A data fusion processing method of an intelligent fusion terminal

Abstract

The application relates to a data fusion processing method of an intelligent fusion terminal and belongs to the technical field of power system data processing. The method comprises the following steps: collecting multi-source data to construct current, voltage and electric quantity time sequence; calculating target difference values and mutation degree evaluation values of each data point, thereby adaptively segmenting each dimension and dividing sub-sequences; calculating the range and information entropy of each sub-sequence, obtaining quality scores through negative exponential mapping and weighted summation, and normalizing to calculate the fusion weight of each dimension at each time; searching and determining a reference value based on a zero electric quantity change time, and calculating a power usage degree index of each dimension; weighting and summing the fusion weight and the power usage degree index to obtain fusion data, and adaptively adjusting the fusion frequency according to the time length and average power consumption of the power consumption interval. The application realizes dynamic distribution of the fusion weight through adaptive segmentation, realizes adaptive adjustment of the fusion frequency through power consumption interval sensing, and effectively improves the data fusion accuracy and the utilization efficiency of terminal computing resources.

Description

Technical Field

[0001] This invention belongs to the field of power system data processing technology, and specifically relates to a data fusion processing method for an intelligent fusion terminal. Background Technology

[0002] With the rapid advancement of smart grid construction, the smart fusion terminal in the distribution area, as a core device on the edge of the power system, undertakes multiple tasks such as multi-source data acquisition, fusion processing, and intelligent control. Multi-source data fusion technology, by integrating multi-dimensional data such as current, voltage, and power from devices like smart meters, low-voltage circuit breakers, reactive power compensation devices, and charging piles, can overcome the limitations of single data sources and provide more comprehensive and accurate information support for load forecasting, equipment status monitoring, and fault early warning.

[0003] To address the issue of multi-source data fusion, existing technology CN114168268B discloses a smart distribution area data acquisition and fusion method based on container technology, which achieves data fusion through a unified information model. However, it uses fixed-weight fusion and cannot respond to dynamic changes in data quality. Existing technology CN116304950B discloses a multi-source heterogeneous data fusion method for distribution networks, which achieves fusion through anomaly analysis using neural networks combined with time registration and interpolation. However, neural network training relies on a large number of samples and incurs high computational costs. Existing technology CN117556377B discloses a multi-source data fusion processing method for station automation terminals, which uses mutation degree assessment to perform fusion. The method segments and uses PCA for dimensionality reduction, but only processes a single data sequence and does not involve multi-dimensional collaborative fusion weight allocation; the existing technology CN118117742A discloses a data processing method for a smart fusion terminal for distribution areas, which obtains the classification results of distribution areas through a convolutional neural network to decide the data processing method, but it relies on a black box model and cannot accurately quantify the differences in data quality of each dimension; the existing technology CN120408534B discloses a concentrator smart fusion terminal based on multi-source fusion data regulation, which obtains power quality scores and fusion weights through the fluctuation characteristics of a fixed-length sequence, but the fixed window cannot adapt to the dynamic transformation of data change patterns and does not involve adaptive adjustment of the fusion frequency.

[0004] In summary, existing technologies for multi-source data processing in intelligent fusion terminals urgently need to address technical issues such as the inability of fusion weights to dynamically and adaptively adjust with data quality, the inability of fixed data quality assessment granularity to adapt to changing modes, reliance on black-box models with high computational overhead, and the waste of computational resources due to fixed fusion frequency. Summary of the Invention

[0005] The purpose of this invention is to provide a data fusion processing method for intelligent fusion terminals to solve the above-mentioned problems in the prior art. The specific technical solution is as follows: A data fusion processing method for an intelligent fusion terminal includes the following steps: Time series data of current, voltage and power consumption of the smart converged terminal in the distribution area are collected. The abrupt change evaluation value of each data point in each data dimension is calculated. Data points with a abrupt change evaluation value greater than a preset threshold in each data dimension are selected as candidate segmentation nodes. After screening the candidate segmentation nodes in each data dimension, a segmentation node set for each dimension is formed. The original data sequence of the corresponding dimension is divided into several subsequences with the position of each node in the segmentation node set as the boundary, so that each data dimension forms its own independent subsequence set. The data quality score for each data dimension is calculated separately for each subsequence, and the data quality score is used to characterize the stability of changes in data within the subsequence. For any data fusion moment, the subsequence to which the moment belongs in each data dimension is determined, and the data quality score of the subsequence to which each dimension belongs is used as the real-time credibility representation of the corresponding dimension. The real-time credibility representation of each dimension is normalized, and the normalized ratio is used as the dynamic fusion weight of each dimension at the current moment. The deviation of the measured value of each data dimension from its respective reference value at the current moment is calculated, and the deviation is mapped to the interval [0,1) through a monotonically increasing function to obtain a normalized energy use index. The energy usage index is weighted and summed using the dynamic fusion weights of each dimension to obtain the fusion data at the current moment; Based on historical load data, the entire day is divided into multiple power consumption intervals. The data fusion frequency for each power consumption interval is calculated based on the duration and average power consumption of each interval. Within the current power consumption interval, the fusion data calculation for subsequent times is triggered at equal intervals according to the corresponding data fusion frequency.

[0006] As a preferred embodiment of this application, the step of calculating the mutation degree evaluation value of each data point in each data dimension includes: for the k-th data point in any data dimension, calculating the absolute value of the difference between the data point and its previous data point as a first difference value, calculating the absolute value of the difference between the data point and its next data point as a second difference value, and taking the larger of the first difference value and the second difference value as the target difference value of the data point; taking the data point as the center, taking three consecutive data points to the left and three consecutive data points to the right as neighboring data points, calculating the arithmetic mean of the target difference values ​​of all neighboring data points as the neighborhood difference mean; calculating the absolute value of the difference between the target difference value of the data point and the neighborhood difference mean, and using the result of minimum-maximum normalization as the mutation degree evaluation value of the data point.

[0007] As a preferred embodiment of this application, the step of filtering candidate segmentation nodes for each data dimension to form a segmentation node set for each dimension includes: determining the time interval between adjacent candidate segmentation nodes in chronological order; if the time interval is less than a preset time interval threshold, retaining the node with the larger mutation degree evaluation value and deleting the node with the smaller mutation degree evaluation value, until the time interval between all adjacent nodes is not less than the preset time interval threshold, and the retained nodes are used as segmentation nodes; adding the first endpoint and the last endpoint of the data sequence to the segmentation nodes to form a segmentation node set.

[0008] As a preferred embodiment of this application, the step of calculating the data quality score for each data dimension on a subsequence basis includes: for any subsequence, calculating the difference between the maximum and minimum values ​​in the subsequence as the range, and taking a negative exponentiation of the range to obtain a first stability index; calculating the distribution information entropy of the absolute value of the difference in the ordinate between adjacent data points in the subsequence, and taking a negative exponentiation of the information entropy to obtain a second stability index; and weighted summing the first stability index and the second stability index, and using the resulting value as the data quality score for the subsequence.

[0009] As a preferred embodiment of this application, the step of calculating the deviation of the measured value of each data dimension at the current moment from its respective reference value includes: for the current dimension and voltage dimension, searching for the nearest moment in the power data sequence before and after the current moment when the power change is zero, using the arithmetic mean of the corresponding dimension data within the interval between the two zero-change moments as the reference value; if the search fails, using the arithmetic mean of the corresponding dimension data during the historical stable period as the reference value; for the power dimension, the reference value is fixed at zero; calculating the absolute value of the difference between the measured value at the current moment and the reference value to obtain the deviation.

[0010] As a preferred embodiment of this application, mapping the degree of deviation to the interval [0,1) using a monotonically increasing function specifically involves substituting the degree of deviation into a transformation function. , and This yields a normalized index of electricity usage; among which, , , These are indicators of energy usage in terms of current, voltage, and quantity, respectively. This is the difference between the measured current value and the reference value. The difference between the measured voltage value and the reference value. This is the actual measured power level; the reference power level is 0.

[0011] As a preferred embodiment of this application, after normalizing the real-time credibility representation of each dimension, the method further includes: performing a moving average filter on the original dynamic fusion weights of each dimension calculated at multiple consecutive data fusion times, and using the filtered values ​​as the final dynamic fusion weights of each dimension at the current time.

[0012] As a preferred embodiment of this application, the step of calculating the data fusion frequency of each power consumption interval based on the duration and average power consumption of each power consumption interval includes: for any power consumption interval, obtaining its duration T and average power consumption E, and calculating the time index. Calculate electricity consumption index ,like If the value is negative, it is set to zero, and the average of the time index and the electricity consumption index is taken as the frequency weight. The frequency weights are then combined with a preset base fusion frequency. Multiply to obtain the data fusion frequency for that power consumption area. .

[0013] As a preferred embodiment of this application, the step of triggering the calculation of fused data at equal intervals according to the corresponding data fusion frequency within the current power consumption zone specifically includes: starting from the current moment, generating a linked list of fused time points for the next hour according to the fusion interval corresponding to the data fusion frequency; the system task scheduler triggering the calculation of the weighted summation to obtain the fused data once when the system clock reaches any fused time point in the linked list, and deleting that time point from the linked list; automatically supplementing and generating subsequent time points when the remaining nodes in the linked list are less than 10; and clearing the current linked list and regenerating it when the system crosses the boundary of the power consumption zone.

[0014] As a preferred embodiment of this application, after collecting the time-series data of current, voltage and power of the intelligent fusion terminal in the collection area, the method further includes performing three cleaning operations on the original data: physical range verification, linear interpolation to complete missing values, and jump suppression based on the middle value of the sliding window.

[0015] The beneficial effects of this invention are as follows: 1) This invention independently calculates the target difference value and abrupt change assessment value for each data point in the three dimensions of current, voltage, and quantity. Based on this, segment nodes are selected and merged, so that the segment boundaries accurately correspond to the turning points of the change patterns of each dimension. Since the change patterns of the three dimensions are different, the number of segment nodes and subsequences are also different. Each dimension is divided into subsequences with consistent internal characteristics at the most suitable granularity, avoiding the problem of different change trends masking each other within a fixed window, and providing an accurate statistical basis for quality scoring.

[0016] 2) This invention calculates the range and information entropy of each subsequence, characterizing data quality from the magnitude and orderliness dimensions respectively. After negative exponential mapping and weighted summation, the quality score of each subsequence is obtained. At any fusion time, the fusion weight of each dimension is calculated by normalizing the scores of the subsequences belonging to each of the three dimensions at that time. Since the segmentation nodes of the three dimensions are different, and different subsequences belong to different dimensions at the same time, the weight allocation can reflect the data quality differences of each dimension in a local time interval in real time, realizing fine-grained dynamic adjustment of the weights.

[0017] 3) This invention divides a day into three power consumption zones—high, medium, and low—based on historical load data. Time and power consumption indicators are calculated for each zone, and the average of these two indicators is multiplied by a base frequency to obtain the fusion frequency for that zone. In the high power consumption zone, key change information is captured intensively at a higher frequency, while in the low power consumption zone, redundant calculations are reduced at a lower frequency. The terminal performs fusion calculations only at the times specified in the scheduling table; during non-fusion times, it only collects and stores data. This effectively reduces processor load and power consumption while ensuring the integrity of information during critical periods.

[0018] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Of course, implementing any product or method of this application does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a general flowchart of the data fusion processing method provided in the embodiments of the present invention.

[0021] Figure 2 This is a flowchart of multi-source acquisition and time-series sequence construction provided in an embodiment of the present invention.

[0022] Figure 3 This is a flowchart of the evaluation of difference values ​​and mutation degree provided in the embodiments of the present invention.

[0023] Figure 4 This is a flowchart of the adaptive segmentation of subsequences provided in an embodiment of the present invention.

[0024] Figure 5 This is a flowchart of the subsequence stability and quality scoring provided in an embodiment of the present invention.

[0025] Figure 6 This is a flowchart of the fusion weight calculation provided in an embodiment of the present invention.

[0026] Figure 7 This is a flowchart of the calculation of the power usage index provided in the embodiments of the present invention.

[0027] Figure 8 This is a flowchart of weighted fusion and frequency adaptive adjustment provided in the embodiments of the present invention.

[0028] Figure 9 This is an architecture diagram of the data fusion processing system provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Example 1 This invention provides a data fusion processing method for intelligent fusion terminals. For example... Figure 1 As shown, the specific steps include: S101, Multi-source acquisition and time series construction This step aims to acquire multi-source raw data collected by the smart converged terminals in the distribution area and construct time-series sequences across various dimensions, providing a data foundation for subsequent feature extraction and segmented analysis. For example... Figure 2 As shown, in this step, the smart converged terminal in the distribution area establishes communication links with various devices within the distribution area, such as smart meters, low-voltage circuit breakers, reactive power compensation devices, and charging piles, through its integrated multiple data acquisition interfaces. These communication links adopt power industry standard protocols such as DL / T 645, Modbus, or IEC 61850, continuously receiving raw measurement data uploaded by each device at a sampling frequency of once per second. The communication coprocessor inside the smart converged terminal parses the received data frames into physical quantity values, including the effective values ​​of three-phase current, three-phase voltage, and cumulative active power. Before each data point is stored in the memory buffer, a millisecond-level timestamp generated by the high-precision real-time clock inside the smart converged terminal is appended.

[0031] The intelligent fusion terminal allocates a circular queue in memory for each data dimension, with a fixed queue depth of 86,400 data points, to store complete sampling records for the most recent 24 hours. The data structure of each queue is a key-value pair array, and each array element contains a timestamp field and a numeric field, with the numeric field being a single-precision floating-point type.

[0032] Before writing each newly arrived data point into the corresponding queue, the intelligent fusion terminal performs three data cleaning steps.

[0033] The first data cleaning step is physical range verification. The intelligent fusion terminal's internal non-volatile memory has preset upper and lower limits for reasonable values ​​in each data dimension. For current data, the upper limit is set to 1.5 times the rated current, and the lower limit is set to 0 amperes. For voltage data, the upper limit is set to 1.2 times the rated voltage, and the lower limit is set to 0.5 times the rated voltage. For electricity data, the upper limit is set to twice the rated daily electricity consumption, and the lower limit is set to 0 kilowatt-hours. The intelligent fusion terminal compares the value of a newly arrived data point with the corresponding upper and lower limits for each dimension. If the value exceeds these limits, the data point is determined to be invalid, discarded, and replaced with the valid value from the previous sampling time for that dimension. When performing the replacement operation, the intelligent fusion terminal reads the data value from the tail of the corresponding circular queue from the previous time, copies it, and writes it to the current time position.

[0034] The second data cleaning step is missing value completion. The intelligent fusion terminal has an internal sampling timeout counter. This counter increments at the end of each sampling period and resets to zero upon successful receipt of valid data. When the counter value exceeds 1, it indicates consecutive missing values. The intelligent fusion terminal records the timestamp of each missing value. After successfully receiving two valid data points from the preceding and following points, it calculates an estimate of the missing value at the intermediate time using linear interpolation. The linear interpolation formula is: Missing value estimate = Previous valid value + (Later valid value - Previous valid value) × (Missing time - Previous valid time) / (Later valid time - Previous valid time). The intelligent fusion terminal then fills the calculated estimate into the circular queue at the position corresponding to the missing value.

[0035] The third data cleaning step is jump suppression. The intelligent fusion terminal maintains a sliding window of length 5, which stores the values ​​of the current data point and the four data points preceding it. The intelligent fusion terminal sorts the five values ​​within the sliding window and takes the median of the sorted values ​​as the baseline. The intelligent fusion terminal calculates the absolute value of the difference between the current data point and the median. If this absolute value is greater than a preset jump threshold, the current data point is determined to be an abnormal jump point and replaced with the median. The jump threshold is set to 20% of the rated current for the current dimension, 10% of the rated voltage for the voltage dimension, and 30% of the average energy of the most recent 10 sampling points for the energy dimension.

[0036] After the above cleaning process, the valid data from each dimension are queued sequentially according to time order. The intelligent fusion terminal reads the data queue at the current moment into three independent time sequences, namely the current data sequence. Voltage data sequence and power data sequence , where n is the total number of data points collected at the current time. The subscripts k (k=1,2,...,n) of these three sequences are strictly aligned to the same acquisition time. That is, the kth current data kth voltage data and the kth power data All correspond to the same acquisition time .

[0037] S102. Assessment of Difference Values ​​and Mutation Level This step aims to calculate the degree of local mutation for each data point in the time series of each data dimension, in order to quantify the drastic change of each point relative to its neighborhood, and provide a quantitative basis for subsequent adaptive segmentation. For example... Figure 3 As shown, this step independently calculates the degree of local mutation for each data point in the three data sequences obtained in S101, aiming to identify potential locations where the change pattern changes from each sequence.

[0038] Taking current data sequences as an example, the intelligent fusion terminal first calculates a target difference value for each data point in the current sequence. For the k-th data point within the sequence... (where 2 ≤ k ≤ n-1), the intelligent fusion terminal reads the previous data point from the current data sequence. and the next data point The value is then used to calculate the first difference value. This is the absolute value of the difference between the ordinates of the current point and the previous point; simultaneously, the second difference value is calculated. This refers to the absolute value of the difference between the ordinates of the current point and the next point. The intelligent fusion terminal will... and The input comparator takes the larger of the two values ​​as the target difference value for that data point. For the first data point of the sequence The intelligent fusion terminal only calculates its connection with... The absolute value of the difference between them is used as For the data points at the end of the sequence Since there are no subsequent data points, the intelligent fusion terminal only calculates its connection with... The absolute value of the difference between them is used as .

[0039] The intelligent fusion terminal then constructs a neighborhood window for each data point and calculates the mean of neighborhood differences. The half-width parameter 'a' of the neighborhood window is set to 3. For the k-th data point, the intelligent fusion terminal extracts the three consecutive data points to its left (i.e., ...) from the current data sequence. , , ) and the three consecutive data points on the right (i.e. , , The set of neighboring data points constitutes the set of neighboring data points for that data point. When a data point is near the beginning of the sequence and there are fewer than 3 neighboring data points on the left (i.e., k value is less than 4), the intelligent fusion terminal only takes the actual left-side data points; when a data point is near the end of the sequence and there are fewer than 3 neighboring data points on the right (i.e., k value is greater than n-3), the intelligent fusion terminal only takes the actual right-side data points. The intelligent fusion terminal records the total number of neighboring data points. Then, the target difference values ​​of these neighboring data points are read from the cache. This constitutes a neighborhood difference set.

[0040] The intelligent fusion terminal calculates the arithmetic mean of the set of differences in the neighborhood. Then, the target difference value of the current data point. The initial mutation severity assessment value is obtained by subtracting the average value and taking the absolute value. .

[0041] To eliminate the incomparability caused by differences in units and amplitudes between different data sequences, the intelligent fusion terminal also needs to normalize the initial abrupt change assessment values ​​of all data points within the same sequence. The intelligent fusion terminal first traverses the entire current sequence to identify all... The maximum value in and minimum value Then for each According to the minimum-maximum normalization formula The final mutation severity assessment value δ(k) is calculated. When equal When the mutation level is equal, it indicates that all points in the sequence have the same degree of mutation, and the intelligent fusion terminal sets all δ(k) to 0. The normalized mutation level evaluation value δ(k) is strictly limited to the closed interval [0,1].

[0042] The intelligent fusion terminal independently repeats all the above calculation processes for the voltage data sequence and the power data sequence to obtain the abrupt change assessment value of each data point in the voltage sequence and the abrupt change assessment value of each data point in the power sequence, respectively.

[0043] S103. Adaptively segment and divide subsequences based on mutation severity assessment values. This step aims to adaptively segment the complete time-series sequence of each data dimension into several subsequences with consistent internal variation characteristics using the mutation degree assessment values ​​of each data point, as the basic unit for subsequent quality assessment. For example... Figure 4As shown, this step uses the mutation degree evaluation value of each data point obtained in S102 to perform adaptive segmentation on each data dimension, and determines the points with evaluation values ​​higher than the set threshold as segmentation nodes, while avoiding nodes being too dense in local areas.

[0044] Taking current data sequences as an example, the intelligent fusion terminal first sets a threshold for the degree of abrupt change. Starting from the first data point of the current sequence, the algorithm iterates through the sequence and, for each data point, reads its abrupt change assessment value. and will and Compare. If The intelligent fusion terminal records the index k of the data point into a list of nodes to be determined. After traversal, the list of nodes to be determined... The list contains the position indices of all data points that meet the conditions in the original sequence, where r is the total number of data points that meet the conditions, and the indices in the list are arranged in ascending order.

[0045] The intelligent fusion terminal then performs deduplication and merging. The intelligent fusion terminal starts from the first index of the list of nodes to be determined. Begin by calculating the second index. and The time interval between the two indices is obtained by comparing their respective timestamps. The intelligent fusion terminal has a preset time interval threshold ΔT. and If the time interval is less than ΔT, the intelligent fusion terminal determines that the two nodes are too close and should retain only one of them. At this point, the intelligent fusion terminal reads and compares... Evaluation value of the degree of mutation of location data points and Evaluation value of the degree of mutation of location data points The system retains the indexes with higher mutation severity assessment values ​​and removes the indexes with lower mutation severity assessment values ​​from the list of nodes to be determined. The intelligent fusion terminal then uses the retained indexes to continue working with the next index in the list of nodes to be determined. The same time interval judgment is performed. If the time interval between two adjacent indices is greater than or equal to ΔT, the intelligent fusion terminal retains both indices and moves the comparison pointer to the next index, continuing the comparison with the index after that. The intelligent fusion terminal repeats the above iterative process until the time interval between all adjacent indices in the list of nodes to be determined is not less than ΔT.

[0046] To ensure the integrity of the segmentation, the intelligent fusion terminal also needs to forcibly add the first index 1 and the last index n of the current data sequence to the list of nodes to be determined. The intelligent fusion terminal performs a merge, inserting index 1 and index n into the list of nodes to be determined, and simultaneously reordering the list of nodes to be determined in ascending order. The final list is the set of segment node indices for this data dimension, denoted as […]. ,in , .

[0047] After obtaining the set of segmented node indices, the intelligent fusion terminal performs sequence partitioning. The intelligent fusion terminal starts from the first node index. Begin reading and All consecutive indices between, i.e. , , , ..., The data points corresponding to these indices are extracted from the current sequence to form the first subsequence. The terminal continues reading. and The data points corresponding to all consecutive indices between them form the second subsequence. The terminal processes them sequentially until... and The interval between [m-1] is defined. After the above operations, the entire original current sequence is divided into (m-1) subsequences. Adjacent data points within each subsequence are continuous in time index, and the subsequences are connected end-to-end through segment node indices. The intelligent fusion terminal assigns a number to each subsequence, starting from 1 and incrementing, and records the start and end indices of each subsequence.

[0048] The voltage data sequence and the power data sequence each independently perform all the operations described above, from threshold comparison to sequence partitioning, to obtain their respective set of segment node indices and subsequence partitioning results.

[0049] S104. Calculate the stability index and quality score of each subsequence. This step aims to calculate a stability index for each subsequence from two dimensions: the magnitude of amplitude change and the orderliness of data changes. The resulting comprehensive data quality score serves as the basis for subsequent weight allocation in the fusion process. Figure 5 As shown, this step calculates the stability index and quality score for each subsequence obtained in S103. Taking a subsequence in the current dimension as an example, let this subsequence be... , where L is the number of data points contained in the subsequence.

[0050] The intelligent fusion terminal first calculates the first stability index of the subsequence, which reflects the drastic degree of change in the overall amplitude of the subsequence. The intelligent fusion terminal then reads the first data point from the subsequence X. Store its value in a temporary variable. and Then the intelligent fusion terminal from Begin reading the remaining data points one by one. For each data point read... (j=2,3,...,L), compare the value of this data point with... If a comparison is made, Then Updated to Compare the value of this data point with... If a comparison is made, Then Updated to After the traversal is complete, the intelligent fusion terminal performs a subtraction operation. The range R is obtained. The intelligent fusion terminal performs a negative exponential mapping on the range R, calling the exponential function in the math library to calculate... exp() is an exponential function with the natural constant e as its base. When R=0, Stab1=1, indicating that the data has no amplitude fluctuations and has the highest stability. As R increases, Stab1 will monotonically decrease and always be greater than 0, but it will quickly approach 0, indicating that the stability will decrease rapidly.

[0051] The intelligent fusion terminal then calculates a second stability index for the subsequence, which reflects the degree of orderliness of data changes within the subsequence. The intelligent fusion terminal starts from the second data point of subsequence X. Begin by reading the current data points one by one. and the previous data point (k=2,3,...,L), perform subtraction and take the absolute value. ,Will Store the difference values ​​sequentially into the set. The length of set D is L-1. The intelligent fusion terminal performs statistical analysis on the elements in D, iterating through each element in D, counting the frequency of each different value (count(u)), and then calculating the probability of each value. The intelligent fusion terminal then calculates according to the definition of information entropy. The closer the H value is to 0, the more likely the vast majority of the discrepancies are the same, indicating a highly consistent pace of data change and good orderliness. Conversely, a larger H value indicates a greater variety and more dispersed distribution of discrepancies, suggesting stronger randomness in data change and potential influence from noise or irregular fluctuations. The intelligent fusion terminal then performs a negative exponential mapping on H, i.e. This transforms orderliness into a positive indicator: when H=0, Stab2=1, and as H increases, Stab2 monotonically decreases and approaches 0.

[0052] After obtaining the first stability index Stab1 and the second stability index Stab2, the intelligent fusion terminal reads the preset weighting coefficients from the configuration register. and Sum the two values ​​with weights. The calculation result W is stored in a temporary register. A larger W indicates more stable and ordered data changes in the subsequence, and higher data quality. The intelligent fusion terminal directly uses W as the data quality score Q for the subsequence, i.e., Q=W, and binds this score with the subsequence's dimension identifier (current, voltage, or quantity), start index, and end index, writing it to the quality score cache in key-value pairs. Each record in the cache contains four fields: dimension type, start index, end index, and quality score value.

[0053] Each subsequence in the current dimension calculates its own quality score according to the above process. All subsequences in the voltage dimension and all subsequences in the quantity dimension also independently undergo the exact same calculation process to obtain their respective quality scores.

[0054] S105. Calculate the fusion weights of each dimension at each time step based on the quality score. This step aims to dynamically determine the weight ratio of the three dimensions—current, voltage, and quantity—in the fusion calculation at any given fusion time using the quality scores of each subsequence, so that high-quality data dimensions contribute a larger share to the fusion result. For example... Figure 6 As shown, this step uses the quality scores of each subsequence obtained in S104 to dynamically calculate the fusion weights of the three dimensions of current, voltage, and energy for any time t when data fusion is required.

[0055] The intelligent fusion terminal first determines the subsequence number corresponding to time t in the current dimension. It then reads the start and end indices of each subsequence from the subsequence index table in the current dimension, and calculates the corresponding sequence position index based on time t. ,in This represents the start time of the sequence. The intelligent fusion terminal compares this position index with the start and end indices of each sub-sequence to find the one that satisfies the condition. Record the number of that subsequence. The intelligent fusion terminal then identifies the "current" and sub-sequence number by dimension. Using the search criteria, a retrieval is performed in the quality score cache to retrieve the corresponding quality score. Record this value as .

[0056] The intelligent fusion terminal uses the same method described above to search for the subsequence number containing time t in the subsequence index table of the voltage dimension. Read the corresponding quality score from the quality score cache. , recorded as The intelligent fusion terminal then searches the subsequence index table in the power dimension for the subsequence number containing time t. Read the corresponding quality score , recorded as .

[0057] After obtaining quality scores from three dimensions, the intelligent fusion terminal performs normalized summation. The intelligent fusion terminal first calculates the total score. Intelligent converged terminal judgment Is it equal to 0? If If it is not equal to 0, calculate separately. , , ;like If the value is 0, then force the command. The calculated three weights satisfy... .

[0058] To suppress drastic jumps in weight values, the intelligent fusion terminal performs a smoothing process. Internally, the intelligent fusion terminal maintains a weight history table, storing the original weight values ​​calculated over the most recent M time points. After calculating the original weight for the current time point, the intelligent fusion terminal reads the weight values ​​from the previous M-1 time points from this weight history table (if there are fewer than M-1 records, all existing records are used). For each dimension, it calculates the arithmetic mean of the current weight and the weights from the previous M-1 time points, and uses this average as the final smoothed weight for that dimension at time t. After smoothing, the intelligent fusion terminal writes the original weight for the current time point into the weight history table and deletes the oldest record to keep the table size below M. The intelligent fusion terminal then sets the smoothed weight values... , , Send to the fusion computing module.

[0059] S106. Calculate the energy usage indexes for each dimension. This step aims to convert the measured values ​​of current, voltage, and energy, which have different dimensions, into a unified, dimensionless energy usage index, so that the three dimensions can participate in subsequent weighted fusion on the same scale. For example... Figure 7 As shown, this step independently calculates the energy usage index at time t for three dimensions: current, voltage, and energy quantity. This energy usage index is a dimensionless quantity with a value range of [0,1].

[0060] For the current dimension, the intelligent fusion terminal first determines the current reference value at the current time t. The intelligent fusion terminal reads all sampling points before time t from the power data sequence, and then iterates backward through the sampling points from time t. For each sample point k reached, the power difference is calculated. Intelligent converged terminals will Compared with 0, when it first appears When the sampling point is recorded, the time is recorded as follows: Stop searching forward. If no match is found within the 1-hour time window before time t. If the sampling point is not found, the search will be terminated and the sampling point will be... Marked as invalid. The intelligent fusion terminal then iterates through the sampling points one by one from time t, searching using the same method. The moment was recorded as If it is not found within a 1-hour time window after time t, then Marked as invalid.

[0061] like and All values ​​are valid; the intelligent fusion terminal extracts the time interval from the current data sequence. Calculate the arithmetic mean of all current data points within the range as the current reference value. .like and If invalid values ​​are found, the intelligent fusion terminal will use the average current value from historical stable periods as the reference value. The historical stable period is determined as follows: the intelligent fusion terminal selects the longest continuous period of current data from the past 24 hours, where the load fluctuation range does not exceed 5% of the average load during that period. The arithmetic mean of the current within this period is then calculated and used as the current reference value. .

[0062] Obtain reference value Then, the intelligent fusion terminal reads the measured current value I(t) at the current time t, performs a subtraction operation, and takes the absolute value. The intelligent fusion terminal substitutes ΔI(t) as the independent variable into the transformation function. The constant 1 in the denominator is used to prevent division by zero errors. This is the difference between the measured current value and the reference value. The intelligent fusion terminal will calculate the result. The data is stored in a register as an indicator of the degree of electrical energy usage in the current dimension.

[0063] For the voltage dimension, the intelligent fusion terminal uses the exact same method as for the current dimension to determine the voltage reference value. That is, first search in the power data sequence. and If valid, the arithmetic mean of the voltage data within the corresponding time interval is taken as the reference value; if invalid, the average voltage value during historical stable periods is taken as the reference value. The intelligent fusion terminal reads the measured voltage value V(t) at the current time t and calculates... Then through Obtain the energy usage index in the voltage dimension ;in, This is the difference between the measured voltage value and the reference value.

[0064] Regarding power consumption, the smart converged terminal will refer to the value. The value is fixed at 0. The intelligent fusion terminal reads the measured power value E(t) at the current time t, and then substitutes it into... This yields an indicator of energy usage in terms of electricity consumption. After the three dimensions of electricity usage indicators are calculated, the terminal will... , , Stored in the data register, awaiting retrieval in the next step of fusion computing.

[0065] S107. Weighted fusion is used to obtain fused data and the fusion frequency is adaptively adjusted. This step aims to perform a weighted summation of the weights and indicators obtained in the previous steps to obtain the final fused data. Simultaneously, it dynamically adjusts the execution frequency of subsequent fusion calculations based on the temporal distribution patterns of electricity load, thereby achieving a balance between computational resource overhead and data acquisition density. For example... Figure 8 As shown, this step includes two subtasks: weighted calculation of the fused data and adaptive adjustment of the fusion frequency.

[0066] For the weighted calculation of the fused data, the intelligent fusion terminal reads the three fusion weights output by S105 from the data register. , , and the three energy usage indicators output by S106 , , And perform weighted summation. The intelligent converged terminal will first... and Feeding the product into the multiplier yields the first product; and Feed the product into the multiplier to get the second product; and The data is fed into a multiplier to obtain the third product. The intelligent fusion terminal then feeds the three products into an accumulator to obtain the sum F(t). F(t) is the fusion data at time t, and its value range is the same as the α index, which is [0,1). The closer F(t) is to 0, the lower the overall energy utilization level; the closer it is to 1, the higher the overall energy utilization level. The intelligent fusion terminal packages F(t) together with the timestamp of time t into a fusion record and appends it to the fusion result queue in chronological order. The depth of this fusion result queue is 10080, which can store fusion records for the most recent week (calculated at one record per minute).

[0067] For adaptive adjustment of the converged frequency, the intelligent converged terminal first divides the power consumption areas based on historical load data. The intelligent converged terminal reads load data from the data storage area for the same time point over the past 7 days. For each minute of the day (a total of 1440 time points), it calculates the arithmetic mean of the load over the 7 days, obtaining a daily average load curve. The intelligent converged terminal then calculates the overall arithmetic mean of the load across all 1440 time points on this daily average load curve, using this overall arithmetic mean as the baseline load. The intelligent converged terminal traverses every point in time on the daily average load curve, identifying load values ​​greater than... The time point is included in the high power consumption zone set, and the load value is less than The selected time points are assigned to the low-power consumption interval set, while the remaining time points are assigned to the medium-power consumption interval set. The intelligent fusion terminal merges consecutive time points within each set, forming a specific power consumption interval from the continuous sequence of time points, and records the start time, end time, and duration T of each interval. This division operation is performed once daily at 00:00, and the division result overwrites the previous day's result.

[0068] For each defined electricity consumption zone, the intelligent fusion terminal calculates the data fusion frequency for that zone independently. Taking a high-electricity-consumption zone as an example, the intelligent fusion terminal first reads the duration T of that zone and the arithmetic mean E of the electricity consumption data at all points in time within that zone. The intelligent fusion terminal then calculates the time index. Calculate electricity consumption index .like The calculation result is negative, and the intelligent fusion terminal will... The truncation threshold is set to 0. The intelligent fusion terminal calculates the frequency weight. Then, the frequency weight λ is combined with the preset base fusion frequency. The product of these two values ​​is used as the actual fusion frequency for that power consumption area. The intelligent fusion terminal performs the above calculations for the high, medium, and low power consumption ranges respectively, and obtains the corresponding fusion frequency values ​​for each range.

[0069] The intelligent fusion terminal writes the above calculation results into the fusion time scheduling table. The fusion time scheduling table is a linked list structure, where each node contains a timestamp field and a flag field. The intelligent fusion terminal selects the corresponding fusion frequency f based on the current power consumption zone, and then generates a fusion time point every 1 / f minutes, inserting each time point as a node into the linked list, until all fusion time points for the next hour are generated. The task scheduler inside the intelligent fusion terminal periodically scans this linked list. Whenever the system clock reaches the timestamp of a node in the linked list, the scheduler triggers the execution of the first subtask (S105 to S107). After a node is executed, the scheduler removes that node from the linked list. When the number of nodes in the linked list is less than 10, the intelligent fusion terminal automatically triggers a regeneration operation of the scheduling table to supplement subsequent fusion time points. When the intelligent fusion terminal detects that the current system time crosses the boundary of the power consumption zone, it immediately clears the current scheduling table and re-executes the power consumption zone judgment and fusion time point generation operations to switch to the new fusion frequency.

[0070] Example 2 This invention also provides a data fusion processing system for an intelligent fusion terminal. For example... Figure 9 As shown, the system is deployed in the embedded processor of the intelligent converged terminal in the distribution area, and includes a data acquisition module, an adaptive segmentation module, a data quality assessment module, a converged weight determination module, a power usage calculation module, a converged calculation module, and a converged frequency adjustment module. Data is transferred between modules through a shared memory buffer, and inter-module calls are managed uniformly by the system task scheduler. Specifically, it includes: Data acquisition module The data acquisition module is the system's front-end data entry point, responsible for receiving, parsing, and cleaning raw multi-source data. Internally, the data acquisition module integrates the DL / T 645 protocol stack, Modbus RTU protocol stack, and IEC 61850 protocol stack, enabling it to simultaneously establish communication links with various devices within the distribution area, such as smart meters, low-voltage circuit breakers, reactive power compensation devices, and charging piles.

[0071] The data acquisition module allocates an independent communication thread for each communication link. Each thread sends a read request to the corresponding device according to a preset sampling period and receives the returned data frames. The communication thread parses physical quantities such as the effective value of current, effective value of voltage, and cumulative active power from the response frame, and adds a timestamp read from the terminal real-time clock chip to each data point. After generating a data point structure, it sends it to the input buffer queue.

[0072] The data acquisition module includes a data cleaning unit. This unit retrieves data point structures from the input buffer queue and sequentially performs three operations: physical range verification, missing value completion, and jump suppression. The cleaned data is then stored according to its dimension into three separate queues: current data queue, voltage data queue, and energy data queue. All three queues are circular queues with a depth of 86400. After completing the enqueue operation, the data acquisition module sends a data ready signal to the adaptive segmentation module.

[0073] Adaptive segmentation module After receiving the data ready signal sent by the data acquisition module, the adaptive segmentation module reads the current complete data sequence from the three data queues and independently performs segmentation node detection and subsequence division for the three dimensions of current, voltage, and energy.

[0074] The adaptive segmentation module internally includes a difference value calculation unit, a mutation evaluation unit, a node filtering unit, and a sequence segmentation unit. The difference value calculation unit calculates the target difference value point-by-point for each data sequence. For points within the sequence, it takes the larger of the absolute differences between that point and its two preceding and following points; for endpoints, it only takes the absolute difference between that point and its only adjacent point. The mutation evaluation unit constructs a neighborhood window for each data point, calculates the mean of the target difference values ​​in the neighborhood, subtracts the target difference value of the current point from the neighborhood mean, and takes the absolute value to obtain the initial mutation degree evaluation value. Then, it performs min-max normalization on all initial evaluation values ​​within the same sequence, outputting the final mutation degree evaluation value for each data point. The node filtering unit filters data points with evaluation values ​​greater than or equal to a preset threshold as candidate segmentation nodes, performs deduplication and merging operations on the candidate nodes to ensure that the time interval between adjacent nodes is not less than a preset threshold, and finally forcibly adds the first and last endpoints of the sequence to form a segmentation node index set. The sequence segmentation unit divides the original sequence into several subsequences based on the set of segment node indices. Each subsequence contains all the continuous data points between two adjacent segment nodes, and the subsequences are packaged and sent to the data quality assessment module.

[0075] Data quality assessment module The data quality assessment module performs stability and orderliness analysis on each subsequence input from the adaptive segmentation module and calculates a quality score. The data quality assessment module includes a range calculation unit, an information entropy calculation unit, a stability assessment unit, and a score storage unit.

[0076] The range calculation unit iterates through the data points in the subsequence, finds the maximum and minimum values, and calculates the difference between them as the range. The information entropy calculation unit calculates the distribution information entropy of the absolute values ​​of the differences between adjacent data points in the subsequence. The stability evaluation unit performs negative exponential mapping on the range and information entropy respectively to obtain the first stability index and the second stability index, and then weights and sums the two to obtain the data quality score of the subsequence. The score storage unit binds the quality score with the dimension type, start index, and end index of the subsequence and writes it into the quality score cache. At the same time, it maintains a subsequence index table, recording the correspondence between the index range of each subsequence under each dimension and the cache key, for the fusion weight determination module to query.

[0077] Fusion weight determination module The fusion weight determination module calculates the fusion weights across three dimensions in real time for any query time based on the quality scores of each subsequence. The fusion weight determination module includes a subsequence localization unit and a weight calculation unit.

[0078] Upon receiving a query time t, the subsequence location unit calculates the sequence position index corresponding to that time, performs binary search in the subsequence index tables of current, voltage, and quantity dimensions respectively, locates the subsequence to which that time belongs, and then reads the quality score of the corresponding subsequence from the quality score cache to obtain the result. , , The weight calculation unit receives three quality scores, performs normalization to obtain the raw weights for the three dimensions, and then smooths the weights at consecutive time points using a moving average filter to output the final weights. , , .

[0079] Electricity usage calculation module The energy usage calculation module calculates the energy usage intensity of each dimension relative to a reference value at a specified time. The energy usage calculation module includes a reference value determination unit and an index conversion unit.

[0080] For the current and voltage dimensions, the reference value determination unit searches the energy data sequence forward and backward for the most recent moment when the energy change is zero, centered on the query time. If two moments with zero change are successfully found within a 1-hour window, the arithmetic mean of the corresponding dimension data within that time interval is taken as the reference value. If the search fails, the mean of historical stable periods is used as the reference value. For the energy dimension, the reference value is fixed at zero. After receiving the reference value, the index conversion unit reads the measured value at the current moment, calculates the absolute value of the difference between the measured value and the reference value, and then substitutes this difference into a preset monotonically increasing conversion function to obtain an energy usage index normalized to the [0,1) interval, and outputs it. , , .

[0081] Fusion computing module The fusion calculation module calculates the final fused data based on weights and indicators, and includes a weighted summation unit and a result output unit.

[0082] The weighted summation unit reads three weights from the fusion weight determination module. , , Three indicators are read from the electricity usage calculation module. The weighted summation unit first performs a normalization check on the weights, then feeds the normalized weights and corresponding indices into a multiplier to calculate the product, and finally sums the three multipliers to obtain the fusion value F. The result output unit packages the F value and the current time t into a fusion record, appends it to the fusion result circular buffer in chronological order, and provides a standard query interface for external application modules to call.

[0083] Fusion frequency adjustment module The fusion frequency adjustment module dynamically adjusts the fusion calculation frequency according to the time distribution pattern of the power load, and includes a zone division unit, a frequency calculation unit, and a time scheduling unit.

[0084] The interval division unit initiates an update operation once every morning at midnight, reads the load data of the same time in the past 7 days and calculates the daily average load curve, and then divides the time of day into high power consumption intervals, medium power consumption intervals and low power consumption intervals based on the overall average value of the daily average load, and records the start time, end time and duration of each interval.

[0085] The frequency calculation unit calculates the fusion frequency for each power consumption zone. For each zone, the frequency calculation unit reads its duration T and average power consumption E, calculates the time index and power consumption index, uses the average of the two indexes as the frequency weight, multiplies it by the preset base fusion frequency, obtains the actual fusion frequency for that zone, and writes the result into the configuration table.

[0086] The time-scheduling unit responds to system clock interrupts, queries the fusion frequency of its current power consumption zone based on the current system time, calculates the fusion interval, and generates a linked list of fusion time points for the next hour. The system task scheduler periodically scans this linked list. When the system clock reaches the timestamp of the head node of the linked list, a complete fusion calculation process is triggered (sequentially calling the fusion weight determination module, the power usage calculation module, and the fusion calculation module). After execution, the head node is deleted. When the system crosses the boundary of a power consumption zone, the time-scheduling unit clears the current linked list and regenerates the scheduling table to switch to the new fusion frequency.

[0087] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of equivalents of the claims be included within the present invention.

Claims

1. A data fusion processing method for an intelligent fusion terminal, characterized in that, Includes the following steps: Collect time-series data on current, voltage, and power consumption from the smart converged terminals in the distribution area; Calculate the mutation degree assessment value of each data point in each data dimension, and perform adaptive segmentation on each data dimension according to the mutation degree assessment value of each data point in each data dimension. Divide the original data sequence of each data dimension into several subsequences, and each data dimension forms its own independent set of subsequences. The data quality score for each data dimension is calculated separately for each subsequence, and the data quality score is used to characterize the stability of changes in data within the subsequence. For any data fusion moment, the subsequence to which the moment belongs in each data dimension is determined, and the data quality score of the subsequence to which each dimension belongs is used as the real-time credibility representation of the corresponding dimension. The real-time credibility representation of each dimension is normalized, and the normalized ratio is used as the dynamic fusion weight of each dimension at the current moment. The deviation of the measured value of each data dimension from its respective reference value at the current moment is calculated, and the deviation is mapped to the interval [0,1) to obtain the normalized power consumption index. The energy usage index is weighted and summed using the dynamic fusion weights of each dimension to obtain the fusion data at the current moment; Based on historical load data, the entire day is divided into multiple power consumption intervals. The data fusion frequency for each power consumption interval is calculated based on the time and power consumption characteristics of each interval. Within the current power consumption interval, the fusion data calculation for subsequent times is triggered according to the corresponding data fusion frequency.

2. The method according to claim 1, characterized in that, The calculation of the mutation degree assessment value for each data point in each data dimension includes: For the kth data point in any data dimension, a first difference value is determined based on the difference between the data point and the previous data point, a second difference value is determined based on the difference between the data point and the next data point, and a target difference value for the data point is determined based on the first difference value and the second difference value. The mean neighborhood difference is determined based on the target difference values ​​of multiple data points within the neighborhood of the data point. The mutation degree assessment value of the data point is determined based on the degree of difference between the target difference value of the data point and the mean difference value of the neighborhood.

3. The method according to claim 2, characterized in that, The steps of determining a first difference value based on the difference between the data point and its previous data point, determining a second difference value based on the difference between the data point and its next data point, and determining a target difference value for the data point based on the first difference value and the second difference value are as follows: The absolute value of the difference between the data point and the previous data point is calculated as the first difference value, and the absolute value of the difference between the data point and the next data point is calculated as the second difference value. The larger of the first difference value and the second difference value is taken as the target difference value of the data point. The step of determining the mean neighborhood difference based on the target difference values ​​of multiple data points within the neighborhood of the data point specifically involves: Centered on the data point, take the three consecutive data points to the left and the three consecutive data points to the right as the neighboring data points, and calculate the arithmetic mean of the target difference values ​​of all neighboring data points as the neighborhood difference mean. The step of determining the mutation degree assessment value of the data point based on the degree of difference between the target difference value of the data point and the mean difference value of the neighborhood is specifically as follows: The absolute value of the difference between the target difference value of the data point and the mean difference value of the neighborhood is calculated and then normalized by min-max to be used as the evaluation value of the mutation degree of the data point.

4. The method according to claim 1, characterized in that, The adaptive segmentation of each data dimension based on the evaluation value of the degree of mutation of each data point in each data dimension includes: Data points whose mutation rate assessment value is greater than a preset threshold in each data dimension are selected as candidate segmentation nodes; Candidate segment nodes for each data dimension are filtered separately. The time interval between adjacent candidate segment nodes is judged in chronological order. If the time interval is less than a preset time interval threshold, the node with the larger mutation degree evaluation value is retained and the node with the smaller mutation degree evaluation value is deleted until the time interval between all adjacent nodes is not less than the preset time interval threshold. The retained nodes are used as segment nodes. The first and last endpoints of the data sequence are added to the segment nodes to form a set of segment nodes for each dimension. Using the position of each node in each segment node set as the boundary, the original data sequence of the corresponding dimension is divided into several subsequences.

5. The method according to claim 1, characterized in that, The calculation of data quality scores for each data dimension, using subsequences as units, includes: For any subsequence, the first stability index is determined based on the difference between the maximum and minimum values ​​in the subsequence, and the second stability index is determined based on the distribution information entropy of the absolute value of the difference between adjacent data points in the subsequence. The first stability index and the second stability index are weighted and summed, and the resulting value is used as the data quality score of the subsequence.

6. The method according to claim 5, characterized in that, The determination of the first stability index based on the difference between the maximum and minimum values ​​in the subsequence is specifically as follows: The difference between the maximum and minimum values ​​in the subsequence is calculated as the range, and the negative exponent of the range is used to obtain the first stability index. The determination of the second stability index based on the distribution information entropy of the absolute value of the difference between adjacent data points in the subsequence is specifically as follows: Calculate the distribution information entropy of the absolute value of the difference between adjacent data points in the subsequence, and obtain the second stability index by taking the negative exponent of the information entropy.

7. The method according to claim 1, characterized in that, The calculation of the deviation of the measured value of each data dimension from its respective reference value at the current moment includes: For the current and voltage dimensions, the system searches the energy data sequence for the closest time before and after the current time when the energy change is zero. The arithmetic mean of the corresponding dimension data within the interval between the two zero change times is used as the reference value. If the search fails, the arithmetic mean of the corresponding dimension data during the historical stable period is used as the reference value. For the battery level dimension, the reference value is fixed at zero; The degree of deviation is obtained by calculating the absolute value of the difference between the measured value at the current moment and the reference value.

8. The method according to claim 7, characterized in that, Mapping the deviation to the [0,1) interval yields a normalized energy usage index. Specifically, the deviation is substituted into a transformation function. , and This yields a normalized index of electricity usage; among which, , , These are indicators of energy usage in terms of current, voltage, and quantity, respectively. This is the difference between the measured current value and the reference value. This is the difference between the measured voltage value and the reference value. This is the actual measured power level; the reference power level is 0.

9. The method according to claim 1, characterized in that, The calculation of the data fusion frequency for each power consumption interval based on its time and power consumption characteristics includes: for any power consumption interval, obtaining its duration T and average power consumption E, and calculating the time index. Calculate electricity consumption index ,like If the value is negative, it is set to zero, and the average of the time index and the electricity consumption index is taken as the frequency weight. The frequency weights are then combined with a preset base fusion frequency. Multiply to obtain the data fusion frequency for that power consumption area. .

10. The method according to claim 1, characterized in that, The step of triggering the calculation of fused data for subsequent moments within the current power consumption zone according to the corresponding data fusion frequency specifically includes: starting from the current moment, generating a linked list of fused moment points for the next hour according to the fusion interval corresponding to the data fusion frequency; when the system clock reaches any fused moment point in the linked list, triggering the calculation of the weighted summation to obtain the fused data once, and deleting that moment point from the linked list; automatically supplementing and generating subsequent moment points when the remaining nodes in the linked list are less than 10; and clearing and regenerating the current linked list when the system crosses the boundary of the power consumption zone. After collecting time-series data on current, voltage, and power from the intelligent fusion terminal in the collection area, the process also includes three cleaning operations: physical range verification, linear interpolation to complete missing values, and jump suppression based on the middle value of the sliding window. After normalizing the real-time credibility representation of each dimension, the method further includes performing a moving average filter on the original dynamic fusion weights of each dimension calculated at multiple consecutive data fusion times, and using the filtered values ​​as the final dynamic fusion weights of each dimension at the current time.

Images