Dynamic perception and carbon footprint factor library construction system based on regional energy

Abstract

The invention relates to the technical field of carbon emission management, in particular to a regional energy-based dynamic perception and carbon footprint factor library construction system, which comprises an energy consumption extraction module, a floating judgment module, a node comparison module, a factor matching module and a factor filing module. According to the method, a composite mechanism of floating threshold limit and change direction consistency judgment is adopted for the screening mode of the regional power grid node energy consumption data, and the precision of identifying the energy consumption change stability is remarkably improved. In the time sequence characteristic analysis, the continuous judgment of a floating section is combined with a direction trend filtering interpolation algorithm, so that the extraction of an energy consumption change interval is more accurate and representative. In a vector construction and classification link, multi-dimensional feature combinations of maximum and minimum change point coordinates, change time span and the like are introduced, and a time sequence interpolation and inflection point identification technology is combined, so that the three-dimensional expression of change trend identification is effectively enhanced.

Description

Technical Field

[0001] This invention relates to the field of carbon emission management technology, and in particular to a system for dynamic sensing and carbon footprint factor database construction based on regional energy. Background Technology

[0002] The field of carbon emission management technology encompasses methods and systems for identifying, recording, calculating, and managing carbon emissions during energy use through information technology. The core of this technology lies in establishing a precise carbon emission accounting system based on energy consumption data and its corresponding emission factors, and utilizing dynamic monitoring, data collection, and model calculations to develop quantitative analysis and management capabilities for the entire carbon emission process.

[0003] Among them, the system for dynamic sensing and carbon footprint factor library construction based on regional energy refers to a technical solution that extracts carbon emission factors and forms a structured factor library by real-time monitoring of regional energy consumption processes, combined with energy categories and usage scenarios. This includes differences in the collection methods of multiple energy types within the region, conversion standards for energy units such as electricity and heat, accurate establishment of the mapping relationship between carbon emission factors and energy categories, and a dynamic update mechanism for factor data.

[0004] While existing technologies possess preliminary capabilities for real-time monitoring of regional energy consumption, they lack in-depth screening of the combination of directional trends and amplitude continuity in extracting energy consumption data fluctuation characteristics. This leads to biases in the structural stability of the extracted energy consumption segments. For example, screening solely based on energy consumption differences easily includes high-frequency random fluctuation segments, interfering with carbon factor identification. In inter-node energy consumption comparisons, the lack of coupling analysis based on directional consistency and amplitude differences between adjacent nodes results in insufficient representativeness of the master node identification results, reducing the overall matching effectiveness of the factor database. Existing technologies often rely on static factor template matching in carbon footprint factor matching, failing to incorporate similarity analysis based on the dynamic vector characteristics of master nodes, thus limiting the accuracy and adaptability of factor label determination. Most archived structures employ single-point data linking, failing to integrate and aggregate the correlation trends between node groups, resulting in a lack of completeness and temporal continuity in carbon footprint factor records, which is detrimental to subsequent source tracing and intelligent analysis. Overall, existing technologies have significant shortcomings in terms of data extraction accuracy, scientific node selection, factor matching flexibility, and archive structure coherence, which affect the stable operation and intelligent evolution capabilities of the carbon footprint monitoring system. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a system for dynamic sensing of regional energy and construction of a carbon footprint factor library.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a system for dynamic sensing and carbon footprint factor database construction based on regional energy includes:

[0007] The energy consumption extraction module collects energy consumption data from regional power grid nodes, filters out energy consumption data intervals in the floating feature data where the difference is lower than the floating threshold and the direction of change has not reversed, and outputs periodic energy consumption floating data.

[0008] The floating determination module judges and processes the stability of energy consumption fluctuation amplitude and direction based on the periodic energy consumption fluctuation data, constructs a set of energy consumption change vectors, and generates fluctuation level identification parameters according to the frequency of change and the amplitude ratio.

[0009] The node comparison module analyzes the band differences between multiple nodes in the power grid through the fluctuation level identification parameters and performs amplitude similarity judgment. It compares the consistency of the change direction of adjacent nodes and filters the master node, and outputs the master node identification code and node amplitude mapping table.

[0010] The factor matching module uses the master node identification code and the node amplitude mapping table to call the energy consumption change vector set of the master node in the node amplitude mapping table, compares it with the carbon footprint factor library, inputs it into the support vector machine model to classify the similarity index, and outputs the matching factor subclass label.

[0011] The factor archiving module constructs a regional power grid carbon footprint factor library by matching factor subclass tags and generates carbon footprint factor archive records.

[0012] As a further aspect of the present invention, the periodic energy consumption fluctuation data includes a time series number, the start and end times of the fluctuation interval, and the number of data points within the interval; the energy consumption change vector set specifically includes the coordinates of the maximum change point, the coordinates of the minimum change point, and the change time span; the fluctuation level identification parameters include the change frequency level, the unit time amplitude ratio level, and the directional trend stability level; the master node identification code includes a unique node number, a fluctuation cycle identification label, and a master node status mark; the node amplitude mapping table specifically includes the amplitude difference corresponding to the node number, the change direction symbol, and the directional correlation coefficient; the matching factor subclass label specifically refers to the factor subclass number, the band category, and the node source type identifier; the carbon footprint factor archive record includes the subclass label, node number, timestamp, vector value, and change trend code.

[0013] As a further aspect of the present invention, the energy consumption extraction module includes:

[0014] The energy consumption acquisition submodule collects time-series energy consumption data from all monitoring nodes in the regional power grid, records it according to a set period, obtains the original power value sequence of the nodes in a continuous period, and calls the time period number, acquisition timestamp and energy consumption of each group of nodes to generate a continuous period energy consumption sequence set.

[0015] The floating construction submodule calculates the energy consumption difference based on the energy consumption of adjacent data points in the continuous periodic energy consumption sequence set and determines whether the change direction is consistent. Data pairs with energy consumption differences lower than the floating threshold and the same direction are marked as floating sequences, and a floating feature segment sequence table is constructed according to the sliding window order.

[0016] The floating threshold is set by calculating the skewness coefficient of the energy consumption difference distribution within a continuous period and combining it with an empirical benchmark.

[0017] The interval screening submodule determines whether the data pairs meet the requirements of difference and direction conditions in three out of five consecutive sequences based on the duration and number of data pairs of each floating segment in the floating feature segment sequence list. It then extracts the floating segments that meet the conditions and marks the start and end times of the data pairs and the corresponding node numbers to obtain the periodic energy consumption floating data.

[0018] As a further aspect of the present invention, the floating determination module includes:

[0019] The trend smoothing submodule, based on the periodic energy consumption fluctuation data, obtains the continuous energy consumption in each data sequence and sorts it in fixed period units. It performs continuous interpolation on adjacent energy consumption in the sorting results and performs fitting processing through the Savitzky-Golay filtering algorithm with the energy consumption change direction as the reference standard to generate a smooth trend value sequence.

[0020] The segment extraction submodule detects the location of inflection points in the smooth trend value sequence and extracts the energy consumption corresponding to the inflection points. It marks the minimum value before and after the first inflection point and the maximum value before and after the last inflection point as the segment boundary, calculates the start and end times and the distance between points of the data within the segment boundary, and performs combined processing to obtain the energy consumption change vector set.

[0021] The level determination submodule calls the point spacing, maximum and minimum value difference and number of inflection points in the sequence of each segment in the energy consumption change vector set, calculates the ratio of change frequency to energy consumption amplitude within a specified time, maps the results to the level classification standard for interval matching and marks the corresponding level identifier to obtain the fluctuation level identifier parameter.

[0022] As a further aspect of the present invention, the node comparison module includes:

[0023] The amplitude comparison submodule obtains the minimum and maximum fluctuation values ​​of each node within the same continuous period based on the fluctuation level identifier parameter, calculates the amplitude interval difference corresponding to the node, and performs standard deviation calculation and error distribution judgment on all nodes based on the amplitude interval difference result to obtain amplitude difference distribution information.

[0024] The direction correlation submodule classifies nodes into groups based on the energy consumption change direction signs of adjacent periods in the amplitude difference distribution information, and judges the consistency of the energy consumption change direction of nodes in each group based on the Pearson correlation coefficient function, thus obtaining a set of direction consistency coefficients.

[0025] The master node filtering submodule performs a joint evaluation of the amplitude difference and directional consistency value of nodes in the same group based on the set of directional consistency coefficients, selects the node with the smallest amplitude difference and the largest directional consistency value as the master node, extracts the master node number and the corresponding number list of the node group, and obtains the master node identification code and the node amplitude mapping table.

[0026] As a further aspect of the present invention, the factor matching module includes:

[0027] The factor extraction submodule calls the corresponding number of the master node and extracts the associated energy consumption change vector set according to the master node identification code and node amplitude mapping table, obtains the numerical sequence and time index information corresponding to the energy consumption change vector set, and generates the master node vector feature group.

[0028] The feature calculation submodule calls each vector value in the main node vector feature group, calculates the Euclidean distance between the vector value and each standard factor vector in the carbon footprint factor library, and normalizes it to represent it in the form of a similarity set. The similarity set is then input into the support vector machine model for classification boundary judgment to obtain the similarity classification output set.

[0029] The category determination submodule selects the nearest neighbor label of the current main node vector in the classification space based on the classification boundary results in the similarity classification output set, and locates the corresponding factor number and attribute subclass according to the position of the nearest neighbor label in the carbon footprint factor library to obtain the matching factor subclass label.

[0030] As a further aspect of the present invention, the factor archiving module includes:

[0031] The tag collection submodule obtains the node group number to which the corresponding main node belongs based on the matching factor subclass tag and the main node identification code, extracts the node group member identifier from the node amplitude mapping table, constructs the node group group corresponding to the subclass tag, and generates the node group mapping set.

[0032] The vector integration submodule calls each node identifier in the node group mapping set, extracts the node change vector from the current cycle energy consumption monitoring data, and retrieves the change vector record in the power grid monitoring database according to the node identifier. The current node change vector is combined with the retrieved change vector record in chronological order to obtain the node group vector set.

[0033] The structure entry module sets the structured field items as subclass label, node number, timestamp, vector value and trend code according to the node group vector set and the matching factor subclass label corresponding to the node. It writes all data into the carbon footprint factor library storage structure according to the label dimension and generates carbon footprint factor archive records as factor library structure data output.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0035] In this invention, a composite mechanism of floating threshold limitation and consistency judgment of change direction is adopted in the screening method of regional power grid node energy consumption data, which significantly improves the accuracy of identifying the stability of energy consumption changes. In time series feature analysis, the combination of floating segment persistence judgment and directional trend filtering interpolation algorithm makes the extraction of energy consumption change intervals more accurate and representative. In the vector construction and classification stage, multi-dimensional feature combinations such as the coordinates of the maximum and minimum change points and the change time span are introduced, combined with time series interpolation and inflection point identification technology, which effectively enhances the three-dimensional expression of change trend identification. The comparison of band differences and directional consistency between adjacent nodes introduces standard deviation analysis and Pearson correlation coefficient calculation to ensure that the master node screening results are more representative and structurally stable. The similarity calculation between master node vectors and carbon footprint factors adopts a linkage mechanism of normalized Euclidean distance and support vector machine classifier, which enhances the matching accuracy and category judgment sensitivity. In the final factor archiving process, a node group division and time series vector integration mechanism under the label dimension is adopted to realize the dynamic collection and real-time storage of carbon footprint factor structured data, improving the application timeliness and data traceability efficiency of carbon footprint information. The combination of technologies such as multi-dimensional feature fusion, refined trend extraction, and dynamic vector aggregation enables the carbon footprint factor matching and archiving process to not only have the advantages of fast response speed and high matching accuracy, but also good adaptability and evolution, making it suitable for energy carbon emission monitoring needs in various scenarios. Attached Figure Description

[0036] Figure 1 This is a system flowchart of the present invention;

[0037] Figure 2 This is a flowchart of the energy consumption extraction module of the present invention;

[0038] Figure 3 This is a flowchart of the floating determination module of the present invention;

[0039] Figure 4 This is a flowchart of the node comparison module of the present invention;

[0040] Figure 5 This is a flowchart of the factor matching module of the present invention;

[0041] Figure 6This is a flowchart of the factor archiving module of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0043] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0044] Please see Figure 1 The system for dynamic sensing of regional energy and construction of a carbon footprint factor database includes:

[0045] The energy consumption extraction module collects energy consumption data of regional power grid nodes, constructs a floating feature dataset based on the difference and directional trend between energy consumption data points, filters out energy consumption data intervals in the floating feature dataset where the difference is lower than the floating threshold and the direction of change has not reversed, and outputs periodic energy consumption floating data.

[0046] The floating determination module judges and processes the stability of energy consumption fluctuation amplitude and direction based on the periodic energy consumption fluctuation data, calls the Savitzky-Golay filtering algorithm to smooth and fit the periodic energy consumption fluctuation data sequence, extracts the minimum, maximum and time span of the fluctuation segment, constructs the energy consumption change vector set, and generates fluctuation level identification parameters according to the frequency of change and the amplitude ratio.

[0047] The node comparison module analyzes the band differences between multiple nodes in the power grid through the fluctuation level identification parameter and performs amplitude similarity judgment. Based on the maximum and minimum difference error within a continuous period, it performs standard deviation comparison on the node amplitude interval, calls the Pearson correlation coefficient function to determine the consistency of the change direction of adjacent nodes and filters the master node, and outputs the master node identification code and node amplitude mapping table.

[0048] The factor matching module uses the master node identification code and the node amplitude mapping table to call the energy consumption change vector set of the master node in the node amplitude mapping table, compares it with the carbon footprint factor library, inputs it into the support vector machine model to classify the similarity index, and outputs the matching factor subclass label.

[0049] The factor archiving module constructs a regional power grid carbon footprint factor library by matching factor subclass labels, obtains the node group characteristics corresponding to the matching factor subclass labels and master node identification codes, calls the current period of each node group and the energy consumption change vector retrieved from the power grid monitoring database by node identifier, performs aggregation and storage operations by label classification, constructs structured storage fields including subclass labels, node numbers, timestamps, vector values ​​and change trend codes, and generates carbon footprint factor archiving records as the structural data output of the factor library;

[0050] The periodic energy consumption fluctuation data includes the time series number, the start and end times of the fluctuation interval, and the number of data points within the interval; the energy consumption change vector set specifically includes the coordinates of the maximum change point, the coordinates of the minimum change point, and the change time span; the fluctuation level identification parameters include the change frequency level, the unit time amplitude ratio level, and the directional trend stability level; the master node identification code includes the unique node number, the fluctuation cycle identification label, and the master node status mark; the node amplitude mapping table specifically includes the amplitude difference corresponding to each node number, the change direction sign, and the directional correlation coefficient; the matching factor subclass label specifically refers to the factor subclass number, the band category, and the node source type identifier; the carbon footprint factor archive record includes the subclass label, node number, timestamp, vector value, and change trend code.

[0051] Please see Figure 2 The energy extraction module includes:

[0052] The energy consumption acquisition submodule collects time-series energy consumption data from all monitoring nodes in the regional power grid, records it according to a set period, obtains the original power value sequence of the nodes in a continuous period, and calls the time period number, acquisition timestamp and energy consumption of each group of nodes to generate a continuous period energy consumption sequence set.

[0053] First, voltage and current signals from each node are collected in real time via power data sensors or metering devices. These signals are then converted into effective power values ​​by a power parameter conversion unit. Polling sampling is performed every 15 minutes, mapping the effective power data of each node within each cycle to the current timestamp. Next, a time-wheel scheduling mechanism is used to synchronously write these data into a cache area according to node group numbers. Simultaneously, the unique node number, the corresponding collection cycle number, and the timestamp are combined to form the primary key of the data record, identifying the ownership of each group of raw power data. Once the cache is full for the current set cycle (e.g., 96 data sets for 24 hours), the data sequence construction module is triggered to integrate the power value set of each node within that cycle. The integration process arranges the data within each cycle chronologically by sorting the timestamps, according to a fixed cycle window (…). A continuous power sequence is formed (e.g., 15 minutes). This constructs a time series set of the original power values ​​for each node within a complete cycle. For example, for monitoring node N001, between 00:00 and 24:00 on May 6, 2025, 96 time series power values ​​were recorded as [152W, 148W, 160W, ..., 175W]. Combined with the timestamp sequence [00:00, 00:15, ..., 23:45] and the cycle numbers P1, P2, ..., P96, a continuous cycle energy consumption sequence set {(N001, P1, 00:00, 152W), (N001, P2, 00:15, 148W), ..., (N001, P96, 23:45, 175W)} is finally constructed. After the sequence is completed, it can be used for subsequent float analysis processing.

[0054] The floating construction submodule calculates the energy consumption difference based on the energy consumption of adjacent data points in the continuous periodic energy consumption sequence set, and determines whether the change direction is consistent. Data pairs with energy consumption differences lower than the floating threshold and the same direction are marked as floating sequences, and a floating feature segment sequence table is constructed according to the sliding window order. The floating threshold is set by calculating the skewness coefficient of the distribution of energy consumption differences in the continuous period and combining it with empirical benchmarks.

[0055] Based on the energy consumption difference between adjacent data points in a continuous periodic energy consumption sequence set, a pairwise difference calculation is performed. For each node, a power difference calculation is performed between two adjacent sampling points. The difference between the current power value and the power value of the previous period is used for calculation. For example, for node N001, if period P5 is 160W and P4 is 150W, then the difference is ΔP = P5 - P4 = 10W. Then, it is determined whether the difference is greater than or equal to 0 to determine whether the direction of change is consistent. All point pairs with consistent directions and differences not exceeding a set threshold are marked. The floating threshold is obtained by setting the skewness coefficient of the historical power difference sample distribution and an empirical benchmark value. For example, for the difference data set of a certain node, ΔP = 10, -5, 7, -6, 3, -2, the Skewness skewness formula is used to calculate the difference. The sequence skewness result is 0.65. Based on the empirical benchmark, the floating threshold is set to ±8W. Then, the data sequence of each node is sequentially advanced through a sliding window mechanism. The window length is set to 5 sets of data. The floating judgment is performed on the continuous point pairs in each window to form a floating feature segment sequence. For example, if window W1 contains the data point difference set 2W, -4W, 3W, -5W, all within ±8W and the direction changes alternately, it does not constitute a floating segment. If the difference direction of 3 sets of continuous data in a certain window is consistent and within the threshold, such as 5W, 7W, 6W, it is marked as a floating sequence and its start and end time and difference value group are recorded. The floating segment sequence table of the corresponding node finally forms a sequence such as (N001, P5–P7, [5W, 7W, 6W]), which is used in the interval filtering stage.

[0056] The interval screening submodule determines whether the data pairs meet the requirement that at least three out of five consecutive sequences meet the difference and direction conditions based on the duration and number of data pairs of each floating segment in the floating feature segment sequence list. It then extracts the floating segments that meet the conditions and marks the start and end times of the data pairs and the corresponding node numbers to obtain the periodic energy consumption floating data.

[0057] Based on the floating feature segment sequence list, persistence and structure are judged for each floating segment. According to the rule: a segment must contain at least 3 sets of differences in the same direction and whose values ​​fall within the floating threshold range in 5 consecutive data pairs. For example, a floating segment is +6W, +5W, -3W, +4W, +7W, of which 4 sets are positive floating and the differences are all within ±8W. Therefore, this segment meets the extraction criteria. The extraction process starts from the segment's starting point and records consecutive data points that meet the criteria according to a sliding rule. When the number of point pairs meeting the criteria reaches... After three groups are collected, records are retained to form new periodic floating data segments. At the same time, the start and end sampling period numbers and corresponding node numbers are recorded and summarized into a segment table. For example, if the segment sequence (N001, P8–P12, [+6W, +5W, -3W, +4W, +7W]) meets the filtering criteria, the floating segment is extracted and marked as (N001, start P8, end P12), and summarized into the periodic floating data table. This table is used to output the data segments with stable small-amplitude changes within a specific time period for each node, which is convenient for subsequent processing.

[0058] Please see Figure 3 The float determination module includes:

[0059] The trend smoothing submodule is based on periodic energy consumption fluctuation data. It obtains the continuous energy consumption in each data sequence and sorts them in fixed period units. It performs continuous interpolation on adjacent energy consumption in the sorting results and performs fitting processing through Savitzky-Golay filtering algorithm with the direction of energy consumption change as the reference standard to generate a smooth trend value sequence.

[0060] First, extract the continuous energy consumption from each data point in the floating data sequence, and then sort them in ascending order according to a unified period division principle. For example, if the period unit is set to 15 minutes, the power data of monitoring node N002 within a certain time period, 148W, 151W, 150W, 155W, 149W, and 153W, are arranged in chronological order.

[0061] The sorting results for P1–P6 are 148W, 149W, 150W, 151W, 153W, and 155W respectively. Then, interpolation is performed on adjacent energy consumption groups. During interpolation, the power values ​​of two consecutive data points and their corresponding time points are called, and an intermediate point is inserted at a set interval. The value of this interpolation point is calculated using linear interpolation. For example, the interpolation value between 150W (P3) and 151W (P4) is 150.5W, resulting in the interpolation point set being 148W, 148.5W, 149W, 149.5W, 150W, 150.5W, and 151W. ...,155W. The interpolated sequence is then used to determine the direction based on the power increase / decrease trend. If the continuous power values ​​show a monotonically increasing or decreasing trend, it is marked as "positive" or "negative". For example, 149W→150.5W→151W→153W is determined to be positive. During trend fitting, smoothing is performed by using a window of 5 points in the interpolated sequence. The smoothing value is calculated once for each window, and the current point is combined with the two adjacent points on the left and right to perform polynomial curve approximation, finally generating a continuous and smooth trend power value sequence. For example, from the original interpolated sequence...

[0062] 148W, 148.5W, 149W, ..., 155W, generate a smoothed sequence.

[0063] 148.2W, 148.6W, 149.1W, ..., 154.8W were used as input data for subsequent structural analysis.

[0064] The segment extraction submodule detects the location of inflection points in the smooth trend value sequence and extracts the energy consumption corresponding to the inflection points. It marks the minimum value before and after the first inflection point and the maximum value before and after the last inflection point as the segment boundary, calculates the start and end times and the distance between points of the data within the segment boundary, and performs combined processing to obtain the energy consumption change vector set.

[0065] First, read the power value groups in each trend sequence, and sequentially detect the signs of adjacent power value changes to identify the locations where the power change direction reverses. These reversal points are marked as inflection points. For example, in the sequence 147W, 148W, 150W, 149W, 148W, 150W, 151W, the first decrease from 150W to 149W is defined as the first inflection point, and the subsequent increase from 148W to 150W is the second inflection point. For each inflection point, identify three data points before and after it, and find the local minimum and maximum values ​​among these points. Mark the minimum values ​​before and after the first inflection point as the starting boundary, and the maximum values ​​before and after the last inflection point as the ending boundary. For example, in a sequence before the first inflection point... The values ​​are 147W, 148W, 149W, and 150W, with the minimum value being 147W, which is set as the starting segment value. The maximum value before and after the inflection point at the end is 151W, which is set as the segment termination value, thus marking the start and end positions of the energy consumption segment. Then, the time difference between the starting and ending sampling points is calculated as the start and end times. For example, if the starting point is 10:15 on May 6, 2025, and the ending point is 11:30, then the start and end duration is 75 minutes. At the same time, the total number of sampling points between the two points is 6, so the point interval is 75 ÷ 5 = 15 minutes. The results are combined to form the segment vector item {start point: 147W, end point: 151W, duration: 75 minutes, number of points: 6}, which is used for subsequent level determination.

[0066] The level determination submodule calls the point spacing, maximum and minimum value difference and number of inflection points in the sequence of each segment in the energy consumption change vector set, calculates the ratio of change frequency to energy consumption amplitude within a specified time, maps the results to the level classification standard for interval matching and marks the corresponding level identifier to obtain the fluctuation level identifier parameter.

[0067] After obtaining the energy consumption variation vector set, the point spacing, maximum-minimum value difference, and number of inflection points in each segment are extracted one by one. First, the variation frequency of each segment is calculated by dividing the total number of inflection points by the segment duration. For example, if the number of inflection points in a vector is 4 and the start-end duration is 80 minutes, then the variation frequency is 4 ÷ (80 / 15) = 0.75 times / cycle. Next, the magnitude of power value change is calculated by dividing the difference between the ending power value and the starting power value by the total time period of the segment as the energy consumption amplitude ratio. For example, if the power variation is 151W - 147W = 4W and the total time is 80 minutes, then the power amplitude ratio is 4W ÷ 80 = 0.05W. The frequency variation is measured per minute, and then the ratio of frequency variation to energy consumption amplitude is mapped to the level division range for matching and judgment. The level division standard is based on the frequency variation range of 0–0.3, 0.3–0.7, 0.7–1.0, >1.0 and the power amplitude ratio range of 0–0.02, 0.02–0.06, >0.06. The corresponding levels are set as L1 to L4. If the frequency of the current segment is 0.75 and the amplitude ratio is 0.05, then the frequency is in the range of 0.7–1.0 and the amplitude ratio is in the range of 0.02–0.06, corresponding to level L3. This level is output as the fluctuation level identifier parameter, and the segment number and monitoring node number are attached to generate the final identifier set.

[0068] Please see Figure 4 The node comparison module includes:

[0069] The amplitude comparison submodule obtains the minimum and maximum fluctuation values ​​of each node within the same continuous period based on the fluctuation level identifier parameter, calculates the amplitude interval difference corresponding to the node, and performs standard deviation calculation and error distribution judgment on all nodes based on the amplitude interval difference results to obtain amplitude difference distribution information.

[0070] First, the maximum and minimum power values ​​of all monitoring nodes within the same continuous period are extracted to form a node power boundary set. Then, the amplitude interval difference for each node is calculated sequentially as the amplitude index value for that node. For example, if the node set is {N1, N2, N3, N4}, and its corresponding maximum and minimum power values ​​are: N1: 163W / 150W, N2: 160W / 151W, N3: 158W / 143W, N4: 165W / 153W, then the difference set is {13W, 9W, 15W, 12W}. The standard deviation is then calculated using this amplitude difference set, starting with the average value. Then calculate the squared deviation of the amplitude difference at each node from the mean: (13-12.25) 2 =0.5625,(9-12.25) 2 =10.5625, (15-12.25) 2 =

[0071] 7.5625, (12-12.25)2 =0.0625, calculate the average to get the variance 4.6875, standard deviation is Simultaneously, the differences are sorted in ascending order according to the node number to construct an error distribution set. If the set is concentrated in a certain interval and does not show a bimodal trend, it is considered that the amplitude difference distribution is relatively concentrated. If multiple obvious cluster points appear, it is judged that the difference between nodes is large. This kind of amplitude difference distribution information will be used to match the trend in the subsequent direction processing stage.

[0072] The direction correlation submodule classifies nodes into groups based on the energy consumption change direction signs of adjacent periods in the amplitude difference distribution information, and judges the consistency of the energy consumption change direction of nodes in each group based on the Pearson correlation coefficient function, thus obtaining a set of direction consistency coefficients.

[0073] After obtaining the distribution information of the node amplitude differences, the direction of energy consumption change of each monitoring node in adjacent periods is extracted one by one. First, the power change difference ΔP of each node between period t and period t-1 is calculated. t =P t -P t-1 If ΔP t A value greater than 0 corresponds to a period direction marked as +1, and vice versa as -1. A symbol sequence set is created for all direction symbols within a period. For example, the symbol sequences corresponding to the node set {N1, N2, N3} are as follows: N1: {+1, +1, -1, +1, +1}, N2: {+1, +1, -1, +1, +1}, N3: {-1, -1, +1, -1, -1}. Nodes with the same direction (N1, N2) are grouped into group G1. An improved Pearson correlation coefficient function is applied to each pair of node sequences within this group to determine the degree of directional consistency. The formula is as follows:

[0074]

[0075] The parameters are explained as follows: x i : The direction sign of node x in the i-th period, with a value of +1 or -1, unitless (dimensionless), y i : The direction sign of node y in the i-th period, also a dimensionless value. The average sign of the direction at node x represents the average trend of change across the entire sequence. The directional sign average of node y, n: number of periods, representing the sample size, n = 5 in this example, α, β: extended parameters for the degree of response to volatility, set to the same value α = β = 2.5, unitless, only order-of-magnitude adjustment factors. Order-of-magnitude adjustment factors α and β are used in the improved Pearson correlation coefficient function to control the sensitivity to outliers (i.e., data points deviating from the mean). Their settings mainly refer to the following three principles: Basic principles of data volatility analysis: In the directional sign sequence, the value is only ±1, but since the average is not an integer, the deviation value may be 0.4, -1.6, etc. This kind of discrete small sample is easily distorted in conventional L... 2 Norm calculations are dominated by extreme values. Therefore, introducing α>2 can increase the penalty effect on large offset values, reducing the impact of periods with large offset values ​​on the overall consistency assessment and enhancing the stability of the trend alignment signal. Empirical model symmetry control: Setting α=β is to avoid structural skewness in the denominator caused by unequal offset terms between the two sequences, which is especially crucial when one node is more or less than the other. Empirically, α=2.5 is between the Euclidean norm (α=2) and the L... 3 The flexible range between norms balances robustness and fluctuation preservation. Parameter selection method: The value of α = 2.5 is primarily based on the distribution of skewness and kurtosis statistics obtained from actual data experiments. Typically, in typical distribution network monitoring scenarios, the skewness range of periodic fluctuations is -0.5 to +0.5, while the kurtosis is concentrated between 2.5 and 4.5. Based on this, a relatively neutral expansion factor of 2.5 is derived, ensuring that minor abrupt changes are not excessively amplified while retaining the strong influence weight of high-frequency unidirectional fluctuation signals. Therefore, α = β = 2.5 is an empirical setting obtained through statistical induction, balancing sample stability, sequence symmetry, and algorithm convergence accuracy, ensuring that the improvement coefficient has good generalization ability and adaptability in engineering applications. Numerator: Represents the degree of synergy between the x and y sequences on the shift trend; Denominator: Generalized Lx for the x and y shift values, respectively. α L β Norm represents the strength of the fluctuation expansion of a sequence, and its unit is dimensionless.

[0076] Let the symbol sequences of nodes N1 and N2 both be {+1,+1,-1,+1,+1}, then we have: Molecular calculations: Denominator calculation: The final improved Pearson coefficient calculation is as follows:

[0077] Therefore, the directional consistency coefficient between nodes N1 and N2 within this periodic window is 0.859, and the set of directional consistency coefficients is denoted as {(N1,N2):0.859}, which will be used for subsequent master node evaluation and selection.

[0078] The master node selection submodule evaluates the amplitude difference and direction consistency value of nodes in the same group based on the set of direction consistency coefficients, selects the node with the smallest amplitude difference and the largest direction consistency value as the master node, extracts the master node number and the corresponding number list of the node group, and obtains the master node identification code and node amplitude mapping table.

[0079] Based on the set of directional consistency coefficients, the corresponding amplitude difference and directional consistency values ​​are extracted for each node combination in each group to form an evaluation input set. Each node obtains its mean directional consistency with all other nodes in the group and its own amplitude interval difference value pair. Let the node set be G1 = {N1, N2, N4}, with amplitude differences of 12W, 9W, and 10W respectively, and directional consistency values ​​as follows: N1: 0.859 with N2, 0.804 with N4, with a mean of... N2: 0.859 with N1, 0.793 with N4, with a mean of 0.826; N4: 0.804 with N1, 0.793 with N2, with a mean of 0.7985. Form a set:

[0080] The sequence {(N1,12W,0.8315),(N2,9W,0.826),(N4,10W,0.7985)} is sorted in descending order of direction consistency to obtain the order: N1→N2→N4. When N1 and N2 are similar in consistency, they are compared in ascending order of amplitude difference to select N2 as the master node. The master node number N2 and the group number G1 are extracted to construct the master node identification code: {N2,G1}. The node amplitude mapping set {(N2:9W),(N1:12W),(N4:10W)} is output as the final output result.

[0081] Please see Figure 5 The factor matching module includes:

[0082] The factor extraction submodule calls the corresponding number of the master node and extracts the associated energy consumption change vector set according to the master node identification code and node amplitude mapping table, obtains the numerical sequence and time index information corresponding to the energy consumption change vector set, and generates the master node vector feature group.

[0083] Based on the master node identification code and node amplitude mapping table, the corresponding energy consumption vector data is retrieved by sequentially calling the master node numbers. First, the master node identification code is input into the identification function as a string for encoding and matching. The system sets a one-to-one correspondence between master node codes and numbers. For example, the identification code "A203" is mapped to master node number 17 in the amplitude mapping table. This number serves as the index key to retrieve the energy consumption monitoring data corresponding to node number 17 in the distributed energy consumption database. The data source samples at a frequency of 5 minutes, recording 288 sets of data per day. Data from the most recent 48 hours is extracted to obtain a total of 576 sampled data items, which are recorded as the basic vector.

[0084] V base =[v1,v2,…,v 576 ], where the subscript j∈[1,576], and each element v j This represents the energy consumption (in kWh) recorded during the j-th 5-minute time interval. For example, at time 1, v1 = 1.20 kWh, and at time 2, v2 = 1.28 kWh. If a missing sampling point is detected, such as a certain value v... 10 For data without records, the average of adjacent values ​​is used for filling and repair. Let v9 = 1.35, v... 11 =1.25, then the repair result is v 10 = 1.30 kWh. This process iterates through all v. j The item is then filled in with missing parts. The cleaned vector V is then... base A sliding window segmentation was performed, with a window width of 24 sampling points (corresponding to a time length of 2 hours) and a step size of 6 sampling points (corresponding to a 30-minute scrolling time). Multiple overlapping sub-vectors were generated sequentially. The first group of sub-vectors is denoted as X1 = [1.20, 1.28, 1.35, 1.27, ..., 1.42], with a length of 24, corresponding to the starting time period "May 1, 2025, 08:00". The structured record is ("2025-05-01 08:00", X1). A total of 93 such sub-vectors were generated, constructing the master node feature set T. main ={(t1,X1),(t2,X2),…,(t 93 ,X 93 )}, where t k X represents the start time index of each segment. k =[x k,1 ,x k,2 ,…,x k,24 The vector represents the energy consumption change data within the corresponding time period. In addition to the original sampled values, each vector also needs to extract statistical feature information, including the maximum value, minimum value, mean, and standard deviation. These statistical results are used as auxiliary features for comparative analysis in subsequent energy consumption pattern judgment and classification model training.

[0085] The final master node feature group record structure includes: time index, original vector, and statistical features. The data structure is as follows: the time is "2025-05-01 08:00", the corresponding energy consumption sub-vector is a 24-dimensional array, the maximum value is 1.37kWh, the minimum value is 1.20kWh, the mean is 1.3075kWh, and the standard deviation is 0.0418kWh. Similarly, a set of vector sequences covering the entire time period is constructed, serving as the data foundation for subsequent similarity calculations and classification model inputs.

[0086] The feature calculation submodule calls each vector value in the main node vector feature group, calculates the Euclidean distance between the vector value and each standard factor vector in the carbon footprint factor library, and normalizes it to represent it in the form of a similarity set. The similarity set is then input into the support vector machine model for classification boundary judgment to obtain the similarity classification output set.

[0087] From the main node feature group T main Read each set of energy consumption subvectors Q sequentially j This is then matched against the standard factor vectors defined in the carbon footprint factor library. Vector Q j =[q j,1 ,q j,2 ,…,q j,n ] is the energy consumption set of the main node during the j-th time period (e.g., 08:00-10:00), q j,k This represents the energy consumption sample value (in kWh) of the k-th item in the vector, where k = 1, 2, ..., n, and typically n = 24. The standard factor vector is set as Rk. m =[r m,1 ,r m,2 ,…,r m,n ], where m = 1, 2, ..., M, r m,k This represents the reference value of the m-th factor template at the k-th sampling position. Calculate Q. j With R m The Euclidean distance between them is given by the following formula: The parameters are explained as follows: q j,k : Energy consumption of the k-th item in the j-th vector of the master node; r m,k : The standard value of the kth term in the m-th term vector of the factor template; d m : Distance between the master node vector and the m-th factor template; n: Number of sampling points in each vector group.

[0088] Simplified example with n=5: Let the principal node subvector be Q1 = [2.2, 2.4, 2.6, 2.1, 2.5] kWh, and the standard factor template be R1 = [2.1, 2.3, 2.5, 2.2, 2.4] kWh. Substitute these values ​​to calculate the Euclidean distance: Perform the above distance calculation on all M factor templates to form a distance set [d1, d2, ..., d M Then, take the reciprocal of the distance for each item to obtain the similarity set s. m : Similarity normalization uses the linear scaling method, applying it to all s m Mapped to the interval [0,1], let the normalized value be: S j =[0.92,0.86,0.79,0.75,0.68], this vector S j The feature input, serving as the sub-vector of the j-th principal node, is fed into the support vector machine (SVM) model for classification boundary determination. The SVM model employs a C-support vector classifier structure with the following parameter configurations: Input dimension: equal to the number of standard factor templates, M = 5; Kernel function: radial basis function; Kernel function expression: K(Q a Q b )=exp(-γ·||Q a -Q b || 2 The parameters are explained as follows: Q a : Input normalized similarity vector (samples to be classified), where the similarity features of the principal node sub-vectors in group a; Q b : The b-th support vector in the model's support vector set (with known label samples); ||Q a -Q b || 2 : Represents vector Q a With Q b The squared Euclidean distance between them, that is, the degree of difference between them in M-dimensional space, is calculated as follows: Where q a,m It is a vector Q a The value of q in the m-th dimension b,m It is Q b Values ​​within the same dimension; -γ: the bandwidth parameter of the kernel function, used to adjust the mapping scale of data in high-dimensional space. A larger value indicates greater sensitivity to differences; currently set to γ ​​= 0.15; exp: the natural exponential function, which exponentially compresses negative distance values, making samples with smaller distances have a greater impact on the kernel function value. Let the input vector be S. j Given a support vector V1 = [0.88, 0.83, 0.76, 0.70, 0.65], where V is the sum of squared values ​​[0.92, 0.86, 0.79, 0.75, 0.68], the difference of squares between the two vectors is ||S|. j -V1|| 2 =(0.92-0.88) 2 +(0.86-0.83) 2 +(0.79-0.76) 2+(0.75-0.70) 2 +(0.68-0.65) 2 =0.0068, substituting into the kernel function: K(S) j V1) = exp(-0.15·0.0068) = exp(-0.00102) ≈ 0.99898. The support vector machine will process all support vectors V i Perform the above kernel function operation, combined with the corresponding Lagrange multiplier α i The weighted values ​​are used to form a discriminant function for classification. The final output is the score probability for each class. Example model output:

[0089] P output =[Industrial Office: 0.92, Commercial Office Building: 0.76, Residential Community: 0.54, School: 0.38], this output indicates that the main node sub-vector is most likely to belong to "Industrial Office" among all categories. This result will be passed to the next module for label filtering and factor number extraction. Each Q j The above process will be repeated to form a similarity classification output set with consistent structure.

[0090] The category determination submodule outputs the classification boundary results in the similarity classification set, filters the nearest label of the current main node vector in the classification space, and locates the corresponding factor number and attribute subclass according to its position in the carbon footprint factor library to obtain the matching factor subclass label;

[0091] The system receives the classification result set output from the previous module and extracts the optimal label for each master node sub-vector in the classification boundary space. For each item in this process, vector probability sorting is performed, and the classification result set corresponding to each vector is arranged from highest to lowest score. For example, if a master node sub-vector has the following classification probabilities in the model output: "Industrial Office" is 0.92, "Commercial Office Building" is 0.76, "Residential Community" is 0.54, and "School" is 0.38, the item with the highest value in this result set is the nearest neighbor label, corresponding to the label "Industrial Office". The system uses this label as the preliminary classification result for the current master node sub-vector. Then, the system calls the indexing system of the carbon footprint factor library, using the label "Industrial Office" as the query keyword to retrieve all registered factor records under this label. The factor library in the system adopts a hierarchical structure. Each factor consists of a unique number, industry attribute, energy consumption range, and historical matching records. Search results under this label include factor numbers such as F001, F004, F009, etc. The system sequentially reads the attribute data of these factors and performs correspondence analysis with the feature data of the master node sub-vectors. If the average energy consumption value of the master node's sub-vectors falls within the energy consumption range of multiple factors, the vector distance between it and each factor template is further determined. Assuming the average value of the current master node's sub-vectors is 2.45 kWh, and the energy consumption ranges found for F001 are [2.1, 2.8], F004 are [1.5, 2.2], and F009 are [2.3, 2.7], then F001 and F009 simultaneously meet the range constraints. Next, their standard vector templates are called, and a vector difference evaluation is performed with the master node's sub-vectors to determine their nearest-neighbor matching factor. During the evaluation, the system calculates the numerical difference between the master node's sub-vectors and each factor template, and selects the one with the smallest difference as the final assigned factor. Its corresponding number is the matching factor number for that node. For example, if the distance between the master node's sub-vector and the F001 template is 0.18, and the distance for F009 is 0.22, then F001 is selected. The complete record of F001 is read, and the attribute subclass information, such as "Industrial Office," is extracted. This tag is used as the final attribute attribution identifier for the master node's subvector. If the master node contains multiple subvectors throughout the analysis period, all attribution tags are statistically analyzed during the final determination. The primary class tag for the master node throughout the entire period is determined based on the attribution frequency. If tags with the same frequency appear, the earliest tag is selected as the primary identifier. Furthermore, the matching factor's number, industry attribute, and identification tag are retained during the positioning process for subsequent carbon emission analysis and factor tracking. The final output structure includes fields such as master node number, subvector start time, matching factor number, and attribute subclass tag.

[0092] Please see Figure 6 The factor archiving module includes:

[0093] The tag collection submodule obtains the node group number to which the corresponding main node belongs based on the matching factor subclass tag and the main node identification code, and extracts the node group member identifier from the node amplitude mapping table to construct the node group grouping corresponding to the subclass tag and generate the node group mapping set.

[0094] Based on the matching factor subclass label and the master node identification code, the system obtains the node group number to which the corresponding master node belongs. First, the master node identification code is standardized to hexadecimal. After unifying the format, the number is searched against the node amplitude mapping table. Taking the identification code "A203" as an example, its standardized result maps to number 17. The system queries using the label "Industrial Office" combined with number 17 to extract the node group identifier corresponding to this number, for example, "N-IND-07". The system uses this node group number as the group index identifier and searches the node amplitude mapping table for the set of all member node identifiers corresponding to the number "N-IND-07". The search results show that the member node numbers in the group are [17, 18, 21, 26, 29]. The system converts all node numbers in the group into a structured identifier array, i.e., generates a node identifier array ["A203", "A204", "... [A207", "A20C", "A20F"], each number corresponds to a master node identifier. By pairing the node group number with the tag attribute of the master node, a complete set of node groups under the tag "Industrial Office" is constructed. Each tag corresponds to only one unique node group number. If different master nodes have the same tag, they are automatically grouped into the node group mapping under the same tag. The mapping result structure is a key-value pair, where the key is the subclass tag and the value is the number of all node members under that tag. For example, the node set corresponding to the tag "Industrial Office" is ["A203", "A204", "A207", "A20C", "A20F"]. Other tags such as "School" and "Commercial Office Building" are mapped and bound in the same way, finally forming a tag-node group mapping set. This set is used for node-level call logic location in the subsequent vector extraction and classification analysis process.

[0095] The vector integration submodule calls each node identifier in the node group mapping set, extracts the node change vector from the current cycle energy consumption monitoring data, and retrieves the change vector record in the power grid monitoring database according to the node identifier. It then combines the current node change vector with the retrieved change vector record in chronological order to obtain the node group vector set.

[0096] The system calls the identifier of each node in the node group mapping set, extracts the node change vector from the current period's energy consumption monitoring data, and completes the multi-node joint integration operation. First, the system enters the current period's monitoring dataset according to the node identifiers ["A203", "A204", "A207", "A20C", "A20F"]. It extracts the energy consumption change sequence recorded by each node in the current period, sampled every 5 minutes. For example, if the current period is "2025-05-05 00:00 to 2025-05-05 12:00", the corresponding number of sampling points is 144. The system extracts the corresponding change vector from the real-time monitoring database. Each node vector is uniformly in the form of a sequence structure of length 144, and each item is the actual energy consumption value of the node at a certain sampling time point. Taking "A203" as an example, its change vector is denoted as Q. A203 =[1.25,1.30,...,1.42], other nodes are extracted using the same structure, and then the system enters the power grid monitoring database to retrieve the historical change records of the corresponding nodes. It searches for the corresponding historical vectors within the same time interval, comparing them by ensuring complete consistency of timestamps. The record format is [A203:{now:[…],history:[…]}]. The system merges the current vector and historical records of each node in chronological order, combining them by concatenating the first and last elements and annotating the time field, resulting in the integrated structure {A203:[(00: The structure [00:00, 1.25), ..., (12:00, 1.42), (00:00, 1.20), ..., (12:00, 1.35)]} reflects the energy consumption sequence of the same node in two periods. After all nodes are merged, they form a node group vector set. Each node is a first-level key value item, under which is attached the current and historical dual-period complete vector sequence. The set is subjected to preliminary trend labeling processing, including the extraction of parameter values ​​such as incremental change and slope direction for subsequent storage structure use. All data maintains the consistency of timestamp alignment and node number mapping.

[0097] The structure entry module sets the structured field items as subclass label, node number, timestamp, vector value and trend code according to the node group vector set and the matching factor subclass label corresponding to the node. It writes all data into the carbon footprint factor library storage structure according to the label dimension and generates carbon footprint factor archive records as factor library structure data output.

[0098] Based on the node group vector set and the corresponding matching factor subclass labels, a unified structure field is defined, constructing a data structure with five dimensions: "Subclass Label," "Node Number," "Timestamp," "Vector Value," and "Trend Code." The system first iterates through each record in the node group vector set, extracting the associated label for each node's vector data. For example, if the label for node "A203" is "Industrial Office," then the subclass label field for all data items is assigned "Industrial Office," the node number field is filled with "A203," the timestamp field records the current or historical time point value, the vector value field is filled with the corresponding energy consumption value, and the trend code field is calculated by comparing the current value with the previous time point. The system uses the value to determine the direction of change. The positive and negative range of the difference is set as the trend reference value. If the difference is ≥ 0.05 kWh, the trend code is marked as "U"; if the difference is ≤ -0.05 kWh, it is marked as "D"; and if it is between (-0.05, 0.05), it is marked as "F". The system uses this logic to generate a trend code sequence for each node vector. A five-element field structure is added to each sampled record. The system then categorizes records by label dimension, aggregating all records belonging to "Industrial Office" and writing them into the carbon footprint factor library storage structure. The records are indexed at the directory level by label name and sorted by node number and timestamp before being stored in the database. This process constitutes the carbon footprint factor archive record, serving as the structured output data of the carbon footprint factor library for subsequent use.

[0099] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A system for dynamic sensing and carbon footprint factor database construction based on regional energy, characterized in that, The system includes: The energy consumption extraction module collects energy consumption data from regional power grid nodes, filters out energy consumption data intervals in the floating feature data where the difference is lower than the floating threshold and the direction of change has not reversed, and outputs periodic energy consumption floating data. The floating determination module judges and processes the stability of energy consumption fluctuation amplitude and direction based on the periodic energy consumption fluctuation data, constructs a set of energy consumption change vectors, and generates fluctuation level identification parameters according to the frequency of change and the amplitude ratio. The node comparison module analyzes the band differences between multiple nodes in the power grid through the fluctuation level identification parameters and performs amplitude similarity judgment. It compares the consistency of the change direction of adjacent nodes and filters the master node, and outputs the master node identification code and node amplitude mapping table. The factor matching module uses the master node identification code and the node amplitude mapping table to call the energy consumption change vector set of the master node in the node amplitude mapping table, compares it with the carbon footprint factor library, inputs it into the support vector machine model to classify the similarity index, and outputs the matching factor subclass label. The factor archiving module constructs a regional power grid carbon footprint factor library by matching factor subclass tags and generates carbon footprint factor archive records.

2. The system for dynamic sensing and carbon footprint factor database construction based on regional energy as described in claim 1, characterized in that, The periodic energy consumption fluctuation data includes a time series number, the start and end times of the fluctuation interval, and the number of data points within the interval; the energy consumption change vector set specifically includes the coordinates of the maximum change point, the coordinates of the minimum change point, and the change time span; the fluctuation level identification parameters include the change frequency level, the unit time amplitude ratio level, and the directional trend stability level; the master node identification code includes a unique node number, a fluctuation cycle identification label, and a master node status mark; the node amplitude mapping table specifically includes the amplitude difference corresponding to the node number, the change direction symbol, and the directional correlation coefficient; the matching factor subclass label specifically refers to the factor subclass number, the band category, and the node source type identifier; the carbon footprint factor archive record includes the subclass label, node number, timestamp, vector value, and change trend code.

3. The system for dynamic sensing and carbon footprint factor database construction based on regional energy as described in claim 1, characterized in that, The energy consumption extraction module includes: The energy consumption acquisition submodule collects time-series energy consumption data from all monitoring nodes in the regional power grid, records it according to a set period, obtains the original power value sequence of the nodes in a continuous period, and calls the time period number, acquisition timestamp and energy consumption of each group of nodes to generate a continuous period energy consumption sequence set. The floating construction submodule calculates the energy consumption difference based on the energy consumption of adjacent data points in the continuous periodic energy consumption sequence set and determines whether the change direction is consistent. Data pairs with energy consumption differences lower than the floating threshold and the same direction are marked as floating sequences, and a floating feature segment sequence table is constructed according to the sliding window order. The floating threshold is set by calculating the skewness coefficient of the energy consumption difference distribution within a continuous period and combining it with an empirical benchmark. The interval screening submodule determines whether the data pairs meet the requirements of difference and direction conditions in three out of five consecutive sequences based on the duration and number of data pairs of each floating segment in the floating feature segment sequence list. It then extracts the floating segments that meet the conditions and marks the start and end times of the data pairs and the corresponding node numbers to obtain the periodic energy consumption floating data.

4. The system for dynamic sensing and carbon footprint factor database construction based on regional energy as described in claim 3, characterized in that, The floating determination module includes: The trend smoothing submodule, based on the periodic energy consumption fluctuation data, obtains the continuous energy consumption in each data sequence and sorts it in fixed period units. It performs continuous interpolation on adjacent energy consumption in the sorting results and performs fitting processing through the Savitzky-Golay filtering algorithm with the energy consumption change direction as the reference standard to generate a smooth trend value sequence. The segment extraction submodule detects the location of inflection points in the smooth trend value sequence and extracts the energy consumption corresponding to the inflection points. It marks the minimum value before and after the first inflection point and the maximum value before and after the last inflection point as the segment boundary, calculates the start and end times and the distance between points of the data within the segment boundary, and performs combined processing to obtain the energy consumption change vector set. The level determination submodule calls the point spacing, maximum and minimum value difference and number of inflection points in the sequence of each segment in the energy consumption change vector set, calculates the ratio of change frequency to energy consumption amplitude within a specified time, maps the results to the level classification standard for interval matching and marks the corresponding level identifier to obtain the fluctuation level identifier parameter.

5. The system for dynamic sensing and carbon footprint factor database construction based on regional energy as described in claim 4, characterized in that, The node comparison module includes: The amplitude comparison submodule obtains the minimum and maximum fluctuation values ​​of each node within the same continuous period based on the fluctuation level identifier parameter, calculates the amplitude interval difference corresponding to the node, and performs standard deviation calculation and error distribution judgment on all nodes based on the amplitude interval difference result to obtain amplitude difference distribution information. The direction correlation submodule classifies nodes into groups based on the energy consumption change direction signs of adjacent periods in the amplitude difference distribution information, and judges the consistency of the energy consumption change direction of nodes in each group based on the Pearson correlation coefficient function, thus obtaining a set of direction consistency coefficients. The master node filtering submodule performs a joint evaluation of the amplitude difference and directional consistency value of nodes in the same group based on the set of directional consistency coefficients, selects the node with the smallest amplitude difference and the largest directional consistency value as the master node, extracts the master node number and the corresponding number list of the node group, and obtains the master node identification code and the node amplitude mapping table.

6. The system for dynamic sensing and carbon footprint factor database construction based on regional energy as described in claim 5, characterized in that, The factor matching module includes: The factor extraction submodule calls the corresponding number of the master node and extracts the associated energy consumption change vector set according to the master node identification code and node amplitude mapping table, obtains the numerical sequence and time index information corresponding to the energy consumption change vector set, and generates the master node vector feature group. The feature calculation submodule calls each vector value in the main node vector feature group, calculates the Euclidean distance between the vector value and each standard factor vector in the carbon footprint factor library, and normalizes it to represent it in the form of a similarity set. The similarity set is then input into the support vector machine model for classification boundary judgment to obtain the similarity classification output set. The category determination submodule selects the nearest neighbor label of the current main node vector in the classification space based on the classification boundary results in the similarity classification output set, and locates the corresponding factor number and attribute subclass according to the position of the nearest neighbor label in the carbon footprint factor library to obtain the matching factor subclass label.

7. The system for dynamic sensing and carbon footprint factor database construction based on regional energy as described in claim 6, characterized in that, The factor archiving module includes: The tag collection submodule obtains the node group number to which the corresponding main node belongs based on the matching factor subclass tag and the main node identification code, extracts the node group member identifier from the node amplitude mapping table, constructs the node group group corresponding to the subclass tag, and generates the node group mapping set. The vector integration submodule calls each node identifier in the node group mapping set, extracts the node change vector from the current cycle energy consumption monitoring data, and retrieves the change vector record in the power grid monitoring database according to the node identifier. The current node change vector is combined with the retrieved change vector record in chronological order to obtain the node group vector set. The structure entry module sets the structured field items as subclass label, node number, timestamp, vector value and trend code according to the node group vector set and the matching factor subclass label corresponding to the node. It writes all data into the carbon footprint factor library storage structure according to the label dimension and generates carbon footprint factor archive records as factor library structure data output.

Images