition and transmission, such as fixed acquisition frequency, lack of differentiated processing of data packets, and single transmission scheduling strategy.

[0005] This invention proposes a data acquisition and transmission method for smart meters, comprising the following steps: A smart sensor is installed inside the smart energy meter, and the smart sensor is used to collect the metering data and operating status data of the smart energy meter. The metering data and operating status data are parsed to obtain metering feature vectors and operating feature vectors. The metering feature vectors and operating feature vectors are then input into the electricity consumption scenario identification model to generate real-time acquisition tags, periodic acquisition tags, and event-triggered acquisition tags. The measurement feature vector, the operation feature vector, and the real-time acquisition tag are input into the data transmission decision model to generate transmission priority parameters, data integrity weights, and abnormal data identifiers. Based on the real-time acquisition tags, periodic acquisition tags, event-triggered acquisition tags, transmission priority parameters, and data integrity weights, a comprehensive transmission load assessment value is calculated. Based on the comprehensive transmission load assessment value, the data acquisition frequency, data packet method, and transmission scheduling strategy for a single communication are adaptively configured between the smart energy meter and the data concentrator or cloud management platform.

[0006] Furthermore, the electricity consumption scenario identification model is constructed based on a multi-feature fusion classification framework and trained through a joint optimization objective function. The joint optimization objective function is constructed by weighting the real-time acquisition labels based on the metering data fluctuation tolerance coefficient and by weighting the event-triggered acquisition labels based on the state change sensitivity coefficient.

[0007] Furthermore, the data transmission decision model is constructed based on a hybrid decision tree and a probabilistic statistical model.

[0008] Furthermore, when generating real-time collection tags, periodic collection tags, and event-triggered collection tags, the following are included: Extract the data fluctuation amplitude features, data fluctuation frequency features, and data change trend features from the metering feature vector, and combine them with the voltage stability index, current load rate, and temperature drift parameters from the operating status feature vector to construct a multi-dimensional power consumption scenario feature matrix; The multi-dimensional electricity consumption scenario feature matrix is ​​input into the feature preprocessing layer of the electricity consumption scenario identification model. After feature dimensionality reduction processing by principal component analysis, an attention mechanism module is introduced to assign dynamic weights to the feature dimensions corresponding to the metering data fluctuation tolerance coefficient and to enhance the weights of the feature dimensions corresponding to the state change sensitivity coefficient. Based on the feature vectors weighted by the attention mechanism, real-time collection labels, periodic collection labels, and event-triggered collection labels are generated through the classification output layer; the confidence thresholds corresponding to each collection label are optimized and determined by cross-validation based on historical electricity consumption scenario datasets.

[0009] Furthermore, when the confidence thresholds corresponding to each collected label are optimized and determined through cross-validation based on historical electricity consumption scenario datasets, the following are included: Construct a historical electricity consumption scenario dataset containing various different types of electricity consumption scenarios, and label each sample in the historical electricity consumption scenario dataset with the corresponding real collection label; The historical electricity consumption scenario dataset is divided into a training subset and a validation subset using cross-validation. In the training subset, multiple candidate confidence threshold combinations are traversed, and in the validation subset, the collection label recognition performance index corresponding to each confidence threshold combination is calculated. The optimal confidence threshold for each collection tag is determined by taking the overall optimal performance index of the collection tag as the optimization objective. Specifically, for the real-time collection tag, a timeliness penalty factor related to the timeliness of data collection is introduced during the threshold optimization process. When a real-time collection tag is misjudged, causing a delay in data collection, the recognition performance index is corrected.

[0010] Furthermore, a timeliness penalty factor related to data acquisition timeliness is introduced during the threshold optimization process. When real-time tag misjudgment causes data acquisition delay, the correction of the recognition performance index includes: During the cross-validation process, the timeliness of the recognition results corresponding to the real-time acquisition tags is evaluated to determine the time delay value between the occurrence of the scene and the completion of data acquisition for the real-time acquisition tags. The time delay value is compared with the real-time acquisition timeliness threshold. When the time delay value exceeds the real-time acquisition timeliness threshold, it is determined to be a real-time acquisition delay event. To address the real-time acquisition delay event, a timeliness penalty factor is introduced to attenuate and correct the performance indicators of the acquisition tag recognition corresponding to the real-time acquisition tag.

[0011] Furthermore, the formula for calculating the timeliness penalty factor is as follows: τ=α×(Δt / Δt0) 2 ; Where τ is the timeliness penalty factor; α is the penalty coefficient, with a value range of 0.5-1; Δt is the actual time delay value; Δt0 is the real-time acquisition timeliness threshold; when Δt≤Δt0, τ=0, and no penalty is imposed.

[0012] Furthermore, when inputting the measurement feature vector, the operational feature vector, and the real-time acquisition tag into the data transmission decision model to generate transmission priority parameters, data integrity weights, and abnormal data identifiers, the following steps are included: The active power deviation feature, reactive power distortion feature and three-phase imbalance feature in the metering feature vector are combined with the communication module signal strength feature, power supply remaining energy feature and historical transmission success rate feature in the operation feature vector to form a transmission decision input matrix. The hybrid decision tree processing layer in the data transmission decision model performs multi-path rule matching on the transmission decision input matrix, outputting preliminary transmission priority parameters and data integrity weight ranges; wherein, the split nodes of the hybrid decision tree processing layer are selected based on the combined evaluation results of information gain ratio and Gini index; Based on the confidence value of the real-time collected tags, the data integrity weight interval is dynamically adjusted through a probabilistic statistical model. When the confidence value of the real-time collected tags is higher than the preset confidence threshold, the lower limit constraint of the data integrity weight interval is increased. The posterior probability of abnormal data occurrence is calculated based on the probabilistic statistical model. When the posterior probability exceeds a preset anomaly judgment threshold, an anomaly data identifier is generated. The anomaly data identifier includes anomaly data type identifier information and corresponding anomaly confidence information.

[0013] Furthermore, when calculating the comprehensive transmission load assessment value based on the real-time acquisition tags, periodic acquisition tags, event-triggered acquisition tags, transmission priority parameters, and data integrity weights, the following steps are included: Perform matrix operations on the scene weight coefficients and transmission priority parameters corresponding to each collection tag to obtain the basic load evaluation vector; The data integrity weight is combined with the amount of data in a single communication to generate a data integrity load component, and then adjusted according to the compression processing efficiency of the data type. The basic load assessment vector is weighted and fused with the data integrity load component, and additional load increments related to anomalies are superimposed when there are anomaly data identifiers. Based on the weighted fusion and abnormal load results, a comprehensive transmission load assessment value is generated to guide communication scheduling.

[0014] Furthermore, based on the comprehensive transmission load assessment value, when adaptively configuring the data acquisition frequency, data packet method, and transmission scheduling strategy for a single communication between the smart energy meter and the data concentrator or cloud management platform, the following is included: The comprehensive transmission load assessment value is compared with the first comprehensive transmission load assessment value and the second comprehensive transmission load assessment value. Based on the comparison result, the data acquisition frequency, data packetization method and transmission scheduling strategy are determined; wherein, the first comprehensive transmission load assessment value is less than the second comprehensive transmission load assessment value. When the comprehensive transmission load assessment value is less than or equal to the first comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the first data acquisition frequency, the first data packetization method, and the priority transmission scheduling strategy, respectively. When the comprehensive transmission load assessment value is greater than the first comprehensive transmission load assessment value and less than or equal to the second comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the second data acquisition frequency, the second data packetization method, and the balanced transmission scheduling strategy, respectively. When the comprehensive transmission load assessment value is greater than the second comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the third data acquisition frequency, the third data packetization method, and the rate-limiting transmission scheduling strategy, respectively.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: The data acquisition and transmission method for smart meters provided by this invention can achieve intelligent and refined management of the data acquisition and transmission process through deep collaboration between multi-dimensional electricity consumption scenario identification and dynamic transmission decision-making. Specifically, it links the analysis of electricity consumption scenario characteristics and data transmission needs, and uses multiple types of acquisition tags generated by the electricity consumption scenario identification model to accurately match the data acquisition needs of different electricity consumption scenarios. For example, in scenarios sensitive to fluctuations in metering data, such as peak electricity consumption periods, real-time acquisition tags ensure the timeliness of key data; during the normal operation phase of the equipment, periodic acquisition tags efficiently and with low power consumption collect data; when the equipment malfunctions, event-triggered acquisition tags respond quickly and collect data in a targeted manner, improving the scenario adaptability and timeliness of data acquisition. At the same time, the data transmission decision model generates transmission priority parameters, data integrity weights, and abnormal data identifiers based on multi-dimensional features, and combines them with comprehensive transmission load assessment values ​​to achieve adaptive communication configuration. Under the premise of ensuring data transmission quality, it dynamically optimizes the data acquisition frequency, packet method, and transmission scheduling strategy, reducing unnecessary communication resource consumption and avoiding network congestion or data redundancy problems of traditional fixed acquisition and transmission modes. Especially with the large-scale increase in the number of smart meters, it can improve the stability and operating efficiency of the electricity data acquisition and transmission system, and provide reliable data support for the refined management of the power system, fault early warning and energy optimization. Attached Figure Description

[0016] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart of a data acquisition and transmission method for a smart energy meter provided in an embodiment of the present invention. Detailed Implementation

[0017] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] See Figure 1 As shown in some embodiments of this application, this embodiment provides a data acquisition and transmission method for smart meters, including the following steps: S100: An intelligent sensor is installed inside the smart energy meter, and the intelligent sensor is used to collect the metering data and operating status data of the smart energy meter; S200: The metering data and operating status data are parsed to obtain metering feature vectors and operating feature vectors, and the metering feature vectors and operating feature vectors are input into the electricity consumption scenario identification model to generate real-time acquisition tags, periodic acquisition tags and event-triggered acquisition tags. S300: Input the metering feature vector, the running feature vector and the real-time acquisition tag into the data transmission decision model to generate transmission priority parameters, data integrity weights and abnormal data identifiers; S400: Based on the real-time acquisition tag, periodic acquisition tag, event-triggered acquisition tag, transmission priority parameter and data integrity weight, calculate the comprehensive transmission load assessment value, and based on the comprehensive transmission load assessment value, adaptively configure the data acquisition frequency, data packet method and transmission scheduling strategy for a single communication between the smart energy meter and the data concentrator or cloud management platform.

[0019] It is understood that the data acquisition and transmission method for smart meters provided in this embodiment can achieve intelligent and refined management of the data acquisition and transmission process through deep collaboration between multi-dimensional electricity consumption scenario identification and dynamic transmission decision-making. Specifically, it links the analysis of electricity consumption scenario characteristics and data transmission needs, and uses multiple types of acquisition tags generated by the electricity consumption scenario identification model to accurately match the data acquisition needs of different electricity consumption scenarios. For example, in scenarios sensitive to fluctuations in metering data, such as peak electricity consumption periods, real-time acquisition tags are used to ensure the timeliness of key data; during the normal operation phase of the equipment, periodic acquisition tags are used to collect data efficiently and with low power consumption; when the equipment malfunctions, event-triggered acquisition tags respond quickly and collect data in a targeted manner, improving the scenario adaptability and timeliness of data acquisition. At the same time, the data transmission decision model generates transmission priority parameters, data integrity weights, and abnormal data identifiers based on multi-dimensional features, and combines them with comprehensive transmission load assessment values ​​to achieve adaptive communication configuration. Under the premise of ensuring data transmission quality, it dynamically optimizes the data acquisition frequency, packet method, and transmission scheduling strategy, reducing unnecessary communication resource consumption and avoiding network congestion or data redundancy problems in traditional fixed acquisition and transmission modes. Especially with the large-scale increase in the number of smart meters, it can improve the stability and operating efficiency of the electricity data acquisition and transmission system, and provide reliable data support for the refined management of the power system, fault early warning and energy optimization.

[0020] Specifically, the electricity consumption scenario identification model is constructed based on a multi-feature fusion classification framework and trained through a joint optimization objective function. The joint optimization objective function is constructed by weighting the real-time acquisition labels based on the metering data fluctuation tolerance coefficient and by weighting the event-triggered acquisition labels based on the state change sensitivity coefficient.

[0021] Understandably, the metering data fluctuation tolerance coefficient is a quantitative indicator of the power management system's tolerance for short-term fluctuations in metering data under specific electricity consumption scenarios. For example, during off-peak periods for residential electricity consumption, user electricity consumption is stable, metering data fluctuations are small, the system's tolerance is high, the coefficient is large, the weight of real-time acquisition tags is reduced, and the collection frequency can be decreased. During peak commercial electricity consumption or critical processes in industrial production, even small fluctuations in metering data may reflect significant changes, the system's tolerance is low, the coefficient is small, the weight of real-time acquisition tags is increased, ensuring the capture of critical changes and meeting high-precision monitoring requirements. The state change sensitivity coefficient is an indicator that measures the system's sensitivity to changes in the operating state parameters of smart meters in an electricity consumption scenario. For stable, healthy smart meters, the coefficient is low, the weight of event-triggered acquisition tags is small, and collection is only triggered when state parameters exceed a relatively large threshold. For older or previously faulty smart meters, the coefficient is high, the weight of event-triggered acquisition tags is large, and collection is triggered even with minor anomalies in state parameters, in order to detect potential faults. By incorporating these two coefficients into the joint optimization objective function, the electricity scenario identification model can dynamically adjust the real-time and event-triggered tag generation logic according to the scenario characteristics, making the data acquisition strategy adaptable to the scenario requirements.

[0022] Specifically, the data transmission decision model is constructed based on a hybrid decision tree and a probabilistic statistical model.

[0023] Understandably, the fusion architecture of hybrid decision trees and probabilistic statistical models can balance the nonlinear mapping of data features with the ability to quantify uncertainty. The hybrid decision tree module first performs hierarchical feature splitting on the input metering, operational feature vectors, and real-time acquisition labels. Key decision nodes are extracted through a multi-branch tree structure, such as using highly discriminative features like "voltage fluctuation amplitude > 5%" as high-level split points to locate the basic priority boundary for data transmission. The probabilistic statistical model module constructs conditional probability distributions between feature variables based on Bayesian networks. For example, it learns the probabilistic correlation between "power factor below 0.9 in metering feature vectors" and "data integrity weight needs to be increased to 0.85," and combines this with Gaussian process regression to predict continuous values ​​of transmission priority parameters, achieving accurate mapping from discrete features to continuous decision parameters. This collaborative mechanism of "tree-structure logical reasoning + probabilistic uncertainty modeling" allows the data transmission decision model to both use decision trees to identify key influencing factors and use probabilistic statistical models to handle random interference in power data transmission, improving the robustness of transmission priority parameters, data integrity weights, and abnormal data identifier generation. For example, when a smart energy meter detects a "current transformer secondary side open circuit" fault, the hybrid decision tree sets the transmission priority parameter to the highest level, and the probabilistic statistical model calculates the probability of data transmission failure. If the probability exceeds the preset threshold, the data integrity weight is automatically increased to ensure that abnormal data transmission has a higher error correction redundancy capability.

[0024] Specifically, the generation of real-time collection tags, periodic collection tags, and event-triggered collection tags includes: Extract the data fluctuation amplitude features, data fluctuation frequency features, and data change trend features from the metering feature vector, and combine them with the voltage stability index, current load rate, and temperature drift parameters from the operating status feature vector to construct a multi-dimensional power consumption scenario feature matrix; The multi-dimensional electricity consumption scenario feature matrix is ​​input into the feature preprocessing layer of the electricity consumption scenario identification model. After feature dimensionality reduction processing by principal component analysis, an attention mechanism module is introduced to assign dynamic weights to the feature dimensions corresponding to the metering data fluctuation tolerance coefficient and to enhance the weights of the feature dimensions corresponding to the state change sensitivity coefficient. Based on the feature vectors weighted by the attention mechanism, real-time collection labels, periodic collection labels, and event-triggered collection labels are generated through the classification output layer; the confidence thresholds corresponding to each collection label are optimized and determined by cross-validation based on historical electricity consumption scenario datasets.

[0025] Understandably, when smart meters in an industrial park operate during peak electricity consumption periods (e.g., 9:00-11:00 AM), the data fluctuation amplitude characteristics in the metering feature vector may manifest as active power fluctuations exceeding ±15% within a short period (e.g., within 5 minutes), data fluctuation frequency characteristics showing up to 8 peak fluctuations per hour, and data change trend characteristics exhibiting a sawtooth curve of rapid rise followed by slow decline. Simultaneously, the voltage stability index in the operating status feature vector detects occasional voltage deviations reaching -3%, current load rates consistently maintained within the 85%-90% range, and temperature drift parameters showing an internal chip temperature 4°C higher than the baseline value. These characteristics collectively constitute a multi-dimensional electricity consumption scenario feature matrix. This matrix first undergoes principal component analysis in the feature preprocessing layer, reducing the original 12-dimensional features to 5 principal components while retaining 92% of the information entropy. Subsequently, the attention mechanism module is activated: Since the current scenario is peak industrial electricity consumption, the tolerance coefficient for metering data fluctuations is relatively small (e.g., 0.3). The system automatically increases the weights of the data fluctuation amplitude and frequency features from the base value of 0.5 to 0.7 and 0.65, respectively. Simultaneously, considering that the equipment is operating at full load during this period, the sensitivity coefficient for state changes is relatively high (e.g., 0.8), and the weights of the temperature drift parameter and current load rate are increased from 0.4 to 0.75. The weighted feature vector is input into the classification output layer. The model calculates a confidence level of 0.89 for real-time acquisition tags (above the preset threshold of 0.8), 0.52 for periodic acquisition tags (below the threshold of 0.6), and 0.31 for event-triggered acquisition tags (below the threshold of 0.5). This ultimately generates real-time acquisition tags, ensuring that key metering data is collected every 2 seconds to capture power surges caused by rapid start-up and shutdown of industrial equipment.

[0026] Specifically, when the confidence thresholds for each collected label are optimized and determined using cross-validation based on historical electricity consumption scenario datasets, the following criteria are included: Construct a historical electricity consumption scenario dataset containing various different types of electricity consumption scenarios, and label each sample in the historical electricity consumption scenario dataset with the corresponding real collection label; The historical electricity consumption scenario dataset is divided into a training subset and a validation subset using cross-validation. In the training subset, multiple candidate confidence threshold combinations are traversed, and in the validation subset, the collection label recognition performance index corresponding to each confidence threshold combination is calculated. The optimal confidence threshold for each collection tag is determined by taking the overall optimal performance index of the collection tag as the optimization objective. Specifically, for the real-time collection tag, a timeliness penalty factor related to the timeliness of data collection is introduced during the threshold optimization process. When a real-time collection tag is misjudged, causing a delay in data collection, the recognition performance index is corrected.

[0027] Understandably, the constructed historical electricity consumption scenario dataset covers eight typical scenarios, including residential electricity consumption (such as off-peak hours on weekdays and peak hours on holidays), commercial complex electricity consumption (such as during business hours and after closing), and industrial production electricity consumption (such as production line start-up and shutdown, and equipment maintenance), containing a total of 12,000 sample data. Each sample is labeled with a manually verified, genuine collection label (such as "real-time collection," "15-minute cycle collection," and "overvoltage event triggered collection"). In the cross-validation phase, a 5-fold cross-validation method is used to divide the dataset into five subsets, selecting four subsets as training subsets and one subset as validation subset each time. For the real-time collection label, cycle collection label, and event triggered collection label, five candidate thresholds are preset (e.g., the candidate threshold range for the real-time collection label is 0.7-0.9, with a step size of 0.05), forming 125 candidate confidence threshold combinations. In the training subset, training is performed iteratively for each combination. For example, when the real-time acquisition label threshold is set to 0.8, the periodic acquisition label threshold is set to 0.6, and the event-triggered acquisition label threshold is set to 0.5, the overall recognition accuracy is calculated to be 89.3%, the precision is 87.6%, and the recall is 88.2% in the validation subset. In particular, for the threshold optimization of real-time acquisition labels, a timeliness penalty factor is introduced: when the model misclassifies a sample that should have generated a real-time acquisition label as a periodic acquisition label, causing the data acquisition delay to exceed the preset 5-second threshold, the penalty coefficient (8 / 5=1.6) is calculated based on the delay duration (e.g., 8 seconds) to correct the recognition performance index (e.g., F1 value) of the sample (original F1 value 0.85×(1-1.6×0.1)=0.724). By evaluating all candidate combinations as described above, the optimal threshold combination was finally selected: real-time collection labels 0.82 (at which point the F1 score after timeliness penalty for real-time label recognition on the validation set reaches as high as 0.89), periodic collection labels 0.58 (periodic scene recognition accuracy of 91%), and event-triggered collection labels 0.45 (event scene recall rate of 92%), ensuring that the electricity consumption scene recognition model can generate accurate collection labels in different scenarios.

[0028] Specifically, during the threshold optimization process, a timeliness penalty factor related to the timeliness of data collection is introduced. When real-time label misjudgment causes data collection delay, the correction of the recognition performance index includes: During the cross-validation process, the timeliness of the recognition results corresponding to the real-time acquisition tags is evaluated to determine the time delay value between the occurrence of the scene and the completion of data acquisition for the real-time acquisition tags. The time delay value is compared with the real-time acquisition timeliness threshold. When the time delay value exceeds the real-time acquisition timeliness threshold, it is determined to be a real-time acquisition delay event. To address the real-time acquisition delay event, a timeliness penalty factor is introduced to attenuate and correct the performance indicators of the acquisition tag recognition corresponding to the real-time acquisition tag.

[0029] Understandably, in a cross-validation experiment targeting the electricity consumption scenario of a commercial complex, the historical dataset contained electricity consumption samples from a shopping mall on weekdays from 10:00 to 12:00. During this period, the air conditioning system frequently started and stopped, resulting in significant fluctuations in active power. Real-time collection labels should have been generated to capture these instantaneous changes. Assuming the real-time collection timeliness threshold is set to 3 seconds, representing the longest allowable delay from the appearance of scene features to the completion of data collection, if a sample in the validation subset has a true real-time collection label, but the model misclassifies it as a periodic collection label (with a period of 15 seconds) due to an improperly set confidence threshold (e.g., the real-time collection label threshold is temporarily set to 0.85), then the time delay from the appearance of scene features to the actual start of collection is 15 seconds. In this case, the 15-second delay far exceeds the real-time collection timeliness threshold of 3 seconds, and is judged as a real-time collection delay event. Therefore, a timeliness penalty factor is introduced, which is positively correlated with the delay duration. The calculation formula is set as penalty factor = 1 + (delay duration - timeliness threshold) / timeliness threshold, i.e., 1 + (15 - 3) / 3 = 5. If the original F1 value of the misjudged sample was 0.88, then the corrected F1 value = original F1 value / (1 + penalty factor) = 0.88 / (1 + 5) ≈ 0.147. This significant performance degradation forces the threshold optimization algorithm to adjust the real-time collection tag threshold in subsequent iterations, for example, by reducing it to 0.82, in order to reduce the timeliness loss caused by such excessively high thresholds and ensure accurate response to timeliness-sensitive scenarios such as peak electricity consumption periods in commercial complexes.

[0030] Specifically, the formula for calculating the timeliness penalty factor is as follows: τ=α×(Δt / Δt0) 2 ; Where τ is the timeliness penalty factor; α is the penalty coefficient, with a value range of 0.5-1; Δt is the actual time delay value; Δt0 is the real-time acquisition timeliness threshold; when Δt≤Δt0, τ=0, and no penalty is imposed.

[0031] It is understandable that if the penalty coefficient α=0.8 and the real-time collection timeliness threshold Δt0=5 seconds are set, when a real-time collection tag is misjudged, resulting in an actual delay of Δt=10 seconds, Δt / Δt0=2, and at this time the timeliness penalty factor τ=0.8×(2) 2=3.2. Assuming the original F1 score for the misclassified sample is 0.9, after introducing the penalty factor, the corrected F1 score can be calculated by multiplying the original F1 score by (1-τ×β) (where β is the penalty weight, with a value of 0.1), i.e., 0.9×(1-3.2×0.1)=0.9×0.68=0.612, which is significantly lower than before correction, thus strengthening the model's focus on the timeliness of real-time label collection. When the actual delay Δt=3 seconds (less than Δt0=5 seconds), τ=0, and no F1 score is corrected, ensuring that misclassifications within the acceptable delay range do not excessively affect performance evaluation, balancing the accuracy and timeliness requirements of data collection.

[0032] Specifically, when inputting the measurement feature vector, the operational feature vector, and the real-time acquisition tag into the data transmission decision model to generate transmission priority parameters, data integrity weights, and abnormal data identifiers, the following steps are included: The active power deviation feature, reactive power distortion feature and three-phase imbalance feature in the metering feature vector are combined with the communication module signal strength feature, power supply remaining energy feature and historical transmission success rate feature in the operation feature vector to form a transmission decision input matrix. The hybrid decision tree processing layer in the data transmission decision model performs multi-path rule matching on the transmission decision input matrix, outputting preliminary transmission priority parameters and data integrity weight ranges; wherein, the split nodes of the hybrid decision tree processing layer are selected based on the combined evaluation results of information gain ratio and Gini index; Based on the confidence value of the real-time collected tags, the data integrity weight interval is dynamically adjusted through a probabilistic statistical model. When the confidence value of the real-time collected tags is higher than the preset confidence threshold, the lower limit constraint of the data integrity weight interval is increased. The posterior probability of abnormal data occurrence is calculated based on the probabilistic statistical model. When the posterior probability exceeds a preset anomaly judgment threshold, an anomaly data identifier is generated. The anomaly data identifier includes anomaly data type identifier information and corresponding anomaly confidence information.

[0033] Understandably, during peak summer electricity consumption (such as 7:00 PM to 9:00 PM), the smart meters in a residential community may exhibit an active power deviation of ±8% (due to concentrated air conditioner startup), a reactive power distortion characteristic showing a total harmonic distortion rate of 5%, and a three-phase imbalance characteristic showing that phase A current is 12% higher than phase C current. Simultaneously, the communication module signal strength characteristic in the operation characteristic vector detects a current signal strength of -75dBm (at a moderate level), the remaining power energy characteristic shows a battery charge of 82%, and the historical transmission success rate averages 96.5% over the past 7 days. These features are concatenated to form an 18-dimensional transmission decision input matrix. After the matrix is ​​input into the hybrid decision tree processing layer, the system first selects "whether the active power deviation exceeds ±5%" as the first split node based on a combined evaluation of information gain ratio (for continuous features such as signal strength) and Gini index (for discrete features such as power level). Subsequently, it matches rules such as "whether the signal strength is ≥-80dBm" and "whether the historical transmission success rate is ≥95%" in each branch path, and finally outputs the initial transmission priority parameter as "high" (corresponding to a weight of 0.8), with a data integrity weight range of [0.7, 0.9]. At this time, since the confidence level of the real-time acquisition tag is 0.89 (higher than the preset confidence threshold of 0.8), the probabilistic statistical model initiates a dynamic adjustment mechanism, increasing the lower limit of the data integrity weight range from 0.7 to 0.75 to ensure the integrity requirements of key real-time data. Simultaneously, the model calculates the posterior probability of abnormal data occurrence: combining the fluctuation amplitude of active power deviation characteristics and the prior probability of historical abnormal data (e.g., the probability of abnormality is 0.3 when the harmonic distortion rate exceeds 5%), the current posterior probability is calculated using Bayes' formula to be 0.28 (lower than the preset abnormality judgment threshold of 0.4), therefore no abnormal data label is generated. If, at this time, an industrial user's smart meter experiences a sudden increase in active power deviation to ±15% and a total harmonic distortion rate reaching 8%, the calculated posterior probability may reach 0.65, and the system will immediately generate an abnormal data label, where the abnormal data type label is "sudden power change + harmonic exceedance", and the abnormality confidence information is 0.65, so that the background system can prioritize processing this abnormal data.

[0034] Specifically, when calculating the comprehensive transmission load assessment value based on the real-time acquisition tags, periodic acquisition tags, event-triggered acquisition tags, transmission priority parameters, and data integrity weights, the following are included: Perform matrix operations on the scene weight coefficients and transmission priority parameters corresponding to each collection tag to obtain the basic load evaluation vector; The data integrity weight is combined with the amount of data in a single communication to generate a data integrity load component, and then adjusted according to the compression processing efficiency of the data type. The basic load assessment vector is weighted and fused with the data integrity load component, and additional load increments related to anomalies are superimposed when there are anomaly data identifiers. Based on the weighted fusion and abnormal load results, a comprehensive transmission load assessment value is generated to guide communication scheduling.

[0035] Understandably, when a smart meter cluster in an industrial park conducts a comprehensive transmission load assessment, it is assumed to include three types of data collection tag scenarios: real-time data collection tags (corresponding to power consumption of key equipment in the industrial production line, scenario weight coefficient 0.9), periodic data collection tags (general lighting and office equipment, scenario weight coefficient 0.5), and event-triggered data collection tags (such as leakage protection activation events, scenario weight coefficient 0.7). The transmission priority parameters corresponding to each scenario are "High" (weight 0.85), "Medium" (weight 0.6), and "Emergency" (weight 0.95), respectively. First, matrix operations are performed to obtain the basic load assessment vector: the contribution value of the real-time data collection scenario is 0.9 × 0.85 = 0.765, the contribution value of the periodic data collection scenario is 0.5 × 0.6 = 0.3, and the contribution value of the event-triggered data collection scenario is 0.7 × 0.95 = 0.665. The sum of the basic load assessment vectors is 0.765 + 0.3 + 0.665 = 1.73. Next, the data integrity load components are calculated: real-time acquired data has an integrity weight of 0.88 (after dynamic adjustment), a single communication data volume of 120KB, and a compression efficiency of 85% (i.e., the compressed data volume is 120×0.85=102KB), so this component is 0.88×102=90.76; periodic acquired data has an integrity weight of 0.6, a data volume of 80KB, and a compression efficiency of 90%, so the component is 0.6×(80×0.9)=43.2; event-triggered data has an integrity weight of 0.92, a data volume of 150KB, and a compression efficiency of 75%, so the component is 0.92×(150×0.75)=103.5. The total data integrity load components are 90.76+43.2+103.5=237.46. Then, the basic load assessment vector (1.73) and the data integrity load component (237.46) are weighted and fused, with the basic load weight set to 0.1 and the integrity load weight to 0.9. The fusion result is 1.73×0.1+237.46×0.9≈0.173+213.714=213.887. If a certain electricity meter triggers an abnormal data flag (abnormal confidence level 0.72), the additional load increment corresponding to the abnormal level is 35 (the preset abnormal confidence level 0.7-0.8 corresponds to an increment of 30-40). The final comprehensive transmission load assessment value is 213.887+35=248.887. When this assessment value exceeds the maximum load threshold of the communication module (e.g., 200), the system will activate a dynamic scheduling strategy, such as temporarily reducing the integrity weight of periodically collected data to 0.5, or prioritizing the transmission of event-triggered and abnormal data to avoid communication congestion.

[0036] Specifically, based on the comprehensive transmission load assessment value, when adaptively configuring the data acquisition frequency, data packet method, and transmission scheduling strategy for a single communication between the smart energy meter and the data concentrator or cloud management platform, the following are included: The comprehensive transmission load assessment value is compared with the first comprehensive transmission load assessment value and the second comprehensive transmission load assessment value. Based on the comparison result, the data acquisition frequency, data packetization method and transmission scheduling strategy are determined; wherein, the first comprehensive transmission load assessment value is less than the second comprehensive transmission load assessment value. When the comprehensive transmission load assessment value is less than or equal to the first comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the first data acquisition frequency, the first data packetization method, and the priority transmission scheduling strategy, respectively. When the comprehensive transmission load assessment value is greater than the first comprehensive transmission load assessment value and less than or equal to the second comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the second data acquisition frequency, the second data packetization method, and the balanced transmission scheduling strategy, respectively. When the comprehensive transmission load assessment value is greater than the second comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the third data acquisition frequency, the third data packetization method, and the rate-limiting transmission scheduling strategy, respectively.

[0037] Understandably, the first data acquisition frequency, the first data encapsulation method, and the priority transmission scheduling strategy are as follows: high-frequency acquisition at 1 minute / time, encapsulated in standard JSON format (including complete fields and verification codes), and priority transmission of real-time acquisition tags and event-triggered acquisition tags (transmission queue priority weight allocation is 0.4 for real-time data, 0.3 for event data, 0.2 for periodic data, and 0.1 for other data); the second data acquisition frequency, the second data encapsulation method, and the balanced transmission scheduling strategy are as follows: medium-frequency acquisition at 5 minutes / time, encapsulated in compressed binary format (only retaining key fields, compression rate of approximately 30%), and transmission bandwidth allocated according to the ratio of "real-time data: periodic data: event data = 3:4:3"; the third data acquisition frequency, the third data encapsulation method, and the rate-limited transmission scheduling strategy are as follows: low-frequency acquisition at 15 minutes / time, encapsulated in a simplified protocol (only transmitting feature values ​​and anomaly identifiers), and paused periodic acquisition data transmission (only retaining real-time and event data, and transmitting only the abnormal fluctuation part of real-time data). For example, when the overall transmission load assessment value of a commercial complex is 180 (the first strategy is activated when it is below the first threshold of 150, which is assumed to be 150 and the second threshold to be 220), the system collects all data at a 1-minute interval and encapsulates it completely, prioritizing the real-time power consumption data (such as power fluctuations) of the mall's air conditioning system; if the assessment value rises to 210 (in the 150-220 range), the system switches to a 5-minute collection interval, compresses and encapsulates the periodic data of non-critical lighting equipment, and balances the transmission needs of various types of data; when the assessment value reaches 250 (above 220), the system collects data only once every 15 minutes, and only transmits abnormal data of critical loads such as elevators and fire-fighting equipment to ensure that the communication link is not congested.

[0038] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0039] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0040] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0041] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0042] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A data acquisition and transmission method for smart energy meters, characterized in that, include: A smart sensor is installed inside the smart energy meter, and the smart sensor is used to collect the metering data and operating status data of the smart energy meter. The metering data and operating status data are parsed to obtain metering feature vectors and operating feature vectors. The metering feature vectors and operating feature vectors are then input into the electricity consumption scenario identification model to generate real-time acquisition tags, periodic acquisition tags, and event-triggered acquisition tags. The measurement feature vector, the operation feature vector, and the real-time acquisition tag are input into the data transmission decision model to generate transmission priority parameters, data integrity weights, and abnormal data identifiers. Based on the real-time acquisition tags, periodic acquisition tags, event-triggered acquisition tags, transmission priority parameters, and data integrity weights, a comprehensive transmission load assessment value is calculated. Based on the comprehensive transmission load assessment value, the data acquisition frequency, data packet method, and transmission scheduling strategy for a single communication are adaptively configured between the smart energy meter and the data concentrator or cloud management platform.

2. The data acquisition and transmission method for smart energy meters according to claim 1, characterized in that, The electricity consumption scenario identification model is constructed based on a multi-feature fusion classification framework and trained through a joint optimization objective function. The joint optimization objective function is constructed by weighting the real-time acquisition labels based on the metering data fluctuation tolerance coefficient and by weighting the event-triggered acquisition labels based on the state change sensitivity coefficient.

3. The data acquisition and transmission method for smart energy meters according to claim 2, characterized in that, The data transmission decision model is constructed based on a hybrid decision tree and a probabilistic statistical model.

4. The data acquisition and transmission method for smart energy meters according to claim 3, characterized in that, When generating real-time acquisition tags, periodic acquisition tags, and event-triggered acquisition tags, the following are included: Extract the data fluctuation amplitude features, data fluctuation frequency features, and data change trend features from the metering feature vector, and combine them with the voltage stability index, current load rate, and temperature drift parameters from the operating status feature vector to construct a multi-dimensional power consumption scenario feature matrix; The multi-dimensional electricity consumption scenario feature matrix is ​​input into the feature preprocessing layer of the electricity consumption scenario identification model. After feature dimensionality reduction processing by principal component analysis, an attention mechanism module is introduced to assign dynamic weights to the feature dimensions corresponding to the metering data fluctuation tolerance coefficient and to enhance the weights of the feature dimensions corresponding to the state change sensitivity coefficient. Based on the feature vectors weighted by the attention mechanism, real-time collection labels, periodic collection labels, and event-triggered collection labels are generated through the classification output layer; the confidence thresholds corresponding to each collection label are optimized and determined by cross-validation based on historical electricity consumption scenario datasets.

5. The data acquisition and transmission method for smart energy meters according to claim 4, characterized in that, When the confidence thresholds for each collected label are optimized and determined using cross-validation based on historical electricity consumption scenario datasets, the following criteria are included: Construct a historical electricity consumption scenario dataset containing various different types of electricity consumption scenarios, and label each sample in the historical electricity consumption scenario dataset with the corresponding real collection label; The historical electricity consumption scenario dataset is divided into a training subset and a validation subset using cross-validation. In the training subset, multiple candidate confidence threshold combinations are traversed, and in the validation subset, the collection label recognition performance index corresponding to each confidence threshold combination is calculated. The optimal confidence threshold for each collection tag is determined by taking the overall optimal performance index of the collection tag as the optimization objective. Specifically, for the real-time collection tag, a timeliness penalty factor related to the timeliness of data collection is introduced during the threshold optimization process. When a real-time collection tag is misjudged, causing a delay in data collection, the recognition performance index is corrected.

6. The data acquisition and transmission method for smart energy meters according to claim 5, characterized in that, In the threshold optimization process, a timeliness penalty factor related to the timeliness of data acquisition is introduced. When real-time tag misjudgment causes data acquisition delay, the correction of the recognition performance index includes: During the cross-validation process, the timeliness of the recognition results corresponding to the real-time acquisition tags is evaluated to determine the time delay value between the occurrence of the scene and the completion of data acquisition for the real-time acquisition tags. The time delay value is compared with the real-time acquisition timeliness threshold. When the time delay value exceeds the real-time acquisition timeliness threshold, it is determined to be a real-time acquisition delay event. To address the real-time acquisition delay event, a timeliness penalty factor is introduced to attenuate and correct the performance indicators of the acquisition tag recognition corresponding to the real-time acquisition tag.

7. The data acquisition and transmission method for smart energy meters according to claim 6, characterized in that, The formula for calculating the timeliness penalty factor is as follows: τ=α×(Δt / Δt0) 2 ; Where τ is the timeliness penalty factor; α is the penalty coefficient, with a value range of 0.5-1; Δt is the actual time delay value; Δt0 is the real-time acquisition timeliness threshold; when Δt≤Δt0, τ=0, and no penalty is imposed.

8. The data acquisition and transmission method for smart energy meters according to claim 7, characterized in that, When inputting the measurement feature vector, the operational feature vector, and the real-time acquisition tag into the data transmission decision model to generate transmission priority parameters, data integrity weights, and abnormal data identifiers, the following steps are included: The active power deviation feature, reactive power distortion feature and three-phase imbalance feature in the metering feature vector are combined with the communication module signal strength feature, power supply remaining energy feature and historical transmission success rate feature in the operation feature vector to form a transmission decision input matrix. The hybrid decision tree processing layer in the data transmission decision model performs multi-path rule matching on the transmission decision input matrix, outputting preliminary transmission priority parameters and data integrity weight ranges; wherein, the split nodes of the hybrid decision tree processing layer are selected based on the combined evaluation results of information gain ratio and Gini index; Based on the confidence value of the real-time collected tags, the data integrity weight interval is dynamically adjusted through a probabilistic statistical model. When the confidence value of the real-time collected tags is higher than the preset confidence threshold, the lower limit constraint of the data integrity weight interval is increased. The posterior probability of abnormal data occurrence is calculated based on the probabilistic statistical model. When the posterior probability exceeds a preset anomaly judgment threshold, an anomaly data identifier is generated. The anomaly data identifier includes anomaly data type identifier information and corresponding anomaly confidence information.

9. The data acquisition and transmission method for smart energy meters according to claim 8, characterized in that, When calculating the comprehensive transmission load assessment value based on the real-time acquisition tags, periodic acquisition tags, event-triggered acquisition tags, transmission priority parameters, and data integrity weights, the following are included: Perform matrix operations on the scene weight coefficients and transmission priority parameters corresponding to each collection tag to obtain the basic load evaluation vector; The data integrity weight is combined with the amount of data in a single communication to generate a data integrity load component, and then adjusted according to the compression processing efficiency of the data type. The basic load assessment vector is weighted and fused with the data integrity load component, and additional load increments related to anomalies are superimposed when there are anomaly data identifiers. Based on the weighted fusion and abnormal load results, a comprehensive transmission load assessment value is generated to guide communication scheduling.

10. The data acquisition and transmission method for smart energy meters according to claim 9, characterized in that, Based on the comprehensive transmission load assessment value, when adaptively configuring the data acquisition frequency, data packet method, and transmission scheduling strategy for a single communication between the smart energy meter and the data concentrator or cloud management platform, the following are included: The comprehensive transmission load assessment value is compared with the first comprehensive transmission load assessment value and the second comprehensive transmission load assessment value. Based on the comparison result, the data acquisition frequency, data packetization method and transmission scheduling strategy are determined; wherein, the first comprehensive transmission load assessment value is less than the second comprehensive transmission load assessment value. When the comprehensive transmission load assessment value is less than or equal to the first comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the first data acquisition frequency, the first data packetization method, and the priority transmission scheduling strategy, respectively. When the comprehensive transmission load assessment value is greater than the first comprehensive transmission load assessment value and less than or equal to the second comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the second data acquisition frequency, the second data packetization method, and the balanced transmission scheduling strategy, respectively. When the comprehensive transmission load assessment value is greater than the second comprehensive transmission load assessment value, the data acquisition frequency, data packetization method, and transmission scheduling strategy are determined to be the third data acquisition frequency, the third data packetization method, and the rate-limiting transmission scheduling strategy, respectively.

Images