Electrical load analysis method and system based on multi-source data fusion

Abstract

The invention discloses an electrical load analysis method and system based on multi-source data fusion. The method comprises the steps of obtaining load data and a meteorological data sequence of a target area; carrying out high-fidelity interpolation on the load missing data by adopting a strategy of combining form matching and meteorological verification; adaptively splitting the complete load sequence into a plurality of internally stable load data subsequences based on a preset change rate threshold; calculating a time weighted average value of the meteorological data corresponding to each sub-sequence as target meteorological data; performing two-stage clustering according to the data quantity and the target meteorological data to form a homogeneous load sub-sequence set; a representative sub-sequence with the meteorological closest to the central value is extracted from each set and input into the load analysis model to obtain a reference result; and efficiently deriving the reference result to all other subsequences in the set according to the meteorological difference by using a preset load correction coefficient. According to the method, on the premise of ensuring the analysis precision, the processing efficiency and the result consistency of the mass load data are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power load analysis, and particularly relates to a power load analysis method and system based on multi-source data fusion. BACKGROUND

[0002] Power load analysis is the core technical support for intelligent power grids to achieve precise scheduling, demand-side management, and energy efficiency optimization. With the popularization of advanced measurement systems, it is possible to obtain massive user-side load data and associated meteorological data. However, when using these multi-source data for in-depth analysis, existing technical solutions often face the following challenges: Firstly, in the data preprocessing link, the actual collected load data often has missing values. Traditional interpolation methods (such as time series interpolation, historical same period mean filling) or models based on simple meteorological regression often fail to fully exploit the self-similar fluctuation patterns within the load sequence and do not consider the dynamic correlation of meteorological conditions in specific time periods, resulting in deficiencies in the morphological fidelity and physical rationality of the repaired data, introducing source errors for subsequent analysis.

[0003] Secondly, in terms of analysis granularity and feature extraction, most methods treat long-time span load sequences as a whole and input them into the model for processing. This approach masks the non-stationary nature of the load and fails to effectively identify and separate the stationary operating phases representing different power states, resulting in rough analysis results that cannot reveal fine-grained user behavior characteristics such as device start-stop and mode switching.

[0004] Finally, in terms of computational efficiency and practicality, facing large-scale users or long-time sequence data, performing independent complex model (such as deep learning model) operations on each data segment or similar working conditions will result in huge computational overhead, making it difficult to meet the needs of real-time or quasi-real-time analysis. At the same time, independent analysis also makes it difficult to ensure the consistency of analysis results for similar power behaviors, restricting the large-scale application of the technology in engineering practice.

[0005] Therefore, how to design a power load analysis method that can simultaneously guarantee data quality, achieve fine-grained analysis, and significantly improve computational efficiency has become a technical problem urgently to be solved in the field. SUMMARY

[0006] The present application aims to overcome the shortcomings of the prior art and provide a power load analysis method and system based on multi-source data fusion. This method realizes efficient, fine, and consistent analysis of power load by intelligently segmenting load data based on stationarity, fine clustering combined with meteorological features, and deriving analysis results based on representative sub-sequences.

[0007] In a first aspect, the present application provides a power load analysis method based on multi-source data fusion, comprising: obtaining a load data sequence and a meteorological data sequence of a target region within a preset time period; based on the meteorological data sequence, using a preset data interpolation strategy to perform an interpolation operation on the load data sequence to obtain a target load data sequence; splitting the target load data sequence according to a preset load data division rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than a preset change rate threshold; determining target meteorological data corresponding to each target load data subsequence, and using a preset data clustering strategy to cluster the at least one target load data subsequence according to the target meteorological data and the data quantity of each target load data subsequence to obtain at least one target load data subsequence set; extracting a target load data subsequence from a target load data subsequence set and inputting the target load data subsequence into a preset power load analysis model, wherein the power load analysis model outputs a certain load analysis result corresponding to the target load data subsequence; determining other load analysis results of other target load data subsequences in the target load data subsequence set according to the certain load analysis result, wherein the other target load data subsequences are any target load data subsequences in the target load data subsequence set except for the target load data subsequence.

[0008] In a second aspect, the present application provides a power load analysis system based on multi-source data fusion, comprising: an acquisition module configured to obtain a load data sequence and a meteorological data sequence of a target region within a preset time period; a processing module configured to, based on the meteorological data sequence, use a preset data interpolation strategy to perform an interpolation operation on the load data sequence to obtain a target load data sequence; a splitting module configured to split the target load data sequence according to a preset load data division rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than a preset change rate threshold; The clustering module is configured to determine the target meteorological data corresponding to each target load data subsequence, and to cluster the at least one target load data subsequence according to the target meteorological data and the data quantity of each target load data subsequence, thereby obtaining at least one set of target load data subsequences. The output module is configured to extract a target load data subsequence from a set of target load data subsequences, and input the target load data subsequence into a preset power load analysis model, wherein the power load analysis model outputs a load analysis result corresponding to the target load data subsequence; The determination module is configured to determine other load analysis results of other target load data subsequences in the set of target load data subsequences based on the certain load analysis result, wherein the other target load data subsequences are any target load data subsequences in the set of target load data subsequences after removing the certain target load data subsequence.

[0009] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the power load analysis method based on multi-source data fusion according to any embodiment of the present invention.

[0010] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the power load analysis method based on multi-source data fusion according to any embodiment of the present invention.

[0011] This application presents a multi-source data fusion-based electricity load analysis method and system. Through a precise matching-repair strategy, it ensures the integrity and physical rationality of load data. It utilizes adaptive threshold segmentation to decompose the load curve into physically meaningful stationary event units, and organizes these units into highly homogeneous sets through a two-layer clustering of duration and weather. Finally, by employing a novel paradigm of representative sequence deep analysis and rapid derivation of load correction coefficients, it improves the efficiency of electricity load analysis while ensuring analytical accuracy, achieving efficient, precise, and interpretable intelligent analysis of electricity load. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 A flowchart illustrating a power load analysis method based on multi-source data fusion, as provided in an embodiment of the present invention; Figure 2 This is a structural block diagram of an electricity load analysis system based on multi-source data fusion, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0015] Please see Figure 1 The diagram shows a flowchart of a power load analysis method based on multi-source data fusion according to this application.

[0016] like Figure 1 As shown, the electricity load analysis method based on multi-source data fusion specifically includes the following steps: Step S101: Obtain the load data sequence and meteorological data sequence of the target area within a preset time period.

[0017] In this step, load data is acquired through smart meters, electricity consumption data collection systems, or Supervisory Control and Data Acquisition (SCADA) systems deployed in the target area (such as a distribution substation, a building, or an industrial park). The load data reflects the change in total active power in the area over time.

[0018] Meteorological data, such as temperature and humidity, is collected by miniature weather stations deployed in and around the target area.

[0019] Specifically, set a clear start and end time according to the analysis task. For example, if performing short-term load pattern analysis, it can be set to "obtain data from the past 7 days (00:00 on October 1, 2023 to 00:00 on October 8, 2023)".

[0020] Pre-set the time resolution of the data (e.g., 15 minutes, 1 hour). Request or configure data to be sampled at this resolution from the load data source and meteorological data source.

[0021] Step S102: Based on the meteorological data sequence, the load data sequence is interpolated using a preset data interpolation strategy to obtain the target load data sequence.

[0022] In this step, it is determined whether there are any missing load data points in the load data sequence; If it does not exist, the load data sequence is directly defined as the target load data sequence; If there is a missing load data point, then in the load data sequence, the load data is extended outward from the missing load data point until a preset number of known data points or sequence boundaries are encountered, and the truncated data segment is defined as a load data subsequence. In the load data sequence, all complete data segments that do not contain missing load data points and have the same length as a given load data subsequence are traversed as candidate mirror segments. The morphological similarity between a given load data subsequence and each candidate mirror segment is calculated, specifically including: The load data subsequence is decentered, and then a trend term sequence is extracted using the moving average method or the locally weighted scatter smoothing method. The load data subsequence is subtracted from the trend term sequence to obtain a residual term sequence of the load data subsequence. Calculate a Pearson correlation coefficient between a residual term sequence of a certain load data subsequence and a candidate residual term sequence of a certain candidate mirror segment, and define the absolute value of the Pearson correlation coefficient as a certain morphological similarity between a certain load data subsequence and a certain candidate mirror segment; The candidate mirror segment with the highest morphological similarity to a certain load data subsequence is selected as the target mirror segment. Based on the proportional relationship between the target mirror segment and the known data at the corresponding positions of a certain load data subsequence, the target residual sequence of the target mirror segment is scaled and shifted to generate an estimated value of a certain load residual at a certain load data missing point. Add the estimated value of a load residual to the trend term of a load data subsequence at a missing load data point to obtain the preliminary imputation value at the missing load data point; Calculate the similarity between a certain meteorological data at the time of the missing load data point and the target meteorological data at the corresponding time of the target mirror segment; If a certain similarity is lower than a preset threshold, the preliminary interpolation value is adjusted based on the meteorological data and the target meteorological data to obtain the final interpolation value; If a certain similarity is not lower than a preset threshold, the preliminary interpolation value is directly determined as the final interpolation value; Data interpolation is performed on a certain load data missing point based on the final interpolation value.

[0023] It should be noted that generating a specific load residual estimate at a point where load data is missing specifically includes: Align the positions of known data points in a certain load data subsequence with the corresponding positions in the target mirror segment; based on all aligned known data point pairs, establish a linear regression relationship between the residual values ​​of the known data points in a certain load data subsequence and the residual values ​​at the corresponding positions in the target mirror segment; according to the slope and intercept parameters of the linear regression relationship, scale and shift the residual values ​​at the corresponding positions of the target mirror segment at a certain load data missing point to obtain a certain load residual estimate for a certain load data missing point.

[0024] In one specific embodiment, the acquired load data sequence A traversal scan is performed to identify all data points marked as "invalid" or "null"; these points are the missing load data points. If no missing points are found, the load data sequence is... Defined as target load data sequence .

[0025] If missing data points are found, then address the missing load data points. (corresponding timestamp) Execute the following cyclic interpolation process: 1. Construct the local sequence segment to be repaired With missing point timestamps Centered on the time axis, expanding outwards at both ends, a "preset number of known data points" is set to 2, meaning the two nearest valid load data points are searched in each direction. If these points are found before reaching the sequence boundary, they are used as the boundary; if the boundary is reached, it is used as the boundary. Load data within this time range is extracted to form a subsequence of load data to be repaired, centered on the missing load data points. For example, if If there are two valid points before and after, then... The sequence is of length 5, and the middle position is the missing load data point.

[0026] 2. Search for candidate mirror segments and calculate morphological similarity. In the complete load data sequence In the middle (excluding periods containing any missing points), slide a subsequence of the load data to be repaired. For windows of the same length, extract all possible complete data segments as candidate mirror segments C.

[0027] Subsequence of load data to be repaired Perform morphological similarity calculations on each candidate mirror segment: Decentralization and Trend Decomposition: First, computation The average of all known points in the range, and from Decentralization is achieved by subtracting the average value from the value of each point; then, a simple moving average method with a window length of 3 is used to adjust the value of the decentralized points. After smoothing, the resulting smoothed sequence is the trend term sequence. .Will (Known points) and Subtracting the corresponding terms yields the sequence of residual terms. For the i-th candidate mirror segment Perform the same processing exactly to obtain the i-th candidate mirror segment. Candidate residual sequence .

[0028] Calculate morphological similarity: Calculate the residual term sequence (Using only values ​​at known locations) and candidate residual sequence Pearson correlation coefficient at the corresponding position .Pick The absolute value is used as a subsequence of the load data to be repaired. With the i-th candidate mirror segment The similarity in morphology between the two. The closer the value is to 1, the more consistent the fluctuation patterns of the two.

[0029] 3. Select the best mirror segment and estimate the residuals at missing points. The candidate mirror segment with the highest morphological similarity is selected as the target mirror segment M. Assume the subsequence of the load data to be repaired... If the position index p (e.g., the 3rd out of 5 points) is missing, the value of M at the corresponding position p is M[p].

[0030] Establish proportional relationships: Subsequences of the load data to be repaired The position indices of all known data points in the target image segment M are aligned with their corresponding positions. For example, a known point is located at index [1,2,4,5].

[0031] Linear regression modeling: using the residual values ​​of the target mirror segment M at these known index positions. Let X be the independent variable, and the subsequence of the load data to be repaired. Known residual values ​​at these locations For the dependent variable (Y), fit a linear model using the least squares method: Where a is the scaling slope and b is the translation intercept.

[0032] Estimate the residual of missing points: Calculate the residual value of the target mirror segment M at position index p. Substituting into the linear model above, the estimated load residuals for the missing load data points are calculated as follows: .

[0033] 4. Synthesize preliminary interpolation values Obtain the subsequence of the load data to be repaired Trend value at location index p Trend value Adding this to the estimated load residual value yields the preliminary interpolated value for the missing load data point at location index p. .

[0034] Specifically, the subsequence of the load data to be repaired The position index p in the sequence is used for a very simple and fast initial estimation, intended only to assist in trend calculation, not for final imputation. For example, a temporary value is obtained by using linear interpolation between the two nearest known points before and after the missing point, resulting in a temporary, missing-free sequence. ; right Apply a moving average algorithm (e.g., a simple moving average with a window size of 3) to calculate the trend value at each position i in the subsequence of the load data to be repaired. This yields the subsequence of the load data to be repaired. Trend value at location index p .

[0035] 5. Final verification and adjustment based on meteorological similarity Calculating meteorological similarity: from meteorological data sequences Extract missing load data points Timestamp temperature And the center time of the target mirror segment M temperature Calculate meteorological similarity The expression is: ,in, To prevent division by zero for extremely small constants. The larger the value, the more similar the weather patterns.

[0036] Judgment and Adjustment: Set a meteorological similarity threshold. For example, 0.8).

[0037] like Not less than If the meteorological conditions are highly similar, the initial interpolation values ​​will be considered... Used directly as the final interpolation value .

[0038] like Less than It was determined that differences in meteorological conditions might affect the load, necessitating adjustments. Adjustment amount. Simple rules can be set based on historical data analysis, for example: , where k is the pre-statistically obtained sensitivity coefficient of temperature to load in this region (unit: kW / °C). Then the final interpolated value... .

[0039] 6. Complete interpolation The calculated Enter the load data sequence Missing load data points .

[0040] Iterate through all missing load data points until all missing load data points are repaired. The resulting complete and continuous load data sequence is the target load data sequence.

[0041] In summary, by introducing morphological similarity matching and residual pattern transplantation, high-precision local waveform reconstruction is achieved. Specifically, this strategy first decomposes the load subsequence to be repaired and its candidate mirror segments into trend terms and residual terms, and specifically calculates the Pearson correlation coefficient at the residual term level to measure morphological similarity. This effectively isolates the differences in the overall load level and focuses on matching micro-fluctuation patterns. After selecting the best mirror segment, a linear regression relationship between its residual and the residual of the segment to be repaired is established based on known data points. Based on this, the residual of the mirror segment at the missing point is quantitatively scaled and shifted to generate a highly suitable residual estimate. Finally, this residual estimate is added to the trend term extracted from the local context of the segment to be repaired (through the moving average method). The synthesized interpolation value not only matches the local macroscopic change direction but also accurately reproduces the expected micro-fluctuation details, avoiding the waveform smoothing distortion or unreasonable abrupt changes often caused by traditional interpolation methods, and greatly improving the morphological fidelity of the repaired data. A meteorological similarity verification and adjustment step has been added, injecting physical constraints into the purely data-driven repair logic. This step calculates the similarity between the meteorological data (such as temperature and humidity) at the time of the missing point and the time of the representative moment of the mirror segment. If the similarity is high, the morphologically based interpolation value is directly adopted. If the similarity is lower than the preset threshold, the initial interpolation value is corrected according to the pre-analyzed meteorological-load response relationship (such as the sensitivity coefficient of temperature change to load). This mechanism ensures that the final interpolation value is not only the optimal solution in terms of data morphology, but also a reasonable result under the consistent consideration of external driving factors (meteorology), which significantly reduces the risk of repairing physical logic contradictions caused by ignoring differences in environmental conditions.

[0042] Step S103: The target load data sequence is split according to the preset load data division rules to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than the preset change rate threshold.

[0043] In this step, the first target load data is the first target load data in the target load data subsequence, and the other target load data are any target load data in the target load data subsequence excluding the first target load data; The target load data sequence is split according to a preset load data partitioning rule to obtain at least one target load data subsequence, including: Starting from the first data point of the target load data sequence, the first data point is taken as the starting point of the current target load data subsequence, and the target load data corresponding to the first data point is taken as the first target load data of the current target load data subsequence; The next data point in the target load data sequence is taken as the data point to be judged in chronological order, and the load change rate between the target load data to be judged corresponding to the data point to be judged and the first target load data in the current target load data subsequence is calculated. Determine whether the load change rate is not greater than a preset change rate threshold; If the change rate is not greater than the preset threshold, the data point to be judged will be included in the current target load data subsequence, and the next data point will be read for judgment. If the change rate is greater than the preset threshold, the current target load data subsequence will be sealed, and the data point to be determined will be used as the starting point of the next target load data subsequence, and the target load data corresponding to the data point to be determined will be used as the first target load data of the next target load data subsequence. After traversing all data points in the target load data sequence, at least one target load data subsequence is finally output.

[0044] In this embodiment, traditional fixed-time-window (e.g., hourly, daily) segmentation completely ignores the inherent dynamic characteristics of the load, potentially fragmenting a stable process or mixing different states. This invention, through its core "rate of change threshold determination" mechanism, uses the starting load value of the sequence as a dynamic benchmark to monitor the relative fluctuations of the load in real time. Segmentation only occurs when the fluctuation exceeds the threshold representing "stability." This ensures that each segmented subsequence internally satisfies a preset definition of stability, essentially corresponding to a continuous and relatively stable power consumption state or event (e.g., continuous operation of a piece of equipment, stable progress of a production process). This segmentation result has clear physical meaning, providing high-quality, highly homogeneous input for subsequent analysis of each subsequence as an independent unit, studying its correlation with meteorology, and performing pattern clustering.

[0045] Because the segmented subsequences exhibit relatively smooth fluctuations, the non-stationarity of the sequences is significantly reduced. This makes subsequent steps (such as clustering in step S104 and extracting a target load data subsequence in step S105) more stable and representative. More importantly, it provides the logical premise for extracting representative subsequences and for deriving the analysis results of other subsequences in the same set based on the representative results. Only when all subsequences within a set are guaranteed to be of the same type of stationary event can the strategy of "analyzing one representative to infer the whole" be reasonable and have high confidence. Therefore, this implementation method is the key foundation for achieving subsequent efficient and consistent analysis.

[0046] Step S104: Determine the target meteorological data corresponding to each target load data subsequence; based on the target meteorological data and the number of data in each target load data subsequence, use a preset data clustering strategy to cluster the at least one target load data subsequence to obtain at least one target load data subsequence set.

[0047] In this step, a start timestamp and an end timestamp of a target load data subsequence are obtained; based on the start timestamp and the end timestamp, all meteorological data points for the corresponding time period are extracted from the meteorological data sequence to form a raw meteorological data segment; the time-weighted average of each meteorological data point in the raw meteorological data segment is calculated as a target meteorological data corresponding to the target load data subsequence, wherein the weighting rule for the time-weighted average is: meteorological data points closer to the midpoint of the target load data subsequence are assigned higher weights.

[0048] In one specific embodiment, the timestamp of the first data point of a certain target load data subsequence is obtained. and the timestamp of the last data point .

[0049] Extracting meteorological segments: from meteorological data sequences The time in the middle is [ , All meteorological data points (e.g., temperature) within the interval constitute the original meteorological data segment. .

[0050] Calculate the time-weighted average: Determine the midpoint of time for a specific target load data subsequence .

[0051] Traversing raw meteorological data fragments For each meteorological data point, calculate the weight of each meteorological data point. .

[0052] The weighting rule adopts the inverse distance weighting concept: , Let j be the weight of the j-th meteorological data point. Let be the timestamp corresponding to the j-th meteorological data point. This means that... The closer to the midpoint of a certain target load data subsequence The greater its weight.

[0053] Calculate meteorological data for a specific target (e.g., target temperature). ): , In the formula, N is the total number of meteorological data points. Let be the temperature value of the j-th meteorological data point.

[0054] Furthermore, based on the number of data in each target load data subsequence, all target load data subsequences with the same number of data are divided into the same initial sequence group to obtain at least one initial sequence group; All target load data subsequences within an initial sequence group are sorted in ascending or descending order according to the values ​​of the target meteorological data corresponding to the target load data subsequences, forming a sequence list. Set an initial grouping window size, and divide the sequence list into several consecutive initial subgroups according to the initial grouping window size; Calculate the meteorological difference between the average value of the first target meteorological data corresponding to the first initial subgroup and the average value of the second target meteorological data corresponding to the second initial subgroup, and determine whether the meteorological difference is greater than a preset difference threshold. The first initial subgroup and the second initial subgroup are two adjacent initial subgroups among a number of consecutive initial subgroups. If the meteorological difference is not less than a preset difference threshold, then the first initial subgroup and the second initial subgroup will not be merged. If the meteorological difference is less than a preset difference threshold, the first initial subgroup and the second initial subgroup are combined to obtain the first target subgroup; Calculate the difference between the maximum and minimum values ​​of the target meteorological data corresponding to all target load data subsequences in the first target subgroup, i.e., the meteorological value range, and determine whether the meteorological value range is greater than a preset splitting threshold; If the range of the meteorological values ​​is not greater than the splitting threshold, then the first target subgroup is directly defined as a new target subgroup; If the range of the meteorological values ​​is greater than the splitting threshold, then within the first target subgroup, according to the median or mean of the target meteorological data, the first target subgroup is split into two new target subgroups, and each target load data subsequence within the same target subgroup is assigned to the same target load data subsequence set, thus obtaining the at least one target load data subsequence set.

[0055] In summary, firstly, regarding the accuracy and dimensionality reduction capabilities of meteorological data association, this method solves the problem of "how to represent the complex meteorological impacts of a period of time with a single data point" by calculating the time-weighted average for each stable load subsequence. Traditional methods either simply take the arithmetic mean or use the entire sequence as input. The former ignores the temporal non-uniformity of meteorological impacts, while the latter results in excessively high data dimensionality. This method assigns higher weights to meteorological data closer to the midpoint of the load event, enabling the calculated target meteorological data to more accurately represent the meteorological conditions that form the core driving force of the stable load state. This processing not only has clear physical meaning but, more importantly, compresses a meteorological sequence into a single feature value, laying the foundation for subsequent scalar-value-based clustering and efficient comparison, thus achieving efficient dimensionality reduction and high-fidelity information condensation. Secondly, the first stage of the clustering strategy groups data based on quantity (duration), capturing the most essential timescale attribute of electricity events (whether they are transient, short-term, or long-term), pre-separating events with vastly different physical meanings. The second stage, within groups of the same duration, divides data based on condensed meteorological characteristics using a clear set of adaptive rules: sorting, merging, and splitting. The core of this approach is controlling two thresholds: a preset difference threshold determines the aggregation of similar meteorological groups, and a splitting threshold ensures the uniformity of meteorological data within a group. Compared to traditional "black box" clustering algorithms (such as K-means), this rule-based method is more transparent and interpretable in its grouping process and criteria. Each resulting set exhibits high consistency in both "duration" and "driving meteorological data," ensuring that any subsequence within a set can effectively represent the common patterns of that set. This provides a solid logical premise and data foundation for the subsequent efficient strategy of analyzing only a representative subsequence. If there are large differences within a set, representativeness analysis becomes meaningless; this step minimizes this problem.

[0056] Step S105: Extract a target load data subsequence from a target load data subsequence set, and input the target load data subsequence into a preset power load analysis model. The power load analysis model outputs a load analysis result corresponding to the target load data subsequence.

[0057] In this step, the average value of the target meteorological data corresponding to all target load data subsequences in the set of target load data subsequences is calculated as the set meteorological center value; Identify a target meteorological data that is closest to the central value of the set of meteorological data, and extract the target load data subsequence corresponding to the target meteorological data from the set of target load data subsequences.

[0058] In summary, the core of the extraction strategy is to find the subsequence "closest to the ensemble's meteorological center". Since the preceding clustering has ensured a high concentration of meteorological conditions within the ensemble, this "closest to the center" sample is the best statistical representative of the entire ensemble in terms of meteorological driving dimensions. Using it as the analysis object, the resulting load analysis results (whether predicted values, classification labels, or anomaly scores) naturally carry the most common and typical electricity consumption behavior response patterns of the entire ensemble. This avoids the bias that may be introduced by random sampling and ensures that a certain target load data subsequence has high representativeness and reliability.

[0059] Step S106: Determine other load analysis results for other target load data subsequences in the target load data subsequence set based on the load analysis result, wherein the other target load data subsequences are any target load data subsequences in the target load data subsequence set after removing the target load data subsequence.

[0060] In this step, a load analysis result corresponding to a certain target load data subsequence is defined as a baseline analysis result, and a target meteorological data corresponding to a certain target load data subsequence is defined as baseline meteorological data; Obtain other target meteorological data corresponding to other target load data subsequences, and calculate the difference between the other target meteorological data and the baseline meteorological data to obtain other meteorological difference values; Based on the preset load correction coefficient, the other meteorological difference values ​​are converted into other expected adjustments to the baseline analysis results, and the other expected adjustments are combined with the baseline analysis results to generate other load analysis results for the other target load data subsequences.

[0061] Specifically, the preset load correction factor is obtained in advance through the following steps: Coefficient training data preparation: acquire historical data, which includes multiple historical load data subsequences and their corresponding historical target meteorological data, as well as the actual historical load analysis results obtained by analyzing each historical load data subsequence through the electricity load analysis model; Response relationship modeling: For load analysis results of the same category (such as load forecast or anomaly probability), establish a statistical relationship model between the result and its corresponding historical target meteorological data; the statistical relationship model is a linear model or a piecewise linear model, and its slope is the load correction coefficient; Coefficient determination and storage: The least squares method is used to fit the relationship model to determine the specific value of the load correction coefficient and store it as a preset parameter.

[0062] In summary, the method in this application, at the data preprocessing level, integrates the interpolation strategies of "internal morphological mirror restoration" and "external meteorological condition verification." It not only utilizes sequence self-similarity to faithfully restore the local fluctuation patterns of missing data but also ensures the physical rationality of the restoration results through meteorological difference correction, providing high-quality input for subsequent analysis. At the feature extraction and structuring level, through adaptive stationarity segmentation based on the load change rate threshold, continuous load curves are intelligently decomposed into stationary event units with clear physical meaning. Furthermore, by combining "meteorological weighted condensation representation" and "duration-meteorological secondary rule clustering," these units are organized into internally highly... Homogeneous sets (with consistent duration and similar meteorological conditions) provide a clear and reliable structural foundation for refined analysis. By employing a "representative subsequence sampling based on meteorological central values" mechanism, complex model calculations are focused on the most typical samples of each homogeneous set. Combined with a pre-set "load correction coefficient," lightweight arithmetic operations efficiently and consistently derive the analysis results to all subsequences within the set. This significantly reduces the overall computational complexity of the analysis from a linear relationship with the number of samples in traditional methods to a relationship only with the number of sets. While ensuring the accuracy of the results, this achieves an order-of-magnitude improvement in efficiency, successfully solving the computational bottleneck of in-depth analysis of large-scale electricity load data.

[0063] Please see Figure 2 The diagram shows a structural block diagram of an electricity load analysis system based on multi-source data fusion according to this application.

[0064] like Figure 2 As shown, the power load analysis system 200 includes an acquisition module 210, a processing module 220, a splitting module 230, a clustering module 240, an output module 250, and a determination module 260.

[0065] The acquisition module 210 is configured to acquire a load data sequence and a meteorological data sequence of a target area within a preset time period; the processing module 220 is configured to perform interpolation on the load data sequence based on the meteorological data sequence using a preset data interpolation strategy to obtain a target load data sequence; the splitting module 230 is configured to split the target load data sequence according to a preset load data partitioning rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than a preset change rate threshold; the clustering module 240 is configured to determine the target meteorological data corresponding to each target load data subsequence, and based on the number of data in the target meteorological data and each target load data subsequence, use a preset... The data clustering strategy clusters the at least one target load data subsequence to obtain at least one set of target load data subsequences; the output module 250 is configured to extract a target load data subsequence from a set of target load data subsequences and input the target load data subsequence into a preset power load analysis model, wherein the power load analysis model outputs a load analysis result corresponding to the target load data subsequence; the determination module 260 is configured to determine other load analysis results of other target load data subsequences in the set of target load data subsequences based on the load analysis result, wherein the other target load data subsequences are any target load data subsequences in the set of target load data subsequences excluding the target load data subsequence.

[0066] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0067] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the power load analysis method based on multi-source data fusion in any of the above method embodiments. In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows: Acquire load data sequences and meteorological data sequences for the target area within a preset time period; Based on the meteorological data sequence, the load data sequence is interpolated using a preset data interpolation strategy to obtain the target load data sequence. The target load data sequence is split according to a preset load data partitioning rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than a preset change rate threshold. Determine the target meteorological data corresponding to each target load data subsequence, and cluster the at least one target load data subsequence using a preset data clustering strategy based on the target meteorological data and the data quantity of each target load data subsequence to obtain at least one target load data subsequence set; A target load data subsequence is extracted from a set of target load data subsequences, and the target load data subsequence is input into a preset power load analysis model. The power load analysis model outputs a load analysis result corresponding to the target load data subsequence. Based on the load analysis result, determine other load analysis results for other target load data subsequences in the target load data subsequence set, wherein the other target load data subsequences are any target load data subsequences in the target load data subsequence set after removing the target load data subsequence.

[0068] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the multi-source data fusion-based power load analysis system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, which can be connected to the multi-source data fusion-based power load analysis system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0069] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the power load analysis method based on multi-source data fusion described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the power load analysis system based on multi-source data fusion. The output device 340 may include a display screen or other display device.

[0070] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0071] In one implementation, the above-described electronic device is applied to a power load analysis system based on multi-source data fusion, for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Acquire load data sequences and meteorological data sequences for the target area within a preset time period; Based on the meteorological data sequence, the load data sequence is interpolated using a preset data interpolation strategy to obtain the target load data sequence. The target load data sequence is split according to a preset load data partitioning rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than a preset change rate threshold. Determine the target meteorological data corresponding to each target load data subsequence, and cluster the at least one target load data subsequence using a preset data clustering strategy based on the target meteorological data and the data quantity of each target load data subsequence to obtain at least one target load data subsequence set; A target load data subsequence is extracted from a set of target load data subsequences, and the target load data subsequence is input into a preset power load analysis model. The power load analysis model outputs a load analysis result corresponding to the target load data subsequence. Based on the load analysis result, determine other load analysis results for other target load data subsequences in the target load data subsequence set, wherein the other target load data subsequences are any target load data subsequences in the target load data subsequence set after removing the target load data subsequence.

[0072] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing electricity load based on multi-source data fusion, characterized in that, include: Acquire load data sequences and meteorological data sequences for the target area within a preset time period; Based on the meteorological data sequence, the load data sequence is interpolated using a preset data interpolation strategy to obtain the target load data sequence. The target load data sequence is split according to a preset load data partitioning rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data and the first target load data in the target load data subsequence is not greater than a preset change rate threshold. Determine the target meteorological data corresponding to each target load data subsequence, and cluster the at least one target load data subsequence using a preset data clustering strategy based on the target meteorological data and the data quantity of each target load data subsequence to obtain at least one target load data subsequence set; A target load data subsequence is extracted from a set of target load data subsequences, and the target load data subsequence is input into a preset power load analysis model. The power load analysis model outputs a load analysis result corresponding to the target load data subsequence. Based on the load analysis result, determine other load analysis results for other target load data subsequences in the target load data subsequence set, wherein the other target load data subsequences are any target load data subsequences in the target load data subsequence set after removing the target load data subsequence.

2. The method for analyzing electricity load based on multi-source data fusion according to claim 1, characterized in that, The step of interpolating the load data sequence based on the meteorological data sequence using a preset data interpolation strategy to obtain the target load data sequence includes: Determine whether there are any missing load data points in the load data sequence; If it does not exist, then the load data sequence is directly defined as the target load data sequence; If there is a missing load data point, then in the load data sequence, the load data is extended outward from the missing load data point until a preset number of known data points or sequence boundaries are encountered, and the truncated data segment is defined as a load data subsequence. In the load data sequence, all complete data segments that do not contain missing load data points and have the same length as a certain load data subsequence are traversed as candidate mirror segments, and the morphological similarity between the certain load data subsequence and each candidate mirror segment is calculated, specifically including: The load data subsequence is decentered, and then a trend term sequence is extracted from the load data subsequence using the moving average method or the locally weighted scatter smoothing method. The load data subsequence is subtracted from the trend term sequence to obtain a residual term sequence of the load data subsequence. Calculate a Pearson correlation coefficient between a residual term sequence of a certain load data subsequence and a candidate residual term sequence of a certain candidate mirror segment, and define the absolute value of the Pearson correlation coefficient as a certain morphological similarity between the certain load data subsequence and the certain candidate mirror segment; The candidate mirror segment with the highest morphological similarity to a certain load data subsequence is selected as the target mirror segment. Based on the proportional relationship between the target mirror segment and the known data at the corresponding positions of the certain load data subsequence, the target residual sequence of the target mirror segment is scaled and shifted to generate a certain load residual estimate at the missing point of the certain load data. The estimated value of a certain load residual is added to the trend term of a certain load data subsequence at a certain load data missing point to obtain the preliminary imputation value at the certain load data missing point; Calculate the similarity between a certain meteorological data at the time of the missing load data point and the target meteorological data at the corresponding time of the target mirror segment; If a certain similarity is lower than a preset threshold, the preliminary interpolation value is adjusted based on the meteorological data and the target meteorological data to obtain the final interpolation value; If a certain similarity is not lower than the preset threshold, then the preliminary interpolation value is directly determined as the final interpolation value; Data interpolation is performed on a certain missing load data point based on the final interpolation value.

3. The method for analyzing electricity load based on multi-source data fusion according to claim 2, characterized in that, The step of scaling and shifting the target residual sequence of the target mirror segment based on the proportional relationship between the target mirror segment and the known data at corresponding positions of a certain load data subsequence, to generate an estimated load residual value at a certain load data missing point, includes: Align the positions of known data points in a certain load data subsequence with the corresponding positions in the target mirror segment; Based on all aligned pairs of known data points, a linear regression relationship is established between the residual values ​​of known data points in a certain load data subsequence and the residual values ​​at the corresponding positions of the target mirror segment. Based on the slope and intercept parameters of the linear regression relationship, the residual value of the target mirror segment at the corresponding position of a certain load data missing point is scaled and shifted to obtain a certain load residual estimate of the certain load data missing point.

4. The method for analyzing electricity load based on multi-source data fusion according to claim 1, characterized in that, The first target load data is the first target load data in the target load data subsequence, and the other target load data is any target load data in the target load data subsequence excluding the first target load data; The step of splitting the target load data sequence according to a preset load data partitioning rule to obtain at least one target load data subsequence includes: Starting from the first data point of the target load data sequence, the first data point is taken as the starting point of the current target load data subsequence, and the target load data corresponding to the first data point is taken as the first target load data of the current target load data subsequence; The next data point in the target load data sequence is taken as the data point to be judged in chronological order, and the load change rate between the target load data to be judged corresponding to the data point to be judged and the first target load data of the current target load data subsequence is calculated. Determine whether the load change rate is not greater than the preset change rate threshold; If the change rate is not greater than the preset change rate threshold, the data point to be determined is included in the current target load data subsequence, and the next data point is read for determination. If the change rate is greater than the preset threshold, the current target load data subsequence will be sealed, and the data point to be determined will be used as the starting point of the next target load data subsequence, and the target load data corresponding to the data point to be determined will be used as the first target load data of the next target load data subsequence. After traversing all data points in the target load data sequence, the at least one target load data subsequence is finally output.

5. The method for analyzing electricity load based on multi-source data fusion according to claim 1, characterized in that, The determination of the target meteorological data corresponding to each target load data subsequence includes: Obtain a start timestamp and an end timestamp of a specific target load data subsequence; Based on a certain start timestamp and a certain end timestamp, all meteorological data points for the corresponding time period are extracted from the meteorological data sequence to form a certain original meteorological data segment; Calculate the time-weighted average of each meteorological data in a certain original meteorological data segment, and use it as a certain target meteorological data corresponding to a certain target load data subsequence. The weighting rule for the time-weighted average is: meteorological data that is closer to the midpoint of the certain target load data subsequence is assigned a higher weight.

6. The method for analyzing electricity load based on multi-source data fusion according to claim 1, characterized in that, The step of clustering the at least one target load data subsequence according to the target meteorological data and the data quantity of each target load data subsequence using a preset data clustering strategy to obtain at least one target load data subsequence set includes: Based on the number of data in each target load data subsequence, all target load data subsequences with the same number of data are divided into the same initial sequence group to obtain at least one initial sequence group; All target load data subsequences within an initial sequence group are sorted in ascending or descending order according to the values ​​of the target meteorological data corresponding to the target load data subsequences, forming a sequence list. Set an initial grouping window size, and divide the sequence list into several consecutive initial subgroups according to the initial grouping window size; Calculate the meteorological difference between the average value of the first target meteorological data corresponding to the first initial subgroup and the average value of the second target meteorological data corresponding to the second initial subgroup, and determine whether the meteorological difference is greater than a preset difference threshold. The first initial subgroup and the second initial subgroup are two adjacent initial subgroups among a number of consecutive initial subgroups. If the meteorological difference is not less than a preset difference threshold, then the first initial subgroup and the second initial subgroup will not be merged. If the meteorological difference is less than a preset difference threshold, the first initial subgroup and the second initial subgroup are combined to obtain the first target subgroup; Calculate the difference between the maximum and minimum values ​​of the target meteorological data corresponding to all target load data subsequences in the first target subgroup, i.e., the meteorological value range, and determine whether the meteorological value range is greater than a preset splitting threshold; If the range of the meteorological values ​​is not greater than the splitting threshold, then the first target subgroup is directly defined as a new target subgroup; If the range of the meteorological values ​​is greater than the splitting threshold, then within the first target subgroup, according to the median or mean of the target meteorological data, the first target subgroup is split into two new target subgroups, and each target load data subsequence within the same target subgroup is assigned to the same target load data subsequence set, thus obtaining the at least one target load data subsequence set.

7. The method for analyzing electricity load based on multi-source data fusion according to claim 1, characterized in that, The step of extracting a target load data subsequence from a set of target load data subsequences includes: Calculate the average value of the target meteorological data corresponding to all target load data subsequences in the set of target load data subsequences, and use it as the set meteorological center value; Identify a target meteorological data that is closest to the central value of the set of meteorological data, and extract the target load data subsequence corresponding to the target meteorological data from the set of target load data subsequences.

8. The method for analyzing electricity load based on multi-source data fusion according to claim 1, characterized in that, The determination of other load analysis results for other target load data subsequences in the set of target load data subsequences based on the certain load analysis result includes: A load analysis result corresponding to a certain target load data subsequence is defined as a baseline analysis result, and a target meteorological data corresponding to a certain target load data subsequence is defined as baseline meteorological data; Obtain other target meteorological data corresponding to other target load data subsequences, and calculate the difference between the other target meteorological data and the baseline meteorological data to obtain other meteorological difference values; Based on the preset load correction coefficient, the other meteorological difference values ​​are converted into other expected adjustments to the baseline analysis results, and the other expected adjustments are combined with the baseline analysis results to generate other load analysis results for the other target load data subsequences.

9. A power load analysis system based on multi-source data fusion, characterized in that, include: The acquisition module is configured to acquire load data sequences and meteorological data sequences for a target area within a preset time period; The processing module is configured to perform interpolation operations on the load data sequence based on the meteorological data sequence using a preset data interpolation strategy to obtain the target load data sequence; The splitting module is configured to split the target load data sequence according to a preset load data division rule to obtain at least one target load data subsequence, wherein the load change rate between other target load data in the target load data subsequence and the first target load data is not greater than a preset change rate threshold. The clustering module is configured to determine the target meteorological data corresponding to each target load data subsequence, and to cluster the at least one target load data subsequence according to the target meteorological data and the data quantity of each target load data subsequence, thereby obtaining at least one set of target load data subsequences. The output module is configured to extract a target load data subsequence from a set of target load data subsequences, and input the target load data subsequence into a preset power load analysis model, wherein the power load analysis model outputs a load analysis result corresponding to the target load data subsequence; The determination module is configured to determine other load analysis results of other target load data subsequences in the set of target load data subsequences based on the certain load analysis result, wherein the other target load data subsequences are any target load data subsequences in the set of target load data subsequences after removing the certain target load data subsequence.

Images