Data acquisition modeling method for electricity utilization information acquisition

Abstract

The invention discloses a data acquisition modeling method for electricity utilization information acquisition, which comprises the following steps of: reading historical electricity meter acquisition records and power grid load data for preprocessing to generate standardized electricity meter response characteristic data; constructing a six-dimensional networking behavior feature vector, and generating an ammeter dynamic partitioning result through a density clustering algorithm; generating a partition-level resource allocation strategy in combination with the ammeter metadata and the data value evaluation model; and a double-layer time wheel scheduling mechanism is constructed, the outer layer manages resource allocation among partitions, the inner layer controls electricity meter decline time slice allocation, and an optimized acquisition scheduling execution plan is generated. The problem that a traditional method neglects power grid topology constraints and data value differences is solved, the data acquisition success rate and the resource utilization efficiency are remarkably improved, and an intelligent technical solution is provided for an electric power information acquisition system.

Description

Technical Field

[0001] The present invention relates to power electronics technology, in particular to a data acquisition modeling method for power consumption information acquisition. Background Art

[0002] With the deepening of the construction of the smart grid and the continuous improvement of the power market reform, the power consumption information acquisition system has become the core infrastructure for power enterprises to achieve refined management, improve service quality, and ensure the safe and stable operation of the power grid. This system provides accurate and timely data support for power dispatching, load forecasting, electricity bill calculation, demand response and other services through real-time data acquisition of distributed smart meters. However, in the face of the increasing scale of meters, the complex and changeable network environment, and diverse business requirements, traditional data acquisition methods face severe challenges in terms of efficiency, success rate, and resource utilization. Establishing an efficient and intelligent data acquisition modeling method has important theoretical significance and practical value for improving the overall operation efficiency of the power system, reducing operating costs, and enhancing the intelligent level of the power grid.

[0003] The current power consumption information acquisition modeling method mainly adopts a static scheduling strategy based on a fixed time window, and batches the meters for acquisition through a preset time period. Existing research focuses on the construction of time series prediction models, using historical response time data to establish ARIMA models, neural network models or support vector machine models to predict the response performance of meters, and formulating acquisition strategies based on the prediction results. Some research has introduced a multi-period adaptive mechanism to adjust acquisition parameters according to the network load characteristics in different time periods. In terms of resource allocation, existing methods mainly perform static grouping based on the geographical location of meters or transformer ownership, and use polling or priority queues for scheduling management. Some advanced solutions have begun to consider the statistical characteristics of meter response time, and group meters with similar response characteristics through cluster analysis, but the clustering features are relatively single, mainly based on basic statistics such as the mean and variance of historical response time.

[0004] Existing technical solutions have significant deficiencies in dealing with the complexity and dynamics of meter networking behavior. For example, traditional clustering methods fail to fully consider the internal relationship between meters and the power grid topology, resulting in grouping results that may group meters with similar behaviors but far electrical distances together, violating the physical constraints of the power system, causing unnecessary network overhead and potential system instability. In addition, existing resource allocation strategies lack a deep perception of the data value differences of meters, and fail to establish a comprehensive evaluation model for data timeliness, business importance, and data integrity. In the case of limited resources, it is impossible to make acquisition decisions that maximize value, resulting in the possible absence or delay of key business data due to improper resource allocation. These two core technical problems directly affect the overall performance and reliability of the power consumption information acquisition system. Summary of the Invention

[0005] Objective of the Invention: To provide a data collection and modeling method for power consumption information collection, aiming to solve at least one technical problem existing in the prior art.

[0006] Technical Solution: A data collection and modeling method for power consumption information collection includes: Reading historical electricity meter collection records and grid load data, preprocessing, and generating standardized electricity meter response feature data And performing clustering analysis on it to generate the electricity meter dynamic partition result; Obtaining electricity meter metadata from the business system, combining the standardized electricity meter response feature data and the electricity meter dynamic partition result, and generating a partition-level resource allocation strategy; Based on the partition-level resource allocation strategy, adopting a two-layer time wheel scheduling method, and generating an optimized collection scheduling execution plan by managing the resource allocation between partitions through the outer time wheel and controlling the decreasing time slice allocation of electricity meters within the partition through the inner time wheel.

[0007] Further, generating the optimized collection scheduling execution plan includes: Based on the partition-level resource allocation strategy, constructing the data structure of the outer time wheel, dividing the collection window into multiple time slices and setting the scheduling order according to the partition priority, and generating the outer time wheel scheduling table; According to the pre-calculated electricity meter-level priority index, configuring the data structure of the inner time wheel for the electricity meters within each partition, generating the electricity meter scheduling sequence within the partition, and allocating decreasing time slices to each electricity meter therein, with the electricity meter having a higher collection success rate obtaining a smaller time slice; Integrating the outer time wheel scheduling table, the electricity meter scheduling sequence within the partition, and the pre-configured retry policy table to generate the collection scheduling execution plan.

[0008] Further, allocating decreasing time slices to each electricity meter in the electricity meter scheduling sequence within the partition includes: Reading the electricity meter scheduling sequence within the partition, extracting the historical RTT sequence and the success rate sequence, Combining the real-time RTT value, calculating the RTT change rate, the RTT variance mutation index, and the success rate decrease index, generating a network anomaly score through a multi-dimensional network anomaly scoring fusion model, calculating the network stability index, and adjusting the anomaly detection threshold accordingly; Identifying the anomaly status electricity meters based on the network anomaly score and the anomaly detection threshold, calculating the time slice quality adjustment factor for the stable mode electricity meters, and calculating the real-time quality adjustment factor for the response mode electricity meters based on the real-time network quality; generating the time slice allocation value accordingly and performing boundary constraint processing.

[0009] Further, generating the standardized electricity meter response feature data includes: Filter the historical electricity meter collection records within a time range and handle outliers to generate the cleaned historical data; Perform correlation analysis on the cleaned historical data and the grid load data, calculate the correlation coefficient between the response performance of each electricity meter and the grid load, and generate the electricity meter-load correlation data; Integrate the period performance statistics of the cleaned historical data and the electricity meter-load correlation data to generate the standardized electricity meter response characteristic data.

[0010] Furthermore, perform correlation analysis on the cleaned historical data and the grid load data, including: Obtain the multi-level grid load including transformer-level load, feeder-level load, and regional total load; and perform time series decomposition on it together with the electricity meter response time series in the cleaned historical data, calculate the delay cross-correlation coefficient of each component, and generate the component-level correlation characteristics; Weight the load correlations at each level according to the electrical hop count from the electricity meter to the transformer to generate the topological weighted correlation degree; Perform frequency domain transformation on the cleaned historical data and the multi-level grid load, calculate the coherence spectrum, and generate the frequency domain correlation characteristics; Calculate the change in the correlation between the electricity meter response and the grid load in different time windows, and detect the correlation mutation points to generate the correlation time-varying characteristics; Weightedly integrate the above outputs, including component-level correlation characteristics, topological weighted correlation degree, frequency domain correlation characteristics, and correlation time-varying characteristics, to generate the electricity meter-load correlation index.

[0011] Furthermore, generate the electricity meter dynamic partitioning results, including: Based on the standardized electricity meter response characteristic data, construct a networking behavior feature vector including average response time, standard deviation of response time, short-term success rate, long-term success rate, stability index, and load correlation, and normalize it to generate the normalized feature vector; Calculate the Mahalanobis distance between each pair of electricity meters based on the normalized feature vector to generate the electricity meter behavior similarity matrix; Apply the improved density clustering algorithm to analyze the electricity meter behavior similarity matrix to obtain the clustering results; Generate the electricity meter dynamic partitioning results based on the clustering results, including partition ID, the set of electricity meters in each partition, and the partition center point characteristics.

[0012] Furthermore, the improved density clustering algorithm is a hierarchical clustering algorithm with topological constraints: Read the original grid topology data from the grid management system, construct the grid topology graph structure, calculate the electrical hop count matrix between electricity meters, identify the transformer nodes of different voltage levels as topological anchor points, and generate the topological constraint clustering center set; Based on the similarity matrix of electricity meter behaviors and the electrical distance matrix, through the reconstruction of similarity weighted by electrical distance, a topological constraint similarity matrix is generated; Based on the set of topological constraint clustering centers and the topological constraint similarity matrix, at the level of primary anchor points, coarse-grained clustering is performed, followed by refined partitioning within the clusters, and topological connectivity verification is achieved. A topological connected clustering result is generated and the dynamic boundary is optimized to generate a stabilized partitioning result and clustering output.

[0013] Furthermore, the reconstruction of similarity weighted by electrical distance includes: Extract the behavior similarity and electrical distance of each pair of electricity meters from the similarity matrix of electricity meter behaviors and the electrical distance matrix, construct a dual-constraint input data set, multiply the behavior similarity therein by the exponential decay factor of the electrical distance to generate a topological weighted similarity matrix; When the electricity meters belong to different voltage levels, a level penalty factor is used to perform level correction on the similarity matrix of electricity meter behaviors to generate a level constraint similarity matrix; Integrate the topological weighted similarity matrix and the level constraint similarity matrix to generate a topological constraint similarity matrix.

[0014] Furthermore, the networking behavior feature vectors include: average response time, standard deviation of response time, short-term success rate, long-term success rate, stability index, and correlation coefficient with the grid load.

[0015] Furthermore, generating a partition-level resource allocation strategy includes: Obtain the electricity meter metadata from the business system, and calculate the basic score of business value based on the user type and electricity consumption; Based on the collection time interval in the standardized electricity meter response feature data, calculate the data timeliness value, and calculate the data gap value based on the collection success records; Integrate the basic score of business value, the data timeliness value, and the data gap value, and calculate the data value index of each electricity meter through weighted summation; Based on the electricity meter dynamic partition result and the data value index of each electricity meter, calculate the total partition value and the partition historical success rate, and calculate the partition resource allocation ratio accordingly, and generate a partition-level resource allocation strategy in combination with the total available collection time.

[0016] Furthermore, calculating the data value index includes: Extract the basic score of business value, the data timeliness value, and the data gap value from the historical data, normalize them to generate three types of normalized value indicators; and adjust the indicator values according to the current system state to generate dynamic weight coefficients; Use the dynamic weight coefficients to perform weighted summation on the three types of normalized value indicators to obtain the basic value score, and add the interaction enhancement term between the value indicators to obtain the interaction value score; Set compensation coefficients, importance coefficients, and gain coefficients for electricity meters with consecutive collection failures, key user electricity meters, and electricity meters with specified monitoring statuses respectively, and calculate the scenario adjustment coefficient; Integrate the basic value score, interaction value score, and scenario adjustment coefficient, and calculate the data value index.

[0017] Furthermore, calculate the partition resource allocation ratio, including: Based on the total partition value, partition historical success rate, partition cohesion, and partition scale, comprehensively calculate the partition importance index; Generate a dynamic balance coefficient according to the system global statistical characteristics; Based on the partition importance index and the dynamic balance coefficient, calculate the preliminary resource allocation ratio; Generate the minimum resource quota and the maximum resource quota according to the resource upper and lower limit guarantee mechanism; Integrate the preliminary resource allocation ratio, the minimum resource quota, and the maximum resource quota, generate an optimized resource allocation ratio that meets the constraint conditions, and calculate the partition time quota accordingly. Combine the partition priority sorting to form a partition-level resource allocation strategy.

[0018] Beneficial effects: Solve the problem that traditional methods ignore the power grid topology constraints and data value differences, significantly improve the data collection success rate and resource utilization efficiency, and provide an intelligent technical solution for the power information collection system. Brief Description of the Drawings

[0019] Figure 1 is the flowchart of the present invention.

[0020] Figure 2 is the flowchart of the present invention for generating an optimized collection scheduling execution plan.

[0021] Figure 3 is the flowchart of the present invention for allocating decreasing time slices to each electricity meter in the electricity meter scheduling sequence within a partition.

[0022] Figure 4 is the flowchart of the present invention for generating standardized electricity meter response characteristic data. Detailed Embodiments

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the drawings and specific embodiments. It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0024] Embodiment 1. This embodiment provides a data collection modeling method for power consumption information collection.

[0025] As Figure 1As shown in the figure, this method is an automated processing flow running on the background server of the power information collection system, including the following steps: Step S101: Read historical electricity meter collection records and grid load data, perform preprocessing, generate standardized electricity meter response feature data, and conduct clustering analysis on it to generate the electricity meter dynamic partition result.

[0026] The historical electricity meter collection records refer to the original data obtained from the database of the electricity consumption information collection system, including electricity meter ID, collection time, response time, success / failure status, etc. The grid load data refers to the total load power data of the transformer, feeder, or area related to the area where the electricity meter is located, obtained from the grid dispatching system or SCADA system. The standardized electricity meter response feature data refers to a set of structured data that can comprehensively and comparably describe the response characteristics of each electricity meter after cleaning, statistics, and correlation analysis. The electricity meter dynamic partition refers to the process of dividing electricity meters with similar networking behavior characteristics and similar electrical topological positions into the same group (i.e., partition), and this partition result will be dynamically updated according to the changes in electricity meter behavior.

[0027] In this embodiment, first, read the electricity meter collection records of the most recent 30 days and the grid load data for the corresponding period through the database interface. Then, perform outlier processing on the response time in the collection records. For example, use the improved Z-score method to identify and replace outliers with the median to form the cleaned data. Then, based on the cleaned data and the load data, calculate the correlation coefficient between the response performance (such as average response time, success rate) of each electricity meter and the grid load, and combine the period performance statistics to finally integrate into the standardized electricity meter response feature data. Finally, based on this feature data, apply a density clustering algorithm (such as the topological constraint hierarchical clustering algorithm) to cluster all electricity meters to obtain the dynamic partition result.

[0028] This step constructs feature data that can reflect the internal operation rules of the electricity meter by integrating historical data and load data and conducting in-depth analysis, laying a foundation for subsequent scientific partitioning. The dynamic partition is achieved through clustering analysis, solving the drawback that traditional static grouping cannot adapt to changes in the network environment.

[0029] Step S102: Obtain electricity meter metadata from the business system, and generate a partition-level resource allocation strategy by combining the standardized electricity meter response feature data and the electricity meter dynamic partition result.

[0030] The electricity meter metadata refers to the data that describes the business attributes of the electricity meter, such as user type (industrial, residential), electricity consumption, billing level, etc. The partition-level resource allocation strategy refers to the decision-making plan for allocating specific collection time quotas and priorities to each dynamic partition.

[0031] This step obtains the metadata of the electricity meters from the power marketing system or the customer relationship management system (CRM). Then, combining the metadata (calculating business value), standardizing the electricity meter response characteristic data (calculating data timeliness value and gap value), and the data value evaluation model, a comprehensive data value index is calculated for each electricity meter. After that, according to the electricity meter dynamic partition result generated in step S101, the total value and historical success rate of each partition are summarized. Finally, through the value-success rate balance model, the resource allocation ratio of each partition is calculated, and combined with the total available acquisition time, the final partition-level resource allocation strategy is formed.

[0032] This step introduces a differential evaluation of data value, enabling limited acquisition resources to be tilted towards high-value electricity meters and partitions, and solving the problem that traditional methods apply evenly and cannot achieve maximum value acquisition.

[0033] Step S103: Based on the partition-level resource allocation strategy, adopt a two-layer time wheel scheduling method. Through the outer time wheel to manage the resource allocation between partitions and the inner time wheel to control the decreasing time slice allocation of electricity meters within a partition, an optimized acquisition scheduling execution plan is generated.

[0034] In the two-layer time wheel scheduling, the slots of the outer time wheel correspond to different electricity meter partitions, and the rotation of its pointer determines which partition should be acquired currently. The slots of the inner time wheel correspond to each electricity meter within a partition, and the rotation of its pointer determines the specific acquisition order and time of the electricity meters within the partition. The decreasing time slice allocation means that the higher the acquisition success rate of an electricity meter, the shorter the acquisition timeout time (time slice) allocated, and vice versa.

[0035] For example, first, based on the partition-level resource allocation strategy generated in step S102, construct the outer time wheel. For example, divide the total acquisition window (such as 2 hours) into multiple time periods according to the resource allocation ratio of each partition, and arrange them on the time wheel in the order of partition priority. Then, for each partition, construct the inner time wheel according to the priority index of its internal electricity meters (combining data value and historical response performance), and allocate a decreasing time slice to each electricity meter. Finally, integrate the scheduling table of the outer time wheel, the scheduling sequences of the inner time wheels of all partitions, and the preset retry strategy (such as exponential backoff) to generate a detailed final execution plan including the specific acquisition time of each electricity meter.

[0036] The two-layer time wheel mechanism ensures the fairness and priority of resource allocation between partitions macroscopically and realizes the efficient scheduling of acquisition within a partition microscopically. The decreasing time slice allocation strategy can save time for good electricity meters and leave more time resources for difficult electricity meters, thus improving the overall acquisition success rate and resource utilization efficiency.

[0037] Through the above steps, the present invention provides an intelligent data acquisition and modeling method, which comprehensively considers the dynamic behavior of electric meters, the physical topology of the power grid, the business value of data, and an efficient scheduling algorithm, can significantly improve the success rate of data acquisition and resource utilization efficiency, and solves the technical problems that traditional methods ignore the topological constraints of the power grid and the differences in data value.

[0038] Embodiment 2: Provide a method for generating an optimized acquisition scheduling execution plan and time slice allocation. This embodiment details how to generate an optimized acquisition scheduling execution plan, especially the core decreasing time slice allocation mechanism.

[0039] In this embodiment, the process of generating an optimized acquisition scheduling execution plan specifically includes: Step S201: Based on the partition-level resource allocation strategy, construct an outer time wheel data structure, divide the acquisition window into multiple time slices, set the scheduling order according to the partition priority, and generate an outer time wheel scheduling table.

[0040] Assume that according to Embodiment 1, the partition-level resource allocation strategy has been obtained, which includes the priority sorting and time quotas [1238s, 1080s, 950s, 857s, 900s, 785s, 727s, 663s] of 8 partitions (1B, 1A, 2A, 2B, 3, 4, 5, 6). The total acquisition window is 2 hours (7200 seconds). The outer time wheel is constructed as a scheduling list, and each item in the list defines the scheduling window of a partition.

[0041] For example, the first item is: {zone_id: "1B", priority: 18.5, time_start: 0, time_end: 1238}.

[0042] The second item is: {zone_id: "1A", priority: 16.02, time_start: 1238, time_end: 2318}, and so on. This scheduling table defines the macroscopic acquisition process, ensuring that high-priority areas are collected first, and the resources (time) they obtain are consistent with the strategy.

[0043] Step S202: According to the pre-computed meter-level priority index, configure an inner time wheel data structure for the meters in each partition, generate the meter scheduling sequence within the partition, and allocate decreasing time slices to each meter.

[0044] The meter-level priority index is calculated by integrating the data value index of a single meter and its historical response performance, and is used to determine its acquisition priority within the partition. The decreasing time slice refers to the acquisition timeout time allocated to the meter. The core idea is that the higher the historical acquisition success rate of the meter, the more stable its network condition usually is. Therefore, a shorter timeout time can be set for it, and if it fails, it can be resolved by retrying. On the contrary, meters with a low success rate require a longer timeout time to cope with potential network delays.

[0045] For example, taking partition 1A as an example, its time quota is 1080 seconds and it contains 20 meters. First, calculate the priority index P_meter of each meter in the partition. For example, the P_meter of meter M001 is 0.765 and that of M003 is 0.742. Then, sort all the meters in the partition in descending order according to P_meter to form a basic scheduling sequence. Next, allocate a decreasing time slice to each meter in the sequence. The higher the acquisition success rate of the meter, the smaller the time slice it gets.

[0046] As a preferred implementation, allocating a decreasing time slice to each meter in the meter scheduling sequence for the partition includes: Step S202a: Read the historical RTT (round-trip time) sequence and success rate sequence of the meters in the partition, and combine the real-time obtained RTT value to calculate the RTT change rate, the RTT variance mutation index, and the success rate drop index. Through a multi-dimensional network anomaly scoring fusion model, such as weighted summation Anomaly_score = w1×RTT_change_rate + w2×RTT_variance_spike + w3×success_rate_drop, generate a quantified network anomaly score. At the same time, calculate the network stability index and dynamically adjust the threshold Anomaly_threshold_adaptive for judging whether it is abnormal accordingly.

[0047] Step S202b: Based on the network anomaly score and the anomaly detection threshold, identify the meters that are currently in an abnormal state. The system maintains an operation mode identifier (stable mode or response mode) for each meter. For meters in the stable mode, use a damping smoothing mechanism to calculate a time slice quality adjustment factor damped_quality_factor to avoid drastic changes in the time slice due to accidental network jitters. For meters in the response mode (i.e., network anomalies are detected), calculate a real-time quality adjustment factor real_time_quality_factor based on the real-time network quality (such as queue delay, processing jitter) so that it can quickly adapt to network changes.

[0048] Step S202c: Generate the final time slice allocation value according to the operating mode of the electricity meter, combining the base time slice, success rate, priority, and corresponding adjustment factors. For example, for a stable-mode electricity meter: the final time slice allocation value T_final = T_base × (1 - δ × SR) × damped_quality_factor. For a responsive-mode electricity meter, an emergency multiplier may also be multiplied. Finally, perform boundary constraint processing on the calculated time slice to ensure that its value is within a preset reasonable range (for example, not less than 15 seconds and not greater than 120 seconds).

[0049] Step S203: Integrate the outer time wheel scheduling table, the electricity meter scheduling sequence within the partition, and the pre-configured retry policy table to generate the acquisition scheduling execution plan.

[0050] Integrate the outer scheduling table in Step S201 and the inner scheduling sequences generated for each partition in Step S202. At the same time, configure a general or hierarchical retry policy. Preferably, use the exponential backoff algorithm, that is, the waiting time for each retry increases exponentially, and add random jitter to avoid conflicts. For example, Delay(retry_n) = T_base × 2^(n - 1) × (1 + random(0, 0.1)), where n is the number of retries. The finally generated execution plan is a list of instructions refined to each electricity meter, specifying its initial acquisition time point, allocated timeout (time slice), and retry policy after failure. For example: {meter_id: "M001", start_time: 1283.8, duration: 32.46, retry_config: [51.4, 105.8, 201.7], zone_id: "1A"}.

[0051] Through the above steps, the present invention not only plans the macroscopic acquisition sequence and resources, but also realizes the refined and adaptive control of the acquisition time of each electricity meter at the microscopic level. In particular, the introduced network anomaly detection and dual-mode time slice allocation mechanism enable the scheduling strategy to intelligently balance system stability and rapid response to network changes, solving the technical problem that traditional fixed or simple decreasing time slices cannot adapt to dynamic network environments, thereby further improving the overall operation efficiency of the system while ensuring a high success rate.

[0052] Embodiment 3: This embodiment is used to describe how the basic data as the model input, the standardized electricity meter response characteristic data, is generated, especially the innovative electricity meter-load correlation analysis process among them.

[0053] Specifically, the process of generating the standardized electricity meter response characteristic data includes: Step S301: Filter the historical electricity meter collection records by time range and process outliers to generate cleaned historical data.

[0054] Select the collection records in the most recent cycle (e.g., 30 days) from the electricity consumption information collection system database as the analysis sample. For the response time series, an adaptive threshold model is used for outlier detection. This model combines the median (RT_median) and the interquartile range (RT_IQR), and is dynamically adjusted according to the coefficient of variation (CV) of the data. The threshold formula can be set as: Upper_threshold = RT_median + α × RT_IQR × log(1 + CV). For the identified outliers, local linear interpolation or the median of the same period in the recent past is used for replacement according to whether they are sporadic or continuous, so as to generate a clean and reliable historical data set.

[0055] Step S302: Conduct correlation analysis on the cleaned historical data and the grid load data, calculate the correlation coefficient between the response performance of each electricity meter and the grid load, and generate electricity meter-load correlation data.

[0056] As a preferred implementation, this step specifically includes: Step S302a: Obtain multi-level grid loads and perform time series decomposition. Obtain multi-level load data such as transformer-level, feeder-level, and regional total loads from the grid management system. Then, apply time series decomposition techniques (such as STL decomposition) to decompose the response time series of the electricity meter and each level of load series into trend terms, periodic terms, and residual terms. This is aimed at separating the change patterns at different time scales for more refined correlation analysis.

[0057] Step S302b: Calculate component-level correlation features.

[0058] For each decomposed component, multi-scale cross-correlation analysis is used to calculate the delayed cross-correlation coefficient between them. For example, calculate the correlation between the trend term of the electricity meter response time and the trend term of the transformer load within the delay window from -24 hours to +24 hours, and find the maximum correlation and its corresponding time delay. This solves the problem that traditional correlation analysis ignores the time delay effect.

[0059] Step S302c: Generate topological weighted correlation. According to the position of the electricity meter in the power grid topology diagram, calculate its electrical hop count (Hop_distance) to each level of transformer. Then, perform non-linear weighting on the correlation of each level of load, and the formula is: WC(meter_i, load_j)=Corr(meter_i, load_j)×exp(-k×Hop_distance). Where k is the attenuation coefficient. This makes the loads closer in electrical connection contribute more to the correlation, reflecting the physical constraints.

[0060] Step S302d: Generate frequency-domain correlation features and time-varying features. Perform fast Fourier transform (FFT) on the response time series and load series, calculate the coherence spectrum in the frequency domain to identify strong correlation patterns at specific frequencies (such as daily cycle, weekly cycle). At the same time, adopt the sliding window method to calculate the change of correlation in different time windows, and detect the mutation points of correlation to generate time-varying features of correlation.

[0061] Step S302e: Weighted integration to generate the electricity meter-load correlation index. Finally, integrate the component-level correlation features, topological weighted correlation, frequency-domain correlation features, and time-varying features of correlation generated in the above steps through a multi-feature fusion model (such as weighted summation) to generate a comprehensive electricity meter-load correlation index that can comprehensively describe the association relationship between the electricity meter and the load.

[0062] Step S303: Integrate the period performance statistics and electricity meter-load correlation data of the cleaned historical data to generate standardized electricity meter response feature data.

[0063] Merge the period performance statistics (such as average response time, success rate by period) of the cleaned data generated in Step S301 with the electricity meter-load correlation index generated in Step S302. The finally formed standardized electricity meter response feature data is a structured table or dataset, where each row represents an electricity meter and each column represents a feature (such as average response time from 0 to 4 o'clock, load correlation index, etc.). This dataset will be used as the basis for subsequent construction of networking behavior feature vectors and dynamic partitioning.

[0064] Through the above steps, the present invention can generate a set of high-quality and high-dimensional electricity meter response feature data. Compared with the prior art, its innovation lies in constructing a load correlation analysis model that takes into account time delay, topological structure, frequency-domain characteristics, and dynamic changes, so that the understanding of the electricity meter response behavior is no longer limited to itself, but is closely combined with the macroscopic operating state of the power grid, providing a solid data support for more accurate subsequent modeling and decision-making.

[0065] Embodiment 4. This embodiment describes the process of dynamically partitioning electric meters based on standardized electric meter response characteristic data, especially the hierarchical clustering algorithm with topological constraints used therein.

[0066] Step S401: Based on the standardized electric meter response characteristic data, construct a networking behavior feature vector including average response time, standard deviation of response time, short-term success rate, long-term success rate, stability index, and load correlation, and normalize it.

[0067] The networking behavior feature vector is a multi-dimensional vector, and each dimension thereof is an index extracted or calculated from the standardized electric meter response characteristic data and capable of characterizing a certain aspect of the electric meter network behavior.

[0068] For each electric meter, construct a six-dimensional networking behavior feature vector V = [RT_avg, RT_std, SR_short, SR_long, Stab_idx, Load_corr]. Among them, the calculation of each feature has been optimized: RT_avg uses time decay weighting to make the weight of recent data higher; RT_std synthesizes the fluctuations of multiple time scales; SR_short and SR_long also use the time decay function for calculation to achieve a smooth transition; Stab_idx evaluates stability from three dimensions of time, fluctuation, and period; Load_corr is the comprehensive index calculated in Embodiment 3. After constructing the vector, to eliminate the influence of different feature dimensions, an improved MinMax method is used to normalize the feature vectors of all electric meters.

[0069] Optionally, the interaction relationship between features can also be mined to generate auxiliary combined features such as the response-success rate composite index RT_SR = RT_avg × (1 - SR_short) to enhance the feature expression ability and add them to the feature vector.

[0070] Step S402: Calculate the Mahalanobis distance between each pair of electric meters based on the normalized feature vector to generate an electric meter behavior similarity matrix.

[0071] Different from the Euclidean distance, this embodiment uses the Mahalanobis distance to calculate the similarity between electric meters. The Mahalanobis distance can consider the correlation between features and is more suitable for processing the high-dimensional correlated features constructed by the present invention. Calculate the Mahalanobis distance between all pairs of electric meters and convert it into a similarity (for example, sim = 1 / (1 + dist_Mahal)), and finally form an N×N electric meter behavior similarity matrix, where N is the total number of electric meters.

[0072] Step S403: Apply an improved density clustering algorithm to analyze the electric meter behavior similarity matrix to obtain a clustering result.

[0073] As a preferred embodiment, this step adopts a hierarchical clustering algorithm with topological constraints, specifically including: Step S403a: Parse the power grid topology and preset anchor points. Read the topology data from the power grid management system, construct a topology graph, and calculate the electrical hop count matrix between all electricity meters. Identify transformer nodes of different voltage levels (such as 10kV transformers) as topological anchor points, and calculate the characteristic centroid of the downstream electricity meter group as the anchor point feature, so as to construct a set of topological constraint clustering centers consistent with the physical structure of the power grid.

[0074] Step S403b: Reconstruct the similarity with electrical distance weighting. This step corrects the behavior similarity matrix generated in step S402 to incorporate topological constraints.

[0075] First, multiply the behavior similarity sim ij by an exponential decay factor of an electrical distance ed ij exp(-λ×ed ij / ed_max) to generate a topologically weighted similarity matrix. This makes the final similarity of two electricity meters with a greater electrical distance be significantly reduced even if their behaviors are similar.

[0076] Secondly, introduce a level penalty factor. When two electricity meters belong to different voltage levels, their behavior similarity will be multiplied by a penalty factor less than 1 to generate a level constraint similarity matrix.

[0077] Finally, fuse the above two corrected similarity matrices through an adaptive weight to generate the final topologically constrained similarity matrix.

[0078] Step S403c: Perform hierarchical constrained clustering and conduct connectivity verification. Based on the set of topological constraint clustering centers and the topologically constrained similarity matrix, first perform coarse-grained clustering at the level of first-level anchor points. Then, within each coarse-grained cluster, refer to the second-level anchor points for refined partitioning. Importantly, after partitioning, topological connectivity verification will be carried out to ensure that all electricity meters within each partition are physically connected, and non-connected partitions will be re-segmented.

[0079] Step S404: Generate the dynamic partitioning result of electricity meters based on the clustering result, including the partition ID, the set of electricity meters in each partition, and the feature of the partition center point.

[0080] For the topological connected clustering results generated in step S403c, dynamic boundary optimization is also performed. Fuzzy electricity meters located at the partition boundaries are identified and redistributed based on their topological connection relationships (such as whether they are on the same feeder) and considerations of partition load balancing. At the same time, to avoid frequent and drastic changes in the partition results, a partition stability enhancement mechanism is introduced to smooth the current partition results and the historical results. Finally, a stabilized partition result is output, which includes the ID of each partition, the list of electricity meters, and the updated partition center features and topological attributes, etc.

[0081] For example, in a certain run, 200 electricity meters are divided into 8 partitions. Partition 1A contains 20 electricity meters, with topological characteristics of an average electrical hop count of 2.4, a connectivity of 0.132, and a partition quality index of 0.282.

[0082] Through the above steps, the core advantage of the dynamic partitioning method constructed by the present invention is the deep integration of data-driven behavioral similarity and physical rule-driven topological constraints. This solves the fundamental problem in the prior art that the clustering results may violate the physical constraints of the power system (such as classifying two electricity meters that are far apart on the power grid into one category), ensuring the electrical rationality and practical operability of the partitioning scheme, and laying a solid foundation for subsequent efficient resource allocation and scheduling.

[0083] Example 5. This example is used to describe how to formulate a reasonable resource allocation strategy for each dynamic partition.

[0084] Step S501: Obtain electricity meter metadata from the business system and calculate the data value index of each electricity meter in combination with the standardized electricity meter response feature data.

[0085] The data value index is used to measure the comprehensive importance and urgency of collecting data from a certain electricity meter at the current moment.

[0086] Step S501a: Calculate three types of basic values.

[0087] Business value basic score (BV_base): Based on electricity meter metadata, such as user type (industrial users have a higher weight than residential users) and historical electricity consumption (higher electricity consumption means a higher weight), it is calculated through weighted calculation.

[0088] Data timeliness value (TV): According to the time interval since the last successful collection, it is calculated using an exponential decay function. The longer the time, the higher the value. The formula is TV = 1 / (1 + λ × (t_current - t_last_successful)). t_current and t_last_successful are the current time and the last successful collection time respectively.

[0089] Data Gap Value (GV): Calculated based on historical successful acquisition records using a data integrity assessment model. The more severe and critical the historical data is missing, the higher the value.

[0090] Step S501b: Perform non - linear fusion and context adjustment.

[0091] First, perform adaptive normalization on the above three types of value indicators.

[0092] Then, determine the dynamic weight coefficients (w1, w2, w3) of the three types of indicators according to the current business scenario (such as whether it is a key point in the billing cycle), and perform weighted summation to obtain the basic value score V_base. Optionally, add an interaction enhancement term V_interact between value indicators to reflect the synergistic effect of value.

[0093] Next, introduce a context reward mechanism to calculate a context adjustment coefficient SAF for meters in special situations. For example, set a failure compensation coefficient for meters with consecutive acquisition failures, a user importance coefficient for key user meters, and a monitoring gain coefficient for meters in special monitoring states.

[0094] Finally, integrate and calculate the final comprehensive data value indicator V = (V_base + V_interact) × SAF.

[0095] For example, meter M001 is an industrial user and has had one acquisition failure recently, and its timeliness value is relatively high. After calculation, its BV_norm = 0.528, TV_norm = 0.9999, GV_norm = 0.001. During the billing cycle, the weight of BV increases, and finally its comprehensive data value indicator V is calculated to be 0.800.

[0096] Step S502: Based on the dynamic partitioning results of meters and the data value indicators of each meter, calculate the total value of the partition and the historical success rate of the partition, and calculate the partition resource allocation ratio accordingly.

[0097] Specifically as follows: Step S502a: Calculate the partition importance index. First, sum the data value indicators V of all meters within the partition to obtain the total value of the partition V_zone. Then, comprehensively consider the total value of the partition, the partition size (Size), and the partition cohesion (Cohesion) to calculate a partition importance index Imp_zone. For example, Imp_zone = V_zone × (1 + λ1 × log(Size / Size_avg)) × (1 + λ2 × (Cohesion - Cohesion_avg)). This is to balance the impacts of value, size, and partition quality.

[0098] Step S502b: Adaptive calculation of resource allocation ratio. Dynamically determine the value - success rate balance coefficients α and β according to the system global state (such as the current acquisition load, data value difference degree). Then, use the formula R_raw(zone_i)=(α×Imp_zone_i + β×SR_zone_i×Size_i) / Σ(...) to calculate the preliminary resource allocation ratio for each zone.

[0099] Step S502c: Perform constraint optimization. To prevent low - value zones from getting no resources at all or high - value zones from occupying too many resources, a resource lower - limit guarantee and upper - limit control mechanism is designed to calculate the minimum resource quota R_min and the maximum resource quota R_max for each zone. Finally, apply a constraint optimization algorithm (such as the Lagrange multiplier method) to adjust the preliminary resource allocation ratio under the conditions of meeting the upper and lower limit constraints and the sum being 1, and obtain the final optimized resource allocation ratio R_opt.

[0100] Step S503: Generate a zone - level resource allocation strategy in combination with the total available acquisition time.

[0101] Multiply the total available acquisition time (such as 7200 seconds) by the optimized resource allocation ratio R_opt of each zone to obtain the specific time quota T_zone for that zone. At the same time, calculate the priority ranking of the zones according to the zone importance index, etc. Finally, integrate the optimized resource allocation ratio, the zone time quota table, and the zone priority ranking to form a complete zone - level resource allocation strategy as the input for the double - layer time - wheel scheduling.

[0102] For example, the importance index of zone 1A is 15.03, its optimized resource allocation ratio is 15.0%, and it obtains a quota of 1080 seconds within a total time of 7200 seconds, and its priority ranks 2nd among 8 zones.

[0103] Through the above steps, the present invention establishes a complete decision - making chain from the evaluation of the data value of a single electric meter to the macro - control of zone - level resources. It ensures the purposefulness of the acquisition resource allocation through a dynamic and multi - dimensional value evaluation model; through value - success rate balance and constraint optimization, it realizes the scientificity and robustness of resource allocation. This solves the technical problems of single resource allocation strategy and lack of value orientation in the prior art, enabling limited communication and computing resources to be utilized where they are most needed, thereby maximizing the overall benefit of the system.

[0104] Embodiment VI. Describe the preferred implementation method of multi - dimensional electric meter - load correlation analysis This Embodiment VI is a more detailed and preferred implementation method for the correlation analysis of the historical data after cleaning and the power grid load data in step S302 of Embodiment III.

[0105] Step S601: Obtain the multi-level power grid load and perform time series decomposition.

[0106] The multi-level power grid load refers to the load data obtained from different levels of the power grid (such as transformers, feeders, regions), which can reflect the local operation status of the power grid more precisely than a single load value.

[0107] Specifically, obtain a multi-dimensional load matrix including transformer-level load (such as T1: 850 kW), feeder-level load (such as F1: 280 kW), and regional total load (2850 kW) from the power grid management system. Then, apply time series decomposition technology (preferably, use the STL decomposition method) to decompose the response time series of each electricity meter and the load series at each level into a trend term (Trend), a periodic term (Seasonal), and a residual term (Residual).

[0108] For example, the response time series of electricity meter M001 is decomposed into three sub-series: Trend_RT, Seasonal_RT, and Residual_RT. The output of this step is multiple sets of decomposed time series components for subsequent refined correlation analysis.

[0109] Step S602: Calculate the component-level correlation, topological weighted correlation degree, and frequency domain / time-varying characteristics.

[0110] Calculate the component-level correlation. For each component decomposed in step S601, use multi-scale cross-correlation analysis to calculate the correlation coefficient at different time delays (such as -24h to +24h), and find the maximum correlation and the corresponding time delay. For example, it is calculated that the trend term of the response time of electricity meter M001 has the highest correlation with the trend term of the load of transformer T1 at a delay of -2 hours, reaching 0.82.

[0111] Calculate the topological weighted correlation degree: Calculate the electrical hop count Hop_distance from the electricity meter to the transformer according to the power grid topology diagram, and then perform exponential decay weighting on the original correlation coefficient: WC(meter_i, load_j) = Corr(meter_i, load_j) × exp(-k × Hop_distance). For example, the electrical hop count from electricity meter M001 to T1 is 3, and its topological weighted correlation degree is corrected to 0.417. Corr represents the correlation coefficient.

[0112] Calculate the frequency domain correlation characteristics: Perform a fast Fourier transform (FFT) on the original time series and calculate the coherence spectrum in the frequency domain to identify strong coupling relationships at specific periods (such as 24 hours).

[0113] Calculate the time-varying features of correlation: Use algorithms such as sliding window and CUSUM to detect the changes and mutation points of the correlation between the electricity meter response and the load over time.

[0114] Step S603: Generate the electricity meter-load correlation index through weighted integration.

[0115] Construct a multi-feature fusion model. Through weighted integration, fuse the component-level correlation features, topology-weighted correlation degree, frequency-domain correlation features, and time-varying correlation features generated in step S602 into a comprehensive electricity meter-load correlation index.

[0116] For example, the final load correlation index of electricity meter M001 is calculated as follows: Load_corr_index = 0.3×Max_corr + 0.25×WC + 0.25×Max_coherence + 0.2×Corr_variance = 0.64. This index will be used as a key dimension of the networking behavior feature vector.

[0117] Through the above steps, this embodiment elaborates in detail an innovative method for analyzing the correlation between electricity meters and loads. It is fundamentally different from the single and static correlation calculation of traditional methods. By introducing multi-dimensional perspectives such as time series decomposition, physical topology weighting, and frequency-domain analysis, it can more profoundly and accurately describe the internal relationship between electricity meter communication behavior and grid operation status, providing unprecedented high-quality data features for subsequent intelligent zoning and scheduling.

[0118] Embodiment Seven: Describe the preferred implementation of the hierarchical clustering algorithm with topological constraints This Embodiment Seven is a more detailed and preferred implementation of applying the improved density clustering algorithm to step S403 in Embodiment Four. It details how to deeply integrate the power grid topology structure into the clustering process.

[0119] Step S701: Analyze the power grid topology structure and preset topological anchor points.

[0120] Read the topology data from the power grid management system to construct a topology graph structure containing nodes such as transformers, feeders, and electricity meters. Based on this graph, use the multi-source shortest path algorithm (such as BFS) to calculate the electrical hop count matrix edij between all electricity meters. Then, identify the key nodes in the network, such as transformers above 10 kV, as first-level topological anchor points, and calculate the weighted feature centroid of the downstream electricity meter groups as the feature vectors of these anchor points to form an initial clustering center set that is strongly correlated with the physical structure.

[0121] Step S702: Reconstruct the similarity with electrical distance weighting.

[0122] Similarity reconstruction refers to the correction of a pure data-driven behavior similarity matrix to forcefully consider physical topology constraints.

[0123] From the electricity meter behavior similarity matrix and the electrical distance matrix, extract the behavior similarity simij and electrical distance edij for each pair of electricity meters. Apply an exponential decay function to correct them: Corrected_simij = simij × exp(-λ × edij / ed_max). This step reduces the final similarity of electricity meters with a large physical distance, even if their behavior characteristics are similar. For example, the original behavior similarity between M001 and M045 is 0.72, but due to an electrical distance of 5, the corrected similarity drops to 0.207.

[0124] When two electricity meters belong to different voltage levels, apply a level penalty factor Level_penaltyij to correct their similarity again. For example, the similarity between M001 (10 kV) and M085 (400 V) is multiplied by a penalty factor of 0.9.

[0125] Finally, through adaptive weights α and β, fuse the above topological weighted similarity and level constraint similarity to generate the final topological constraint similarity matrix.

[0126] Step S703: Perform hierarchical constraint clustering and dynamic optimization.

[0127] Based on the topological anchors in step S701 and the topological constraint similarity matrix in step S702, use an improved hierarchical clustering algorithm to first perform coarse-grained clustering at the high-level anchor level and then perform refined partitioning within the clusters.

[0128] For all generated partitions, perform topological connectivity checks to ensure that all electricity meters within a partition physically form a connected subgraph, and re-partition the unconnected partitions with topological awareness.

[0129] Identify electricity meters located at the partition boundaries with ambiguous ownership, and perform final allocation based on their upstream and downstream topological relationships (such as whether they belong to the same feeder) and considerations of partition load balancing, and introduce a historical weight smoothing mechanism to enhance the stability of the partition results.

[0130] Through the above steps, in this embodiment, the physical topology structure of the power grid is deeply and fully embedded in the electricity meter clustering algorithm in a quantitative and computable manner. This fundamentally solves the technical problem that existing data-driven clustering methods ignore the specific constraints of the power system and result in partition results that are not practical. The final partition results are not only cohesive in data behavior but also reasonable in physical connection, providing a high-quality decision-making basis for subsequent resource allocation and network management.

[0131] Example VIII: Preferred Implementation of Describing Comprehensive Data Value Evaluation and Partition Resource Allocation Strategy This Example VIII is a more detailed and preferred implementation of steps S501 and S502 in Example V. It details the whole process from the value evaluation of a single electricity meter to the formulation of a macro resource strategy.

[0132] Step S801: Calculate the comprehensive data value index.

[0133] The comprehensive data value index V is a quantitative score that dynamically and multi-dimensionally measures the importance and urgency of collecting the data of an electricity meter at the current moment.

[0134] Calculate the basic values of three dimensions: Calculate the basic business value score BV_base based on user type and electricity consumption; Calculate the data timeliness value TV by applying an exponential decay function according to the time since the last successful collection; Evaluate the data gap value GV based on historical missing situations.

[0135] After performing adaptive normalization on the three types of value indexes, dynamically determine the weight coefficients (w1, w2, w3) according to the current business scenario (such as whether it is a billing cycle, whether the data integrity rate alarms) for weighted summation to obtain the basic value score V_base. Optionally, to reflect the synergy effect, an interaction enhancement term V_interact between value indexes can be added.

[0136] To cope with special situations, introduce a situation adjustment coefficient SAF. For example, assign a failure compensation coefficient FC to electricity meters with consecutive failed collections; Assign a user importance coefficient UIC to electricity meters of key users such as hospitals and large factories; Assign a monitoring gain coefficient MBC to electricity meters in special monitoring states such as power protection and negative control.

[0137] Integrate all components through the formula V = (V_base + V_interact) × SAF to calculate the final comprehensive data value index for each electricity meter.

[0138] Step S802: Calculate the partition-level resource allocation ratio.

[0139] Sum up the values V of all electricity meters in the partition to obtain the total partition value V_zone. Then, comprehensively consider the total partition value, partition size Size, and partition cohesion Cohesion (from clustering quality evaluation) to calculate the partition importance index Imp_zone. This ensures that partitions with high value, large scale, and good quality obtain higher importance scores.

[0140] Dynamically determine the coefficients α and β for balancing value and success rate according to the overall operating state of the system. For example, when network resources are scarce, increase the weight of β to prioritize ensuring the overall acquisition success rate; when there are significant differences in data value, increase the weight of α to prioritize collecting high-value data.

[0141] Based on the importance index and balance coefficient, calculate the preliminary resource allocation ratio R_raw. Then, apply the resource lower limit guarantee mechanism (to ensure that each partition has a minimum acquisition opportunity) and the upper limit control mechanism (to prevent a single partition from occupying too many resources) to generate the minimum R_min and maximum R_max quotas. Finally, through a constrained optimization algorithm, generate the final optimized resource allocation ratio R_opt that satisfies all constraints and has a sum of 1.

[0142] Step S803: Generate a complete partition-level resource allocation strategy.

[0143] Multiply the total available acquisition time T_total by R_opt to obtain the specific time quota T_zone for each partition. Then, calculate the partition priority ranking in combination with the partition importance index and urgency, etc. The finally formed strategy includes the resource ratio, time quota, and scheduling order of each partition.

[0144] Through the above steps, this embodiment establishes a scientific and intelligent resource allocation decision-making system. It solves the problems of blind and rigid resource allocation in traditional methods through refined data value evaluation and global adaptive allocation algorithms. It ensures that valuable acquisition resources can be accurately allocated to the places where the most value can be generated, realizing the transformation from completing acquisition tasks to maximizing acquisition value.

[0145] Embodiment Nine: Preferred implementation for describing adaptive hybrid response time slice allocation and parameter self-learning This Embodiment Nine is a more detailed and preferred implementation for allocating decreasing time slices to each electric meter in Step S202 of Embodiment Two, and incorporates the concept of parameter self-learning optimization.

[0146] Step S901: Construct a network anomaly detection trigger.

[0147] The network anomaly detection trigger is a real-time monitoring module used to determine whether there are abnormal fluctuations in the network communication quality of a single electric meter.

[0148] Monitor the RTT (Round-Trip Time) and acquisition success rate of each electricity meter in real time, and calculate three key anomaly metrics: the RTT change rate RTT_change_rate, the RTT variance spike metric RTT_variance_spike (the ratio of the current variance to the historical benchmark), and the success rate drop metric success_rate_drop (the decline in the short-term success rate relative to the long-term).

[0149] Fuse the above three metrics into a comprehensive Anomaly_score through a multi-dimensional network anomaly scoring fusion model (such as weighted summation). At the same time, calculate a network stability index network_stability_index based on recent network quality statistics and use it to dynamically adjust the anomaly determination threshold Anomaly_threshold_adaptive. This enables the threshold to be appropriately relaxed when the overall network is unstable, avoiding false alarms.

[0150] Step S902: Implement a dual-mode time slice allocation strategy.

[0151] Based on Anomaly_score and Anomaly_threshold_adaptive, determine whether the electricity meter is currently in an abnormal state. The system maintains an operating mode identifier for each electricity meter. When an anomaly is continuously detected, it switches to the response mode; when the network has been stable for a period of time, it switches back to the stable mode.

[0152] For electricity meters in the stable mode, the time slice adjustment factor is calculated using a damping smoothing mechanism to absorb occasional network jitters and maintain the stability of acquisition. The formula is: T_final = T_base × (1 - δ × SR) × damped_quality_factor.

[0153] For electricity meters in the response mode, the time slice adjustment factor is directly calculated based on the real-time network quality (such as the queue delay analyzed by RTT difference) and can be additionally multiplied by an emergency multiplier emergency_multiplier determined by the severity of the anomaly. This enables the time slice to be quickly amplified to cope with network deterioration and strive to ensure successful acquisition. The formula is T_final = T_base × (1 - δ × SR) × real_time_quality_factor × emergency_multiplier.

[0154] Step S903: Final calculation of time slices and self-learning optimization of parameters.

[0155] Perform boundary constraints (maximum / minimum time slice check) on T_final calculated in step S902, and perform time resource balance within the partition to ensure that the total time consumption of the partition does not exceed the quota.

[0156] After the scheduling plan is executed, the system will track the actual acquisition success rate actual_success_rate of each electric meter under the new time slice.

[0157] Compare the actual success rate with the predicted success rate of the model to form a success rate loss function success_loss. Then, adopt optimization algorithms such as gradient descent to automatically fine-tune the key parameters in the model according to the loss function, such as the weights w1, w2, w3 in the anomaly scoring model, and the adjustment coefficients α, β, γ, etc. in the time slice calculation formula. Its update logic can be expressed as w_new = w_old - learning_rate × d(success_loss) / dw. learning_rate is the learning rate. success_loss is the success rate loss function, learning_rate is the learning rate that controls the update step size of the parameters, and w includes the weight coefficients w1, w2, w3 and the adjustment coefficients α, β, γ, etc.

[0158] Through the above steps, this embodiment proposes an adaptive time slice allocation mechanism. By real-time anomaly detection and dual-mode switching, it solves the fundamental problem that traditional static or simple decreasing time slices cannot adapt to complex dynamic network environments. More importantly, by introducing a closed-loop of success rate feedback and parameter self-learning, the entire scheduling system has the ability of self-evolution and continuous optimization, and can maintain the optimal performance during long-term operation.

[0159] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A data acquisition modeling method for power consumption information acquisition, characterized in that Including: Read historical electricity meter collection records and grid load data, preprocess them, generate standardized electricity meter response feature data, and perform clustering analysis on it to generate electricity meter dynamic partition results; Obtain electricity meter metadata from the business system, combine the standardized electricity meter response feature data and the electricity meter dynamic partition results to generate a partition-level resource allocation strategy; Based on the partition-level resource allocation strategy, adopt a two-layer time wheel scheduling method. Through the outer time wheel to manage the resource allocation between partitions and the inner time wheel to control the decreasing time slice allocation of electricity meters within the partition, generate an optimized acquisition scheduling execution plan.

2. The method according to claim 1, wherein Generate an optimized acquisition scheduling execution plan, including: Based on the partition-level resource allocation strategy, construct the data structure of the outer time wheel, divide the acquisition window into multiple time slices, and set the scheduling order according to the partition priority to generate the outer time wheel scheduling table; According to the pre-computed electricity meter-level priority index, configure the data structure of the inner time wheel for the electricity meters within each partition, generate the electricity meter scheduling sequence within the partition, and allocate decreasing time slices to each of the electricity meters. The electricity meter with a higher acquisition success rate gets a smaller time slice; Integrate the outer time wheel scheduling table, the electricity meter scheduling sequence within the partition, and the pre-configured retry policy table to generate the acquisition scheduling execution plan.

3. The method according to claim 2, characterized in that Allocate decreasing time slices to each electricity meter in the electricity meter scheduling sequence within the partition, including: Read the electricity meter scheduling sequence within the partition, extract the historical RTT sequence and success rate sequence; Combine the real-time RTT value, calculate the RTT change rate, the RTT variance mutation index, and the success rate decrease index, generate a network anomaly score through a multi-dimensional network anomaly scoring fusion model, calculate the network stability index, and adjust the anomaly detection threshold accordingly; Identify abnormal state electricity meters based on the network anomaly score and the anomaly detection threshold, calculate the time slice quality adjustment factor for stable mode electricity meters, and calculate the real-time quality adjustment factor for response mode electricity meters based on the real-time network quality; generate the time slice allocation value accordingly and perform boundary constraint processing.

4. The method according to claim 1, characterized in that, Generate standardized electricity meter response feature data, including: Perform time range screening and outlier processing on historical electricity meter collection records to generate cleaned historical data; Perform correlation analysis on the cleaned historical data and the grid load data, calculate the correlation coefficient between the response performance of each electricity meter and the grid load, and generate electricity meter-load correlation data; Integrate the period performance statistics of the cleaned historical data and the electricity meter-load correlation data to generate standardized electricity meter response feature data.

5. The method according to claim 4, characterized in that, Perform correlation analysis on the cleaned historical data and the grid load data, including: Obtain multi-level grid loads including transformer-level load, feeder-level load, and regional total load; and perform time series decomposition on them together with the electricity meter response time series in the cleaned historical data, calculate the delay cross-correlation coefficient of each component, and generate component-level correlation features; Weight the load correlations at each level according to the electrical hop count from the electricity meter to the transformer to generate the topological weighted correlation degree; Perform frequency domain transformation on the cleaned historical data and the multi-level grid load, calculate the coherence spectrum, and generate frequency domain correlation features; Calculate the correlation change between the electricity meter response and the grid load in different time windows, detect the correlation mutation points, and generate the time-varying correlation features; Weightedly integrate the above outputs to generate the electricity meter-load correlation index.

6. The method according to claim 1, wherein Generate the dynamic partitioning results of the electricity meters, including: Based on the standardized electricity meter response feature data, construct a networking behavior feature vector containing the average response time, standard deviation of the response time, short-term success rate, long-term success rate, stability index, and load correlation, and normalize it to generate a normalized feature vector; Calculate the Mahalanobis distance between each pair of electricity meters based on the normalized feature vector to generate an electricity meter behavior similarity matrix; Perform clustering analysis on the electricity meter behavior similarity matrix to obtain the clustering results; Generate the dynamic partitioning results of the electricity meters based on the clustering results, including the partition ID, the set of electricity meters in each partition, and the feature of the partition center point.

7. The method according to claim 6, wherein The improved density clustering algorithm is a hierarchical clustering algorithm with topological constraints: Read the original grid topology data from the grid management system, construct the grid topology graph structure, calculate the electrical hop count matrix between electricity meters, identify the transformer nodes of different voltage levels as topological anchor points, and generate a set of topological constraint clustering centers; Based on the electricity meter behavior similarity matrix and the electrical distance matrix, generate a topological constraint similarity matrix through the reconstruction of the similarity weighted by the electrical distance; Based on the set of topological constraint clustering centers and the topological constraint similarity matrix, perform coarse-grained clustering at the first-level anchor point level, then perform refined partitioning within the cluster, and implement topological connectivity verification, generate the topological connected clustering results and perform dynamic boundary optimization, generate the stabilized partitioning results and cluster the output.

8. The method according to claim 7, characterized in that The reconstruction of the similarity weighted by the electrical distance includes: Extract the behavior similarity and electrical distance of each pair of electricity meters from the electricity meter behavior similarity matrix and the electrical distance matrix, construct a dual-constraint input data set, multiply the behavior similarity therein by the exponential decay factor of the electrical distance to generate a topological weighted similarity matrix; When the electricity meters belong to different voltage levels, use the level penalty factor to perform level correction on the electricity meter behavior similarity matrix to generate a level constraint similarity matrix; Integrate the topological weighted similarity matrix and the level constraint similarity matrix to generate a topological constraint similarity matrix.

9. The method according to claim 6, wherein The networking behavior feature vector includes: average response time, standard deviation of the response time, short-term success rate, long-term success rate, stability index, and correlation coefficient with the grid load.

10. The method according to claim 1, characterized in that, Generate the partition-level resource allocation strategy, including: Obtain the electricity meter metadata from the business system and calculate the basic score of the business value based on the user type and electricity consumption; Based on the collection time interval in the standardized electricity meter response feature data, calculate the data timeliness value, and calculate the data gap value based on the collection success records; Integrate the basic score of the business value, the data timeliness value, and the data gap value, and calculate the data value index of each electricity meter through weighted summation; Based on the dynamic partitioning results of the electricity meters and the data value indicators of each electricity meter, calculate the total partition value and the historical success rate of the partition, and calculate the partition resource allocation ratio accordingly, and generate the partition-level resource allocation strategy in combination with the total available collection time.

11. The method according to claim 10, wherein Calculate the data value index, including: Extract the business value base score, data timeliness value, and data gap value from historical data, normalize them, and generate three types of normalized value indicators; and adjust the indicator values according to the current system status to generate dynamic weight coefficients; Use the dynamic weight coefficients to perform weighted summation on the three types of normalized value indicators to obtain the base value score, and add the interaction enhancement term between the value indicators to obtain the interaction value score; Set compensation coefficients, importance coefficients, and gain coefficients for the electricity meters with continuous collection failures, key user electricity meters, and electricity meters with specified monitoring status respectively, and calculate the situation adjustment coefficient; Integrate the base value score, interaction value score, and situation adjustment coefficient to calculate the data value indicator.

12. The method according to claim 10, wherein Calculate the partition resource allocation ratio, including: Based on the total partition value, partition historical success rate, partition cohesion, and partition scale, comprehensively calculate the partition importance index; generate a dynamic balance coefficient according to the system global statistical characteristics; Based on the partition importance index and the dynamic balance coefficient, calculate the preliminary resource allocation ratio; generate the minimum resource quota and the maximum resource quota according to the resource upper and lower limit guarantee mechanism; Integrate the preliminary resource allocation ratio, the minimum resource quota, and the maximum resource quota to generate an optimized resource allocation ratio that meets the constraint conditions, calculate the partition time quota accordingly, and form a partition-level resource allocation strategy in combination with the partition priority ranking.

Images