Method, device and equipment for extracting characteristic of special variable user load based on multi-dimensional time sequence data and medium

Abstract

The application relates to a special variable user load feature extraction method, device, equipment and medium based on multi-dimensional time sequence data. The method comprises the following steps: preprocessing an original data set collected to obtain a reference data set, obtaining a to-be-repaired data point in the reference data set, respectively obtaining a spatial repair value and a time repair value of the to-be-repaired data point, fusing the spatial repair value and the time repair value based on a fusion weight to obtain a data repair value, updating a data value corresponding to the to-be-repaired data to the data repair value to obtain a repaired data set, performing time sequence alignment on a load repaired data set and an environment repaired data set to obtain a multi-dimensional time sequence data set, extracting a time sequence feature set of the multi-dimensional time sequence data set, screening the time sequence feature set to obtain a target feature set, and constructing a special variable user load space-time feature sample library based on the target feature set. The method can improve data processing precision.

Description

Technical Field

[0001] This application relates to the field of electrical digital data processing technology, and in particular to a method, apparatus, equipment and medium for extracting load characteristics of dedicated transformer users based on multidimensional time series data. Background Technology

[0002] Dedicated transformer users, as an important component of the power system, are characterized by large load capacity, strong fluctuations, and significant susceptibility to external environmental influences. Accurate load characteristic extraction and preprocessing are key prerequisites for realizing user load identification, power allocation, adjustability potential assessment, and demand-side management.

[0003] Traditional methods primarily focus on identifying loads by collecting electrical data and matching it with a standard feature library. However, the preprocessing steps are described in a relatively simple way, mainly involving signal noise reduction and anomaly removal. They do not take into account the fusion and feature construction of multi-source heterogeneous data, especially ignoring the important influence of external environmental data such as temperature and humidity on load characteristics, resulting in high data noise and a single feature dimension. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, equipment, and medium for extracting the load characteristics of dedicated transformer users based on multidimensional time series data, which can improve the accuracy of data processing, in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data, including:

[0006] The collected raw dataset is preprocessed to obtain a reference dataset, and the data points to be repaired in the reference dataset are obtained; the raw dataset includes the load raw dataset and the environment raw dataset;

[0007] The spatial and temporal repair values ​​of the data points to be repaired are obtained separately, and the spatial and temporal repair values ​​are fused based on the fusion weight to obtain the data repair value; the fusion weight is determined based on the fluctuation coefficient and average correlation coefficient of the data to be repaired.

[0008] Update the data values ​​corresponding to the data to be repaired with the data repair values ​​to obtain the repair dataset; the repair dataset includes the load repair dataset and the environment repair dataset.

[0009] The load repair dataset and the environmental repair dataset are time-series aligned to obtain a multidimensional time-series dataset, and the time-series feature set of the multidimensional time-series dataset is extracted; the time-series feature set includes statistical features, morphological features, transform domain features and environmental coupling features;

[0010] The time-series feature set is filtered to obtain the target feature set, and a spatiotemporal feature sample library of dedicated transformer user load is constructed based on the target feature set.

[0011] In one embodiment, the step of obtaining the data points to be repaired in the reference dataset includes:

[0012] The reference dataset is divided according to a preset time window;

[0013] For reference data points within the current time window, obtain the sequence mean and sequence standard deviation of the reference data points, and identify reference data points whose deviation from the sequence mean exceeds a preset statistical threshold as statistical outliers; the preset statistical threshold is determined based on the sequence standard deviation.

[0014] Get the timestamp of the reference data point, and then use the timestamp to find the missing data points in the reference data point.

[0015] Statistical outliers and missing data points are designated as data points to be repaired.

[0016] In one embodiment, the process of obtaining the space repair value includes:

[0017] Based on the user location of the user to be repaired corresponding to the data point to be repaired, obtain the candidate object associated with the user to be repaired, and obtain the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curve corresponding to the candidate object;

[0018] Candidate objects are filtered based on the correlation coefficient to obtain similar objects. Based on the correlation coefficient, the historical data values ​​of similar objects at the time of repair corresponding to the data point to be repaired are weighted and averaged to obtain the spatial repair value.

[0019] In one embodiment, the process of obtaining the time restoration value includes:

[0020] Based on the time point to be repaired corresponding to the data point to be repaired, obtain the corresponding time window to be repaired;

[0021] The sequence to be repaired corresponding to the time window to be repaired is input into the Long Short-Term Memory network model, and the time repair value is output.

[0022] In one embodiment, the process of obtaining the fusion weights includes:

[0023] Based on the time point to be repaired corresponding to the data point to be repaired, obtain the corresponding fluctuation time window, and obtain the fluctuation coefficient based on the data points within the fluctuation time window;

[0024] Obtain the average correlation coefficient corresponding to the data points to be repaired, and obtain the fusion weight based on the fluctuation coefficient and the average correlation coefficient.

[0025] In one embodiment, the step of time-series aligning the load repair dataset and the environmental repair dataset to obtain a multidimensional time-series dataset includes:

[0026] Obtain the target time point set based on the load repair dataset;

[0027] For each target time point in the target time point set, obtain the first raw data point before the target time point and the second raw data point after the target time point in the original environmental dataset.

[0028] Based on the first and second original data points, a polynomial function is constructed, and environmental data interpolation corresponding to the target time point is obtained according to the polynomial function.

[0029] Based on the interpolation of all environmental data, the environmental remediation dataset is updated, and the load remediation dataset and the updated environmental remediation dataset are time-series aligned to obtain a multidimensional time-series dataset.

[0030] In one embodiment, the step of filtering the time-series feature set to obtain the target feature set includes:

[0031] For each time-series feature in the time-series feature set, the information coefficient between the time-series feature and the target variable is obtained. Based on the information coefficient, the time-series feature set is first filtered to obtain the candidate feature set.

[0032] Using time-series features as input and target variables as output, an importance score is obtained for the time-series features through a decision tree model. Based on the importance score, the time-series feature set is then filtered to obtain the target feature set.

[0033] Secondly, this application also provides a device for extracting the load characteristics of dedicated transformer users based on multidimensional time series data, comprising:

[0034] The data point acquisition module is used to preprocess the collected raw dataset to obtain a reference dataset and to acquire the data points to be repaired in the reference dataset; the raw dataset includes the load raw dataset and the environment raw dataset.

[0035] The weighted fusion module is used to obtain the spatial and temporal repair values ​​of the data points to be repaired, and then fuse the spatial and temporal repair values ​​based on the fusion weight to obtain the data repair value; the fusion weight is determined based on the fluctuation coefficient and average correlation coefficient of the data to be repaired.

[0036] The data repair module is used to update the data values ​​corresponding to the data to be repaired with the data repair values ​​to obtain the repair dataset; the repair dataset includes the load repair dataset and the environment repair dataset;

[0037] The feature extraction module is used to perform time-series alignment of the load repair dataset and the environmental repair dataset to obtain a multidimensional time-series dataset, and to extract the time-series feature set of the multidimensional time-series dataset; the time-series feature set includes statistical features, morphological features, transform domain features and environmental coupling features;

[0038] The feature filtering module is used to filter the time series feature set to obtain the target feature set, and to build a spatiotemporal feature sample library of dedicated transformer user load based on the target feature set.

[0039] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method steps of any one of the first aspects.

[0040] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method steps of any one of the first aspects.

[0041] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements the method steps of any one of the first aspects.

[0042] The aforementioned method, apparatus, equipment, and medium for extracting load features of dedicated transformer users based on multidimensional time series data preprocesses the collected raw dataset to obtain a reference dataset, acquires the data points to be repaired in the reference dataset, obtains the spatial and temporal repair values ​​of the data points to be repaired, fuses the spatial and temporal repair values ​​based on fusion weights to obtain data repair values, updates the data values ​​corresponding to the data to be repaired with the data repair values ​​to obtain a repair dataset, performs temporal alignment on the load repair dataset and the environmental repair dataset to obtain a multidimensional time series dataset, extracts the temporal feature set of the multidimensional time series dataset, filters the temporal feature set to obtain a target feature set, and constructs a spatiotemporal feature sample library of dedicated transformer user load based on the target feature set. This can improve the repair accuracy of abnormal and missing data, comprehensively quantify load features, improve data processing accuracy, and thus improve sample quality. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1This is an application environment diagram of a method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data in one embodiment.

[0045] Figure 2 This is a flowchart illustrating a method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data in one embodiment.

[0046] Figure 3 This is a flowchart illustrating a method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data in another embodiment.

[0047] Figure 4 This is a structural block diagram of a dedicated transformer user load feature extraction device based on multidimensional time series data in one embodiment.

[0048] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0050] The method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Terminal 102 is used to preprocess the collected raw dataset to obtain a reference dataset, and to obtain the data points to be repaired in the reference dataset. It obtains the spatial and temporal repair values ​​of the data points to be repaired, and fuses the spatial and temporal repair values ​​based on fusion weights to obtain data repair values. It updates the data values ​​corresponding to the data to be repaired with the data repair values ​​to obtain a repaired dataset. It performs time-series alignment on the load repair dataset and the environmental repair dataset to obtain a multi-dimensional time-series dataset, extracts the time-series feature set of the multi-dimensional time-series dataset, filters the time-series feature set to obtain a target feature set, and constructs a special transformer user load spatiotemporal feature sample library based on the target feature set. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0051] In one exemplary embodiment, such as Figure 2 As shown, a method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data is provided, and this method is applied to... Figure 1 Taking terminal 102 as an example, the explanation includes the following steps 202 to 210. Wherein:

[0052] S202: Preprocess the collected raw dataset to obtain a reference dataset, and obtain the data points to be repaired in the reference dataset; the raw dataset includes the load raw dataset and the environment raw dataset.

[0053] Optionally, the raw dataset includes a load raw dataset and an environmental raw dataset acquired in parallel. The load raw dataset is acquired by a power consumption information acquisition terminal deployed on the user side of the dedicated transformer. This data is collected continuously at a first preset frequency (e.g., every 15 minutes) from the user's main incoming line, including instantaneous three-phase voltage and current values, and the calculated active and reactive power, forming a load raw dataset arranged in chronological order. The environmental raw dataset is acquired from a meteorological service center or local monitoring station at a second preset frequency (e.g., hourly) via a preset network interface protocol. This data is linked to the user's geographical location and includes key parameters such as temperature, humidity, and light intensity, constituting the environmental raw dataset. These two types of data inherently differ in sampling frequency, timestamp format, and data integrity, constituting initial data heterogeneity. Subsequently, the acquired first raw load dataset and first raw environmental dataset undergo preliminary quality assessment and labeling to identify abnormal data points requiring repair.

[0054] S204: Obtain the spatial repair value and temporal repair value of the data points to be repaired respectively, and fuse the spatial repair value and temporal repair value based on the fusion weight to obtain the data repair value; the fusion weight is determined according to the fluctuation coefficient and average correlation coefficient of the data to be repaired.

[0055] Optionally, for the marked data points to be repaired, precise repair is performed using the spatial and temporal correlations of the data to generate a complete and reliable data sequence. Spatial repair values ​​based on group similarity and temporal repair values ​​based on individual historical patterns are computed in parallel, and the two are dynamically fused through an adaptive mechanism. The spatial repair value refers to the completion value calculated using neighboring similar data points around the point to be repaired, while the temporal repair value refers to the completion value calculated using the historical / subsequent time-series data of the point to be repaired itself. The volatility coefficient measures the stability / severity of the data; the greater the volatility, the more unstable the data, and the less reliable the single-dimensional complement. The average correlation coefficient measures the strength of the spatial / temporal correlation; the higher the correlation, the more reliable the complement result for that dimension. The spatial and temporal repair values ​​are fused using fusion weights to finally obtain the accurate data repair value.

[0056] S206: Update the data values ​​corresponding to the data to be repaired to the data repair values ​​to obtain the repair dataset; the repair dataset includes the load repair dataset and the environment repair dataset.

[0057] Optionally, after calculating the accurate data repair values, the original data points are replaced with the data repair values ​​to obtain a complete repair dataset. The repair process for the original load dataset and the original environment dataset is similar, and the final repair dataset includes the load repair dataset and the environment repair dataset.

[0058] S208: Perform time-series alignment on the load repair dataset and the environmental repair dataset to obtain a multidimensional time-series dataset, and extract the time-series feature set of the multidimensional time-series dataset; the time-series feature set includes statistical features, morphological features, transform domain features and environmental coupling features.

[0059] Optionally, due to the different sampling frequencies of load data and environmental data, a time-series misalignment problem exists. Resampling and interpolation techniques are used to generate a fused dataset where all dimensions of data are completely synchronized in terms of timestamps, providing a unified time reference for subsequent feature extraction. Typically, the time interval of the frequently collected load data is used as the reference to generate a continuously synchronized time stamp sequence, serving as a common time axis for all data alignment. For the complete load dataset, its original sampling interval is already consistent with the unified time axis, so it can be directly mapped. For the complete environmental dataset, its original sampling frequency is usually lower than the unified time axis, requiring upsampling interpolation. The load data and environmental data, already aligned on the same time reference, are horizontally merged according to timestamps to finally generate a synchronized multidimensional time-series dataset. The synchronized multidimensional time-series dataset is a standard two-dimensional data matrix, where each row corresponds to a unified time point, and each column corresponds to a specific data dimension, providing a unified and regular data input for subsequent feature extraction. From the generated synchronous multidimensional time-series data, using a fixed time window as the analysis unit, a set of features that can comprehensively characterize the load behavior of dedicated transformer users is systematically extracted from four dimensions: statistics, morphology, transformation domain, and environmental coupling.

[0060] S210: Filter the time series feature set to obtain the target feature set, and construct a spatiotemporal feature sample library of dedicated transformer user load based on the target feature set.

[0061] Optionally, the extracted high-dimensional original feature vectors are reduced in dimensionality and refined to construct a standardized and structured feature sample library for direct training and use by various downstream machine learning models. After feature selection, data standardization and sample encapsulation are performed. All constructed samples are organized and stored in a unified format to form a dedicated transformer user load spatiotemporal feature sample library. This library has a clear structure, concise feature dimensions, and has undergone standardization, enabling it to directly and efficiently serve subsequent advanced analytical applications such as load identification, power decomposition, and adjustability potential assessment.

[0062] In the above-mentioned method for extracting the load features of special transformer users based on multidimensional time series data, the original dataset is preprocessed to obtain a reference dataset, and the data points to be repaired in the reference dataset are obtained. The spatial repair values ​​and temporal repair values ​​of the data points to be repaired are obtained respectively, and the spatial repair values ​​and temporal repair values ​​are fused based on the fusion weight to obtain the data repair values. The data values ​​corresponding to the data to be repaired are updated with the data repair values ​​to obtain the repaired dataset. The load repair dataset and the environmental repair dataset are time-series aligned to obtain a multidimensional time series dataset, and the temporal feature set of the multidimensional time series dataset is extracted. The temporal feature set is filtered to obtain the target feature set. Based on the target feature set, a spatiotemporal feature sample library of special transformer user load is constructed. This method can improve the repair accuracy of abnormal and missing data, comprehensively quantify load features, improve data processing accuracy, and thus improve sample quality.

[0063] In an exemplary embodiment, the step of obtaining data points to be repaired in a reference dataset includes: dividing the reference dataset according to a preset time window; for the reference data points within the current time window, obtaining the sequence mean and sequence standard deviation of the reference data points, and identifying reference data points whose deviation from the sequence mean exceeds a preset statistical threshold as statistical outliers; the preset statistical threshold is determined based on the sequence standard deviation; obtaining the timestamps of the reference data points, and obtaining missing data points in the reference data points based on the timestamps; and identifying the statistical outliers and missing data points as data points to be repaired.

[0064] Optionally, taking the original load dataset as an example, for the active power in the load data, a physical rationality check is first performed to determine whether the active power value exceeds 1.2 times the rated capacity of the user's transformer or is negative. If so, it is marked as a physical anomaly. Then, outlier identification is performed based on statistical principles. The mean and standard deviation of the sequence are calculated within a sliding time window. If the absolute deviation of a data point from the mean exceeds three times the standard deviation, it is marked as a statistical anomaly. Simultaneously, by verifying the continuity of timestamps, data points that should theoretically exist but were not actually collected are identified and marked as missing values. For unmarked normal data points, a sliding window averaging filter algorithm is used for smoothing. Taking the current point as the center, three points are taken, one before and one after, and the arithmetic mean of these three points is calculated as the new value of the current point to suppress high-frequency random noise. This operation does not change the already marked anomalies or missing points. Similarly, a physical rationality check is performed on the original environmental dataset, and statistical verification is performed by calculating the mean and standard deviation of the sequence within a sliding time window. Finally, all statistical anomalies and missing data points are designated as data points to be repaired.

[0065] In this embodiment, the reference dataset is divided according to a preset time window. For the reference data points within the current time window, the sequence mean and sequence standard deviation of the reference data points are obtained. Reference data points whose deviation from the sequence mean exceeds a preset statistical threshold are identified as statistical outliers. The timestamps of the reference data points are obtained, and missing data points in the reference data points are obtained based on the timestamps. Statistical outliers and missing data points are identified as data points to be repaired. This approach can accurately identify data points to be repaired, improve the quality of the original dataset, and thus improve the quality of subsequent samples.

[0066] In an exemplary embodiment, the process of obtaining the spatial repair value includes: obtaining candidate objects associated with the user to be repaired based on the user location of the user to be repaired corresponding to the data point to be repaired; obtaining the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curve corresponding to the candidate object; filtering the candidate objects based on the correlation coefficient to obtain similar objects; and, based on the correlation coefficient, performing a weighted average of the historical data values ​​of the similar objects at the time to be repaired corresponding to the data point to be repaired to obtain the spatial repair value.

[0067] Optionally, taking the original load dataset as an example, a set of candidate users corresponding to the data points to be repaired, located in the same power supply area or belonging to the same industry, is selected from the user profile database. Then, the Pearson correlation coefficient between the historical load curves of each candidate user and the curve of the user to be repaired is calculated. Specifically, the two daily load curves are treated as two high-dimensional vectors, and a correlation coefficient between -1 and 1 is obtained by calculating the ratio of the product of their covariance and their respective standard deviations. The closer the Pearson correlation coefficient is to 1, the more consistent the trends of the two curves, indicating that the electricity consumption behavior of the two users is more similar. Based on this, several users with the highest similarity are selected as similar neighbors. Finally, the actual load values ​​of these neighboring users at the time of repair are weighted and averaged according to their correlation coefficients with the user to be repaired to obtain the spatial repair value. Similarly, for the original environmental dataset, historical environmental data located in similar geographical locations and altitudes to the user to be repaired are obtained, and the corresponding correlation coefficients are acquired. Then, the environmental data of the candidate locations is filtered, and the spatial repair value of the environmental data is calculated based on the filtered similar data.

[0068] In this embodiment, by obtaining candidate objects associated with the user to be repaired based on the user location of the user to be repaired corresponding to the data point to be repaired, obtaining the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curves corresponding to the candidate objects, filtering the candidate objects based on the correlation coefficient to obtain similar objects, and taking a weighted average of the historical data values ​​of similar objects at the time to be repaired corresponding to the data point to be repaired based on the correlation coefficient to obtain the spatial repair value. This can accurately obtain the spatial repair value, provide high-precision spatial reference data for the data point to be repaired, and thus improve the accuracy of the data to be repaired.

[0069] In an exemplary embodiment, the process of obtaining the time repair value includes: obtaining the corresponding time window to be repaired based on the time point to be repaired corresponding to the data point to be repaired; inputting the sequence to be repaired corresponding to the time window to be repaired into the long short-term memory network model, and outputting the time repair value.

[0070] Optionally, for outliers in the time series, historical load data of the data point to be repaired within a time window before and after the time point to be repaired is extracted. If other outliers or missing points exist within the window, they are temporarily filled using linear interpolation to form a continuous sequence. That is, temporary values ​​are calculated using the two nearest normal points before and after these outliers through a linear relationship, thus forming a continuous input sequence. After the input sequence is constructed, a pre-trained Long Short-Term Memory (LSTM) network model is invoked. The constructed input sequence is input into the LTM network model, and the model calculates through forward propagation, outputting the time-repaired value for the time point to be repaired. The LTM network model is a recurrent neural network specifically designed for processing time series data. Through memory units and gating mechanisms, the LTM network model can effectively capture long-term dependencies in the sequence.

[0071] In this embodiment, by obtaining the corresponding repair time window based on the repair time corresponding to the data point to be repaired, inputting the repair sequence corresponding to the repair time window into the long short-term memory network model, and outputting the time repair value, the time repair value can be accurately obtained, thereby improving the accuracy of the data to be repaired.

[0072] In an exemplary embodiment, the process of obtaining the fusion weight includes: obtaining a corresponding fluctuation time window based on the time to be repaired corresponding to the data point to be repaired; obtaining the fluctuation coefficient based on the data points within the fluctuation time window; obtaining the average correlation coefficient corresponding to the data point to be repaired; and obtaining the fusion weight based on the fluctuation coefficient and the average correlation coefficient.

[0073] Optionally, the fusion weight is calculated using the volatility coefficient and the average correlation coefficient. The volatility coefficient (the ratio of standard deviation to mean) refers to the fluctuation of load data within a short time window near the data point to be repaired, reflecting the stability of the recent load. The average correlation coefficient is the average correlation coefficient of several similar objects, reflecting the consistency of group behavior. The fusion weight ensures that when the average spatial similarity is high and the data volatility coefficient is low, the fusion weight approaches one, meaning lateral repair dominates; conversely, when the average spatial similarity is low or the data volatility coefficient is high, the fusion weight approaches zero, meaning longitudinal prediction dominates. Finally, the repaired value = lateral repaired value × fusion weight + longitudinal predicted value × (1 - fusion weight). This final repaired value replaces the original data points, generating a complete repaired dataset.

[0074] In this embodiment, by obtaining the corresponding fluctuation time window based on the time to be repaired corresponding to the data point to be repaired, obtaining the fluctuation coefficient based on the data point within the fluctuation time window, obtaining the average correlation coefficient corresponding to the data point to be repaired, and obtaining the fusion weight based on the fluctuation coefficient and the average correlation coefficient, the fusion weight can be dynamically adjusted to improve the fusion accuracy, thereby improving the accuracy of the data to be repaired.

[0075] In an exemplary embodiment, the step of temporally aligning the load repair dataset and the environmental repair dataset to obtain a multidimensional time-series dataset includes: obtaining a set of target time points based on the load repair dataset; for each target time point in the set of target time points, obtaining a first original data point before the target time point and a second original data point after the target time point in the original environmental dataset; constructing a polynomial function based on the first and second original data points, and obtaining environmental data interpolation corresponding to the target time point based on the polynomial function; updating the environmental repair dataset based on all environmental data interpolation, and temporally aligning the load repair dataset and the updated environmental repair dataset to obtain a multidimensional time-series dataset.

[0076] Optionally, a continuous sequence of equally spaced timestamps is generated based on the time interval of the high-frequency collected load data, serving as a common time axis for all data alignment. For the complete load dataset, its native sampling interval is already consistent with the unified time axis, so it can be directly mapped. For the complete environment dataset, its original sampling frequency is lower than the unified time axis, requiring upsampling interpolation. A piecewise cubic Hermitian interpolation polynomial method is used for upsampling. Specifically, for each target time point on the unified time axis, two original data points before and after that target time point are found in the original environment dataset. Using the ordinate values ​​and derivative values ​​of these four points, a cubic polynomial function is constructed such that the function passes through these four points, and the derivative values ​​at these points are consistent with preset values. Then, the target time point is substituted into this cubic polynomial to calculate the corresponding interpolation result. While preserving the original data point information, a smooth and continuous environmental data interpolation is calculated for each unified time point based on the values ​​and trends of the preceding and following data points. After resampling and interpolation are completed, multidimensional data splicing is performed. Load data and environmental data that have been aligned on the same time base are horizontally merged according to timestamps to finally generate a synchronous multidimensional time series dataset.

[0077] In this embodiment, a target time point set is obtained based on the load repair dataset. For each target time point in the target time point set, a first original data point before the target time point and a second original data point after the target time point are obtained in the original environmental dataset. Based on the first and second original data points, a polynomial function is constructed. The environmental data interpolation corresponding to the target time point is obtained according to the polynomial function. The environmental repair dataset is updated according to all the environmental data interpolation. The load repair dataset and the updated environmental repair dataset are time-series aligned to obtain a multi-dimensional time-series dataset, which can provide a consistent time dimension for subsequent feature extraction, thereby improving the accuracy of feature extraction.

[0078] In an exemplary embodiment, the process of filtering a time-series feature set to obtain a target feature set includes: for each time-series feature in the time-series feature set, obtaining the information coefficient between the time-series feature and the target variable; performing a first filtering of the time-series feature set based on the information coefficient to obtain a candidate feature set; using the time-series feature as input and the target variable as output, obtaining an importance score corresponding to the time-series feature through a decision tree model; and performing a second filtering of the time-series feature set based on the importance score to obtain the target feature set.

[0079] Optionally, the time-series feature set includes statistical features, morphological features, transform domain features, and environmental coupling features. Statistical features include the maximum active power value (i.e., daily maximum load), the minimum active power value (i.e., daily minimum load), the arithmetic mean of active power values ​​(i.e., daily average load), the ratio of daily average load to daily maximum load (i.e., daily load factor), the difference between the maximum and minimum active power values ​​(i.e., daily load peak-to-valley difference), the variance, skewness, and kurtosis of the series. These statistical features characterize the basic distribution patterns of load data from different perspectives, including central tendency, dispersion, and distribution shape. Morphological features are used to characterize the shape and trend of the load curve, including the time of peak occurrence, the time of trough occurrence, the load ramp-up rate, the load decline rate, and the load change trend. Transform domain features reveal the intrinsic characteristics of the load from the perspectives of frequency domain and complexity. By performing three-level wavelet packet decomposition on the daily load sequence, the energy of the original sequence is decomposed into multiple different frequency bands, and the proportion of energy in each frequency band to the total energy is calculated, forming a set of wavelet packet energy spectrum features to characterize the distribution of load fluctuations at different frequency scales. Simultaneously, the singular spectral entropy of the daily load sequence is calculated. By performing singular value decomposition on the trajectory matrix constructed from the sequence, the singular value spectrum is obtained, and its normalized information entropy value is then calculated, reflecting the complexity and regularity of the load time series. Environmental coupling features are used to quantify the dynamic relationship between the load and external environmental factors, jointly analyzing load data with temperature data within the same window. By calculating the correlation coefficient between the load sequence and the temperature sequence at different time offsets, the offset with the strongest correlation is identified, reflecting the intensity and delay of the temperature's influence on the load. Then, the mutual information between the load and temperature sequences is calculated to measure the overall correlation strength, including nonlinear relationships. Simultaneously, the mean standard deviation of the regression residuals within the sliding window is calculated. This involves establishing a linear regression model of the load with respect to temperature within the sliding time window, calculating the standard deviation of the regression residuals within each window, and then averaging the standard deviations across all windows. This feature value reflects the severity of the residual fluctuations in the load itself after excluding the linear influence of temperature. Through these four feature extraction methods, a high-dimensional original feature vector is generated for a given time window, comprehensively characterizing the user's load characteristics on that day.

[0080] Furthermore, feature dimensionality reduction is achieved through a two-stage hybrid feature selection strategy. The first stage is filtering selection, which calculates the maximum information coefficient between each feature in the original feature vector and the target variable. The maximum information coefficient effectively captures the linear and non-linear correlation strength between the feature and the target. Features with a maximum information coefficient below a preset threshold are considered weakly correlated and discarded, resulting in a candidate feature subset. The second stage is wrapping selection, which uses all features in the candidate feature subset as input and the target variable as output to train a gradient boosting decision tree model. After model training, the model's built-in feature importance evaluation function (typically based on the frequency of feature usage in decision tree node splitting and the resulting reduction in impurity) is called to obtain an importance score for each feature. Features are sorted from highest to lowest score, and a predetermined number of top-ranked features are selected to form the final refined feature subset. This hybrid selection strategy of filtering followed by wrapping ensures both the efficiency of feature selection and the effectiveness of the selected features for the modeling task.

[0081] In this embodiment, for each time-series feature in the time-series feature set, the information coefficient between the time-series feature and the target variable is obtained. The time-series feature set is then first screened based on the information coefficient to obtain a candidate feature set. Using the time-series feature as input and the target variable as output, the importance score corresponding to the obtained time-series feature is obtained through a decision tree model. Based on the importance score, the time-series feature set is then second screened to obtain the target feature set. This approach can comprehensively quantify load features, improve data processing accuracy, and thus improve sample quality.

[0082] In one exemplary embodiment, such as Figure 3 As shown, a method for extracting user load features of dedicated transformers based on multidimensional time series data is provided. This method includes the following steps:

[0083] The collected raw dataset is preprocessed to obtain a reference dataset, which is then divided according to a preset time window. For each reference data point within the current time window, the sequence mean and standard deviation are obtained. Reference data points whose deviation from the sequence mean exceeds a preset statistical threshold are identified as statistical outliers. The preset statistical threshold is determined based on the sequence standard deviation. The timestamps of the reference data points are obtained, and missing data points are identified based on these timestamps. Statistical outliers and missing data points are designated as data points to be repaired. The raw dataset includes the load raw dataset and the environment raw dataset.

[0084] Based on the user location of the user to be repaired corresponding to the data point to be repaired, candidate objects associated with the user to be repaired are obtained, and the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curves corresponding to the candidate objects is obtained. The candidate objects are filtered according to the correlation coefficient to obtain similar objects. Based on the correlation coefficient, the historical data values ​​of similar objects at the time to be repaired corresponding to the data point to be repaired are weighted and averaged to obtain the spatial repair value. According to the time to be repaired corresponding to the data point to be repaired, the corresponding time window to be repaired is obtained. The sequence to be repaired corresponding to the time window to be repaired is input into the Long Short-Term Memory network model, and the time repair value is output.

[0085] Based on the time point to be repaired corresponding to the data point to be repaired, obtain the corresponding fluctuation time window, and obtain the fluctuation coefficient based on the data point within the fluctuation time window; obtain the average correlation coefficient corresponding to the data point to be repaired, obtain the fusion weight based on the fluctuation coefficient and the average correlation coefficient, and fuse the spatial repair value and the temporal repair value based on the fusion weight to obtain the data repair value; the fusion weight is determined based on the fluctuation coefficient and the average correlation coefficient of the data to be repaired.

[0086] The data values ​​corresponding to the data to be repaired are updated with the data repair values ​​to obtain the repair dataset; the repair dataset includes the load repair dataset and the environment repair dataset.

[0087] Obtain a set of target time points from the load repair dataset; for each target time point in the target time point set, obtain the first original data point before the target time point and the second original data point after the target time point in the original environmental dataset; based on the first and second original data points, construct a polynomial function, and obtain the environmental data interpolation corresponding to the target time point according to the polynomial function; update the environmental repair dataset according to all environmental data interpolation, and perform time-series alignment between the load repair dataset and the updated environmental repair dataset to obtain a multidimensional time-series dataset, and extract the time-series feature set of the multidimensional time-series dataset; the time-series feature set includes statistical features, morphological features, transform domain features, and environmental coupling features.

[0088] For each time-series feature in the time-series feature set, the information coefficient between the time-series feature and the target variable is obtained. The time-series feature set is then first screened based on the information coefficient to obtain a candidate feature set. Using the time-series feature as input and the target variable as output, the importance score corresponding to the obtained time-series feature is scored through a decision tree model. The time-series feature set is then second screened based on the importance score to obtain the target feature set. A spatiotemporal feature sample library of dedicated transformer user load is then constructed based on the target feature set.

[0089] In this embodiment, the collected raw dataset is preprocessed to obtain a reference dataset, and the data points to be repaired in the reference dataset are obtained. The spatial repair value and temporal repair value of the data points to be repaired are obtained respectively, and the spatial repair value and temporal repair value are fused based on the fusion weight to obtain the data repair value. The data value corresponding to the data to be repaired is updated with the data repair value to obtain the repair dataset. The load repair dataset and the environmental repair dataset are time-series aligned to obtain a multidimensional time-series dataset, and the time-series feature set of the multidimensional time-series dataset is extracted. The time-series feature set is filtered to obtain the target feature set. Based on the target feature set, a spatiotemporal feature sample library of special transformer user load is constructed, which can improve the repair accuracy of abnormal and missing data, comprehensively quantify load characteristics, improve data processing accuracy, and thus improve sample quality.

[0090] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0091] Based on the same inventive concept, this application also provides a device for extracting the load features of dedicated transformer users based on multidimensional time series data, used to implement the above-mentioned method for extracting load features of dedicated transformer users based on multidimensional time series data. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for extracting load features of dedicated transformer users based on multidimensional time series data provided below can be found in the limitations of the method for extracting load features of dedicated transformer users based on multidimensional time series data described above, and will not be repeated here.

[0092] In one exemplary embodiment, such as Figure 4 As shown, a device for extracting features of dedicated transformer user load based on multidimensional time series data is provided, including: a data point acquisition module 10, a weight fusion module 20, a data repair module 30, a feature extraction module 40, and a feature filtering module 50, wherein:

[0093] The data point acquisition module 10 is used to preprocess the collected raw dataset to obtain a reference dataset and to acquire the data points to be repaired in the reference dataset; the raw dataset includes the load raw dataset and the environment raw dataset.

[0094] The weighted fusion module 20 is used to obtain the spatial repair value and temporal repair value of the data points to be repaired, and to fuse the spatial repair value and temporal repair value based on the fusion weight to obtain the data repair value; the fusion weight is determined according to the fluctuation coefficient and average correlation coefficient of the data to be repaired.

[0095] The data repair module 30 is used to update the data values ​​corresponding to the data to be repaired to the data repair values, so as to obtain the repair dataset; the repair dataset includes the load repair dataset and the environment repair dataset.

[0096] The feature extraction module 40 is used to perform time-series alignment on the load repair dataset and the environmental repair dataset to obtain a multidimensional time-series dataset, and to extract the time-series feature set of the multidimensional time-series dataset; the time-series feature set includes statistical features, morphological features, transform domain features and environmental coupling features.

[0097] The feature filtering module 50 is used to filter the time series feature set to obtain the target feature set, and to build a spatiotemporal feature sample library of special transformer user load based on the target feature set.

[0098] In an exemplary embodiment, the data point acquisition module 10 is further configured to divide the reference dataset according to a preset time window; for the reference data points within the current time window, acquire the sequence mean and sequence standard deviation of the reference data points, and designate the reference data points whose deviation from the sequence mean exceeds a preset statistical threshold as statistical outliers; the preset statistical threshold is determined based on the sequence standard deviation; acquire the timestamp of the reference data points, and acquire the missing data points in the reference data points based on the timestamp; and designate the statistical outliers and missing data points as data points to be repaired.

[0099] In an exemplary embodiment, the weighted fusion module 20 is further configured to obtain candidate objects associated with the user to be repaired based on the user location of the user to be repaired corresponding to the data point to be repaired, obtain the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curve corresponding to the candidate object; filter the candidate objects according to the correlation coefficient to obtain similar objects, and perform a weighted average of the historical data values ​​of the similar objects at the time to be repaired corresponding to the data point to be repaired based on the correlation coefficient to obtain the spatial repair value.

[0100] In an exemplary embodiment, the weight fusion module 20 is further configured to obtain the corresponding repair time window based on the repair time corresponding to the data point to be repaired; input the repair sequence corresponding to the repair time window to the long short-term memory network model, and output the time repair value.

[0101] In an exemplary embodiment, the weight fusion module 20 is further configured to obtain a corresponding fluctuation time window based on the time to be repaired corresponding to the data point to be repaired, obtain the fluctuation coefficient based on the data points within the fluctuation time window, obtain the average correlation coefficient corresponding to the data point to be repaired, and obtain the fusion weight based on the fluctuation coefficient and the average correlation coefficient.

[0102] In an exemplary embodiment, the feature extraction module 40 is further configured to obtain a set of target time points based on the load repair dataset; for each target time point in the set of target time points, obtain a first original data point before the target time point and a second original data point after the target time point in the original environmental dataset; construct a polynomial function based on the first and second original data points, and obtain the environmental data interpolation corresponding to the target time point based on the polynomial function; update the environmental repair dataset based on all the environmental data interpolation, and perform time-series alignment between the load repair dataset and the updated environmental repair dataset to obtain a multidimensional time-series dataset.

[0103] In an exemplary embodiment, the feature selection module 50 is further configured to obtain the information coefficient between the time-series feature and the target variable for each time-series feature in the time-series feature set, perform a first selection on the time-series feature set based on the information coefficient to obtain a candidate feature set; use the time-series feature as input and the target variable as output, obtain the importance score corresponding to the time-series feature through a decision tree model, and perform a second selection on the time-series feature set based on the importance score to obtain a target feature set.

[0104] The modules in the aforementioned device for extracting the load characteristics of dedicated transformer users based on multidimensional time series data can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0105] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a method for extracting the characteristics of a special-purpose transformer user load based on multi-dimensional time-series data. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0106] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0107] In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps: preprocessing the collected raw dataset to obtain a reference dataset, and obtaining the data points to be repaired in the reference dataset; the raw dataset includes a load raw dataset and an environmental raw dataset; obtaining the spatial repair value and temporal repair value of the data points to be repaired respectively, and fusing the spatial repair value and temporal repair value based on the fusion weight to obtain the data repair value; the fusion weight is determined according to the fluctuation coefficient and average correlation coefficient of the data to be repaired; updating the data value corresponding to the data to be repaired to the data repair value to obtain the repair dataset; the repair dataset includes a load repair dataset and an environmental repair dataset; performing time-series alignment on the load repair dataset and the environmental repair dataset to obtain a multi-dimensional time-series dataset, and extracting the time-series feature set of the multi-dimensional time-series dataset; the time-series feature set includes statistical features, morphological features, transform domain features, and environmental coupling features; filtering the time-series feature set to obtain a target feature set, and constructing a special transformer user load spatiotemporal feature sample library based on the target feature set.

[0108] In one embodiment, the process of obtaining data points to be repaired in a reference dataset when the processor executes a computer program includes: dividing the reference dataset according to a preset time window; for the reference data points within the current time window, obtaining the sequence mean and sequence standard deviation of the reference data points, and identifying reference data points whose deviation from the sequence mean exceeds a preset statistical threshold as statistical outliers; the preset statistical threshold is determined based on the sequence standard deviation; obtaining the timestamps of the reference data points, and obtaining missing data points in the reference data points based on the timestamps; and identifying the statistical outliers and missing data points as data points to be repaired.

[0109] In one embodiment, the process of obtaining spatial repair values ​​when the processor executes a computer program includes: obtaining candidate objects associated with the user to be repaired based on the user location of the user to be repaired corresponding to the data point to be repaired; obtaining the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curve corresponding to the candidate object; filtering the candidate objects based on the correlation coefficient to obtain similar objects; and, based on the correlation coefficient, performing a weighted average of the historical data values ​​of the similar objects at the time to be repaired corresponding to the data point to be repaired to obtain the spatial repair value.

[0110] In one embodiment, the process of obtaining time-repair values ​​when the processor executes a computer program includes: obtaining a corresponding time window to be repaired based on the time point to be repaired corresponding to the data point to be repaired; inputting the sequence to be repaired corresponding to the time window to be repaired into a long short-term memory network model, and outputting the time-repair value.

[0111] In one embodiment, the process of obtaining the fusion weights involved when the processor executes a computer program includes: obtaining a corresponding fluctuation time window based on the time to be repaired corresponding to the data point to be repaired; obtaining the fluctuation coefficient based on the data points within the fluctuation time window; obtaining the average correlation coefficient corresponding to the data point to be repaired; and obtaining the fusion weights based on the fluctuation coefficient and the average correlation coefficient.

[0112] In one embodiment, the time-series alignment of a load repair dataset and an environmental repair dataset to obtain a multidimensional time-series dataset when the processor executes a computer program includes: obtaining a set of target time points based on the load repair dataset; for each target time point in the set of target time points, obtaining a first original data point before the target time point and a second original data point after the target time point in the original environmental dataset; constructing a polynomial function based on the first and second original data points, and obtaining environmental data interpolation corresponding to the target time point based on the polynomial function; updating the environmental repair dataset based on all environmental data interpolation, and performing time-series alignment between the load repair dataset and the updated environmental repair dataset to obtain a multidimensional time-series dataset.

[0113] In one embodiment, the process of a processor executing a computer program to filter a set of temporal features to obtain a target feature set includes: for each temporal feature in the set of temporal features, obtaining the information coefficient between the temporal feature and the target variable; performing a first screening of the temporal feature set based on the information coefficient to obtain a candidate feature set; using the temporal feature as input and the target variable as output, obtaining an importance score corresponding to the temporal feature through a decision tree model; and performing a second screening of the temporal feature set based on the importance score to obtain the target feature set.

[0114] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0115] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0116] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0117] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0118] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for extracting the load characteristics of dedicated transformer users based on multidimensional time series data, characterized in that, The method includes: The collected raw dataset is preprocessed to obtain a reference dataset, and the data points to be repaired in the reference dataset are obtained; the raw dataset includes the load raw dataset and the environment raw dataset; The spatial repair value and the temporal repair value of the data point to be repaired are obtained respectively, and the spatial repair value and the temporal repair value are fused based on the fusion weight to obtain the data repair value; the fusion weight is determined according to the fluctuation coefficient and the average correlation coefficient of the data to be repaired. The data values ​​corresponding to the data to be repaired are updated with the data repair values ​​to obtain a repair dataset; the repair dataset includes a load repair dataset and an environment repair dataset. The load repair dataset and the environment repair dataset are time-series aligned to obtain a multidimensional time-series dataset, and the time-series feature set of the multidimensional time-series dataset is extracted; the time-series feature set includes statistical features, morphological features, transform domain features, and environmental coupling features; The time-series feature set is filtered to obtain the target feature set, and a spatiotemporal feature sample library of dedicated transformer user load is constructed based on the target feature set.

2. The method according to claim 1, characterized in that, The step of obtaining the data points to be repaired in the reference dataset includes: The reference dataset is divided according to a preset time window; For reference data points within the current time window, the sequence mean and sequence standard deviation of the reference data points are obtained. Reference data points whose deviation from the sequence mean exceeds a preset statistical threshold are identified as statistical outliers. The preset statistical threshold is determined based on the sequence standard deviation. Obtain the timestamp of the reference data point, and obtain the missing data points in the reference data point based on the timestamp; The statistical outliers and the missing data points are designated as data points to be repaired.

3. The method according to claim 1, characterized in that, The process of obtaining the spatial repair value includes: Based on the user location of the user to be repaired corresponding to the data point to be repaired, obtain the candidate object associated with the user to be repaired, and obtain the correlation coefficient between the data curve to be repaired corresponding to the user to be repaired and the candidate data curve corresponding to the candidate object; The candidate objects are filtered according to the correlation coefficient to obtain similar objects. Based on the correlation coefficient, the historical data values ​​of the similar objects at the time to be repaired corresponding to the data point to be repaired are weighted and averaged to obtain the spatial repair value.

4. The method according to claim 1, characterized in that, The process of obtaining the time restoration value includes: Based on the time point to be repaired corresponding to the data point to be repaired, obtain the corresponding time window to be repaired; The sequence to be repaired corresponding to the time window to be repaired is input into the long short-term memory network model, and the time repair value is output.

5. The method according to claim 1, characterized in that, The process of obtaining the fusion weights includes: Based on the time to be repaired corresponding to the data point to be repaired, obtain the corresponding fluctuation time window, and obtain the fluctuation coefficient based on the data points within the fluctuation time window; Obtain the average correlation coefficient corresponding to the data point to be repaired, and obtain the fusion weight based on the fluctuation coefficient and the average correlation coefficient.

6. The method according to claim 1, characterized in that, The process of time-series aligning the load repair dataset and the environmental repair dataset yields a multidimensional time-series dataset, including: Obtain the target time point set based on the load repair dataset; For each target time point in the set of target time points, obtain a first raw data point before the target time point and a second raw data point after the target time point in the original environmental dataset; Based on the first original data point and the second original data point, a polynomial function is constructed, and environmental data interpolation corresponding to the target time point is obtained according to the polynomial function; The environmental remediation dataset is updated based on all environmental data interpolation, and the load remediation dataset and the updated environmental remediation dataset are time-series aligned to obtain a multidimensional time-series dataset.

7. The method according to claim 1, characterized in that, The step of filtering the time-series feature set to obtain the target feature set includes: For each time-series feature in the time-series feature set, the information coefficient between the time-series feature and the target variable is obtained, and the time-series feature set is first filtered according to the information coefficient to obtain a candidate feature set; Using the time-series features as input and the target variable as output, an importance score corresponding to the time-series features is obtained through a decision tree model. Based on the importance score, the time-series feature set is further filtered to obtain the target feature set.

8. A device for extracting load characteristics of dedicated transformer users based on multidimensional time series data, characterized in that, The device includes: The data point acquisition module is used to preprocess the collected raw dataset to obtain a reference dataset, and to acquire the data points to be repaired in the reference dataset; the raw dataset includes the load raw dataset and the environment raw dataset. The weighted fusion module is used to obtain the spatial repair value and the temporal repair value of the data point to be repaired, and to fuse the spatial repair value and the temporal repair value based on the fusion weight to obtain the data repair value; the fusion weight is determined according to the fluctuation coefficient and the average correlation coefficient of the data to be repaired. The data repair module is used to update the data value corresponding to the data to be repaired to the data repair value, thereby obtaining a repair dataset; the repair dataset includes a load repair dataset and an environment repair dataset; The feature extraction module is used to perform time-series alignment on the load repair dataset and the environment repair dataset to obtain a multi-dimensional time-series dataset, and to extract the time-series feature set of the multi-dimensional time-series dataset; the time-series feature set includes statistical features, morphological features, transform domain features and environmental coupling features; The feature filtering module is used to filter the time-series feature set to obtain the target feature set, and to construct a spatiotemporal feature sample library of dedicated transformer user load based on the target feature set.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images