Method for identifying abnormity of electric power measurement data based on PCA-IF model

Abstract

The invention discloses a PCA-IF model-based electric power measurement data abnormity identification method, which comprises the following steps of data preprocessing: acquiring multi-dimensional original data of voltage, current, power factor, environment temperature and operation time of an electric power measurement device, and performing interpolation filling based on time sequence characteristics and spatial correlation on missing values; feature dimension reduction: extracting orthogonal principal components from the preprocessed data by using principal component analysis (PCA), and selecting the number of the principal components after dimension reduction according to the cumulative variance contribution rate; building an anomaly detection model: building a plurality of random trees based on an isolated forest algorithm, and dynamically adjusting segmentation points of each tree according to the principal component characteristics after dimension reduction; aiming at the physical characteristics of the electric power measurement data, setting a characteristic segmentation range in combination with an industrial standard; and abnormal score calculation: calculating an abnormal score according to a weighting result of the path length of the sample in the tree and the principal component variance contribution rate, and judging that the sample with the score higher than a set threshold value is abnormal.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-dimensional metering data anomaly detection in intelligent operation and maintenance of power systems, and in particular to a method for identifying power metering data anomalies based on a PCA-IF model. Background Art

[0002] Driven by smart grids and advanced metering systems (AMI), the scale of electricity metering data has surged, but existing anomaly detection methods still have the following bottlenecks: Data quality flaws: Voltage, current, and other data collected by metering devices are susceptible to environmental interference (such as temperature fluctuations and equipment failures), resulting in widespread missing values ​​or noise. For example, the solution in publication number CN119354254A directly uses difference calculations and threshold comparisons without cleaning or filling the raw data. As the missing data rate increases, model error increases significantly (see step 2).

[0003] Inefficient high-dimensional data processing: Power data contains multi-dimensional features such as voltage, current, and power factor. Directly applying traditional algorithms (such as line chart analysis and statistical thresholds) is susceptible to interference from redundant features. CN119354254A uses line chart trends to identify anomalies, which improves intuitiveness but does not address the problem of misjudgment caused by the coupling of multi-dimensional data (see step 1).

[0004] Uneven distribution of anomalies: Abnormal electricity usage samples account for a low proportion and have widely varying characteristics, making it difficult for conventional methods (such as statistical thresholding) to capture implicit patterns. CN119354254A uses a fixed threshold to determine anomalies, but fails to consider the dynamic changes in data distribution under different operating conditions (such as the nonlinear impact of temperature on metering errors).

[0005] Limitations of existing technologies Existing solutions, such as CN119354254A, optimize detection logic through difference calculation and temperature compensation, but their core flaws are: Lack of data preprocessing: No cleaning and filling rules were designed for power metering data, resulting in insufficient robustness of the model in data-missing scenarios. Insufficient algorithm adaptability: It relies on manual threshold setting and intuitive analysis, does not introduce dimensionality reduction and unsupervised learning mechanisms, and has difficulty processing high-dimensional and nonlinear data. Summary of the Invention

[0006] To address the shortcomings and deficiencies of existing technologies (such as lack of data preprocessing, interference from high-dimensional redundant features, and high false alarm rate of anomaly detection), the present invention provides a method and system for detecting anomalies in power metering data based on collaborative optimization of PCA and isolation forest. The core technologies and innovations of the system include: Multi-dimensional data adaptation preprocessing mechanism: Targeting the temporal and spatial correlation of power metering data, differentiated rules such as cubic spline interpolation and adjacent device data filling are used to address model errors caused by missing values. Dimensionality reduction optimization driven by the physical meaning of principal components: Orthogonal principal components are extracted based on the cumulative variance contribution rate, and features strongly correlated with voltage and power factor are retained through load matrix analysis, improving the industry interpretability of the reduced features. Isolation forest dynamic segmentation based on domain knowledge fusion: Integrating with power industry standards (such as GB / T 12325-2008 voltage deviation range and reactive power compensation threshold), the tree structure's segmentation points and depth are dynamically adjusted to suppress overfitting of high-dimensional data. Contribution-weighted anomaly scoring mechanism: Dynamic weights are assigned based on the variance contribution rate of the principal components. Anomaly scores are calculated based on the combined path length and weight, achieving accurate anomaly determination and a low false alarm rate.

[0007] This solution effectively solves the limitations of traditional methods in power scenarios, such as data redundancy and uneven distribution of abnormalities, and provides reliable technical support for optimizing line loss rates and preventing and controlling electricity theft risks in smart grids.

[0008] The present invention specifically adopts the following technical solutions: A method for identifying anomalies in power metering data based on a PCA-IF model, characterized by comprising the following steps: Data preprocessing: Collect multi-dimensional raw data on the voltage, current, power factor, ambient temperature, and operating time of the power metering device, and interpolate and fill missing values ​​based on temporal characteristics and spatial correlation; Feature dimensionality reduction: Use principal component analysis (PCA) to extract orthogonal principal components from preprocessed data, and select the number of principal components after dimensionality reduction based on the cumulative variance contribution rate; Anomaly detection model construction: Multiple random trees are constructed based on the isolation forest algorithm, and the split points of each tree are dynamically adjusted according to the principal component characteristics after dimensionality reduction; Based on the physical characteristics of power metering data and industry standards, the feature segmentation range is set; Anomaly score calculation: The anomaly score is calculated based on the weighted result of the sample's path length in the tree and the principal component variance contribution rate. Samples with scores higher than the set threshold are considered anomalies.

[0009] Furthermore, according to the variance contribution rate of the principal components, higher segmentation priorities are assigned to principal components with high contribution rates, and the segmentation depth of principal components with low contribution rates is limited.

[0010] Furthermore, the industry standards include: The voltage characteristic segmentation range is set based on the power system voltage tolerance standard; The split range of the power factor characteristic is set based on the reactive power compensation threshold.

[0011] Preferably, the allowable voltage deviation standard is the ±7% deviation range of the three-phase power supply voltage of 20 kV and below specified in "GB / T 12325-2008 Power Quality Supply Voltage Deviation".

[0012] Furthermore, the anomaly score threshold was determined through experimental verification to balance detection sensitivity and false alarm rate, with a value range of 0.7-0.9.

[0013] Furthermore, the interpolation filling includes: Cubic spline interpolation is used for voltage and current time series data; The ambient temperature data is filled with the average data of the adjacent metering devices in the same power supply area during the same period.

[0014] And, a system for identifying anomalies in power metering data based on a PCA-IF model, comprising: a data acquisition module configured to collect multi-dimensional raw data of voltage, current, power factor, ambient temperature and operating time of the power metering device; A preprocessing module is configured to perform interpolation to fill missing values ​​based on temporal characteristics and spatial correlation; The feature dimensionality reduction module is configured to extract orthogonal principal components using principal component analysis (PCA) and select the number of principal components after dimensionality reduction based on the cumulative variance contribution rate; Anomaly detection model module, including: An isolation forest construction unit, configured to dynamically adjust the split point of each random tree based on the principal component features after dimensionality reduction; a segmentation range setting unit configured to set a feature segmentation range in combination with physical characteristics of the power metering data and industry standards; The anomaly determination module is configured to calculate an anomaly score based on a weighted result of the path length of the sample in the tree and the principal component variance contribution rate, and determine that the sample with a score higher than a set threshold is an anomaly.

[0015] Furthermore, the preprocessing module includes: A time series interpolation submodule is configured to fill missing values ​​in voltage and current time series data using cubic spline interpolation; A spatial correlation filling submodule is configured to fill in the missing values ​​of ambient temperature based on the mean data of adjacent metering devices in the same power supply area during the same period; The regression prediction submodule is configured to use a random forest regression model to fill in the missing values ​​of power factor and running time.

[0016] As a preferred solution, the cumulative variance contribution rate in the feature dimension reduction module is ≥85%, and the principal components that are strongly correlated with voltage and power factor are preferentially retained.

[0017] Furthermore, the cumulative variance contribution rate in the feature dimension reduction module is ≥85%, and the principal components that are strongly correlated with voltage and power factor are preferentially retained.

[0018] Furthermore, in the isolated forest construction unit: The number of random trees is 100; The maximum depth of each tree is limited to 10 levels; In the abnormality determination module: When the principal component variance contribution rate is ≥10%, the weight is 1.2; When the variance contribution rate is 5%-10%, the weight is 1.0; When the variance contribution rate is less than 5%, the weight is 0.8; The anomaly score threshold is 0.8.

[0019] Furthermore, it also includes: a visualization module configured to generate an abnormality detection report and mark the main component contribution ratio of voltage and power factor abnormalities.

[0020] And, an electronic device includes a memory, a processor, and a computer program stored in the memory and runnable on the processor, characterized in that the processor implements the steps of the above method when executing the program.

[0021] A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0022] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects: Improve the accuracy and robustness of anomaly detection: Through a multi-dimensional data adaptation preprocessing mechanism, combined with time series interpolation and spatial correlation filling rules, it effectively solves the problems of missing data and noise interference, ensures the integrity and reliability of input data, and lays a high-quality foundation for subsequent model training; Optimizing high-dimensional data processing efficiency: Based on the synergistic mechanism of principal component analysis (PCA) and isolation forest (IF), orthogonal principal component dimensionality reduction is used to reduce redundant features. At the same time, by dynamically adjusting the tree structure and segmentation strategy, computational complexity is significantly reduced and detection efficiency is improved. Enhanced domain adaptability and detection accuracy: By setting feature segmentation ranges (such as voltage deviation and power factor thresholds) in accordance with power industry standards, model parameters are deeply coupled with the physical characteristics of power scenarios, avoiding misjudgments caused by parameter generalization in traditional methods. Achieve accurate anomaly determination and low false alarm rate: Through an anomaly scoring mechanism weighted by the principal component variance contribution rate, it dynamically balances detection sensitivity and false alarm rate, ensuring that the identification of abnormal samples is neither overly conservative nor rigorous. Expand the practicality of the technical solution: The tree structure depth limit and dynamic weight allocation designs in the preferred solution further suppress the risk of overfitting and improve the model's generalization ability, making it suitable for power metering datasets of different sizes. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments: Figure 1 This is a flow chart of a method for identifying abnormalities in power metering data based on the PCA-IF model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] Hereinafter, specific embodiments of the present application will be described in detail with reference to the accompanying drawings. Based on these detailed descriptions, those skilled in the art will be able to clearly understand the present application and implement the present application. Without violating the principles of the present application, the features of different embodiments may be combined to obtain new implementations, or certain features of certain embodiments may be substituted to obtain other preferred implementations.

[0025] To make the features and advantages of the present invention more clearly understood, the following embodiments are specifically described in detail with reference to the accompanying drawings. This embodiment of the present invention provides a method for identifying anomalies in electricity metering data based on the PCA-IF model. This method collects raw data on characteristic factors such as voltage, current, power factor, ambient temperature, and operating time of metering devices, uses PCA for feature dimensionality reduction, and utilizes the IF algorithm to identify anomalies in metering data. The PCA-IF model is then used to identify anomalies in electricity metering data. The method preprocesses the raw data set through data cleaning and missing value filling. Principal component analysis (PCA) is then used to reduce dimensionality and analyze feature correlations. The isolation forest (IF) algorithm is then used to randomly construct multiple trees, with each tree randomly selecting features and split points to partition the data space. The model uses an anomaly scoring formula to identify anomalies in the electricity metering device.

[0026] The following is a detailed description of the design of the embodiment of the present invention: 1.1 Principal Component Analysis (PCA) algorithm: Principal Component Analysis (PCA) uses a series of mathematical transformations to reduce the original high-dimensional feature vectors into low-dimensional, mutually orthogonal feature vectors. The weakened relationships between low-dimensional feature vectors significantly reduce the degree of data duplication, which not only ensures the validity of the original data but also reduces resource requirements during training, resulting in better detection results. The calculation steps are as follows: (1) The original data matrix is , indicating that there are m pieces of data, including n dimensions; (2) Yes Normalize according to the following formula so that the expectations of different eigenvectors are all equal to zero; Where, 、 、 They represent the original data before normalization, the maximum value of the dimension where the original data is located, and the minimum value of the dimension where the original data is located, respectively. Represents the normalized eigenvalue.

[0027] (3) Calculation The covariance matrix of is: ; (4) Calculate the eigenvalues ​​and eigenvectors of the covariance matrix Y; (5) Arrange the eigenvectors by row according to the eigenvalues ​​of the covariance matrix Y, and take the first Z rows to obtain a new matrix ; (6) It is the low-dimensional data set obtained after PCA processing.

[0028] Isolation Forest (Isolaon tiForest) algorithm: The anomaly identification model uses the Isolation Forest algorithm, which randomly partitions data and creates a tree structure, using the degree of data isolation as the basis for determining anomalies. The algorithm gradually constructs a binary tree (i-Tree) by randomly selecting data features and random partitioning points within the feature range. The path length of a sample in the tree is used to measure its degree of anomaly, with a shorter average path length indicating a higher degree of anomaly. The partitioning process is random, and normal samples require multiple partitions to be isolated. Compared to normal samples, outliers are fewer in number and have larger eigenvalue variations. This makes outliers more likely to be randomly partitioned into fewer subspaces, resulting in shorter paths in the tree. By randomly constructing multiple i-Trees and calculating the average path length of each sample in these trees, samples with higher anomaly scores are ultimately considered outliers.

[0029] Method for identifying abnormalities in power metering data based on PCA-IF model: 1.3.1 Data Preprocessing In order to reduce the impact of data on the proposed model, this embodiment preprocesses the original data of characteristic factors such as voltage, current, power factor, ambient temperature, and operating time of the metering devices in the power system, and fills in some missing data to ensure the accuracy of the model operation.

[0030] As a preferred solution, in this embodiment: For time series characteristics such as voltage and current, the cubic spline interpolation method is used to fill in missing values; For spatial correlation features such as ambient temperature, fill in the gaps based on the mean data of adjacent metering devices in the same power supply area during the same period; For non-time series features such as power factor and operating time, random forest regression model is used to predict and fill in.

[0031] Through multi-dimensional data classification processing, differentiated filling rules are formed according to the characteristics of different dimensions of power data (temporal, spatial correlation, and static attributes) to improve the preprocessing accuracy.

[0032] Taking into account the adaptability of power scenarios, this embodiment combines the characteristics of data collection in the power industry (such as regional deployment and strong time series correlation) to avoid the error accumulation caused by general interpolation methods.

[0033] Building a predictive model This embodiment proposes a method for identifying abnormalities in power metering data by combining a principal component analysis (PCA) model with an isolation forest algorithm to establish a PCA-IF model. The method uses characteristic factor data such as voltage, current, power factor, ambient temperature, and operating time of the power system, and uses a principal component analysis (PCA) model to first reduce the dimension of training sample features to improve the abnormality recognition and detection effect. The isolation forest algorithm is then used to identify abnormal metering data again. This model can more accurately identify abnormal data and reduce errors. The model includes data preprocessing, PCA for feature dimensionality reduction, and IF algorithm for identifying abnormal metering data. The flow chart of the method for identifying abnormalities in power metering data using the PCA-IF model is shown below. Figure 1 shown.

[0034] The PCA-IF model identifies abnormal power metering data by preprocessing the original data, filling in abnormal data and null value data, and then performing feature selection. PCA is used for feature dimensionality reduction, and the IF algorithm can identify abnormal metering data more accurately.

[0035] As a preferred solution, in this embodiment: After extracting the orthogonal principal components, their physical meaning is analyzed based on the principal component loading matrix, for example: The first principal component corresponds to the comprehensive characterization of voltage fluctuation and power factor; The second principal component corresponds to the coupled effect of ambient temperature and equipment operating time.

[0036] The principal components with cumulative variance contribution rate ≥ 85% were selected, and the principal components with strong correlation with voltage and power factor were retained first.

[0037] This embodiment uses physical meaning to drive dimensionality reduction, converting the mathematical dimensionality reduction of PCA into feature interpretation in the power scenario, ensuring that the features after dimensionality reduction have clear industry interpretability.

[0038] Redundant features are eliminated in a targeted manner through load matrix analysis to avoid retaining principal components that are not relevant to anomaly detection (such as static attributes such as device ID).

[0039] Based on PCA, the features of the training samples are classified into leaf nodes with a score corresponding to each tree. The isolation forest algorithm consists of multiple random trees. Each tree divides the data space by the features and split points obtained by PCA dimensionality reduction. The model uses the anomaly score formula to measure the degree of abnormality of the sample as follows: in, represents the abnormality score of sample x, is the distance from the sample to the nearest leaf node, is the average height of the tree. The value range of the anomaly score is [0, 1]. Samples with scores close to 1 are considered anomalies, and samples close to 0 are normal points.

[0040] As a preferred solution, in this embodiment: Weights can be assigned based on the variance contribution of the principal components, for example: The principal component weight for variance contribution > 10% is 1.2; The principal component with a variance contribution rate of 5%-10% has a weight of 1.0; The principal component weight for variance contribution rate <5% is 0.8; The anomaly score is dynamically adjusted through the importance of the principal component to improve the detection sensitivity of features with high contribution rates.

[0041] Specifically, the training process of Isolation Forest includes the following steps: (1) Parameter setting: Set the number of trees in the forest (e.g., 100) and the maximum depth of each tree to ensure that the model can run efficiently.

[0042] Tree structure construction: Select features and split points (such as voltage or power factor) through PCA dimensionality reduction, and build each isolated tree (such as splitting the voltage range [210, 240]).

[0043] (2) Path length calculation: For each sample in the training data, calculate the path length to the leaf node.

[0044] As a preferred solution, in this embodiment: For the principal component features after dimensionality reduction, set the segmentation priority according to their physical meaning: The split points of the voltage-related principal components are based on industry standards (e.g., ±7% deviation range); The split point of the power factor-related principal component is based on the reactive compensation threshold (0.92-1.0); When constructing the isolation forest, the maximum depth of each tree is limited to 10 layers to avoid overfitting of high-dimensional data.

[0045] This embodiment combines the physical meaning of principal components with a tree segmentation strategy to address the misjudgment problem caused by traditional random segmentation of isolation forests. Furthermore, by limiting tree depth, it optimizes the efficiency of high-dimensional data processing, meeting the requirements of large-scale and real-time power data.

[0046] (3) Anomaly score calculation: Calculate the anomaly score of the sample based on the path length and average height.

[0047] The PCA-IF model for identifying anomalies in power metering data mainly uses principal component analysis (PCA) to effectively reduce data dimensions and improve the model's computational efficiency and accuracy. The isolation forest algorithm (IF) successfully solves the problem of uneven distribution of abnormal samples by virtue of its random partitioning characteristics, and uses reasonable parameter optimization to achieve accurate identification of anomalies and a low false alarm rate of normal points.

[0048] In summary, the improvements of this embodiment over the prior art mainly include: 1. Work efficiency: Data models can be used to identify abnormal data, conduct targeted on-site maintenance of metering equipment, and focus on abnormal users of metering equipment, significantly reducing the risk of electricity theft by users.

[0049] 2. Economical: Based on the subjective data characteristics of users, machine learning algorithms are introduced to identify potential users with abnormal data and intervene in on-site operation and maintenance early, which can effectively reduce the occurrence of user complaints and improve power supply reliability.

[0050] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0051] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, performs the above-described method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electrical, magnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0052] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0053] The above shows and describes the basic principles, main features and advantages of the present disclosure. Those skilled in the art should understand that the present disclosure is not limited to the above embodiments. The above embodiments and descriptions are merely illustrative of the principles of the present disclosure. Various changes and improvements may be made to the present disclosure without departing from the spirit and scope of the present disclosure, and such changes and improvements shall fall within the scope of the present disclosure.

[0054] The present invention is not limited to the above-described preferred embodiment. Based on the present invention, anyone can derive various other methods for identifying anomalies in power metering data based on the PCA-IF model. All equivalent changes and modifications made within the scope of the present invention are intended to fall within the scope of the present invention.

Claims

1. A method for identifying abnormalities in power metering data based on the PCA-IF model, characterized in that: The following steps are involved: Data preprocessing: Collect multi-dimensional raw data on the voltage, current, power factor, ambient temperature, and operating time of the power metering device, and interpolate and fill missing values ​​based on temporal characteristics and spatial correlation; Feature dimensionality reduction: Use principal component analysis (PCA) to extract orthogonal principal components from preprocessed data, and select the number of principal components after dimensionality reduction based on the cumulative variance contribution rate; Anomaly detection model construction: Multiple random trees are constructed based on the isolation forest algorithm, and the split points of each tree are dynamically adjusted according to the principal component characteristics after dimensionality reduction; Based on the physical characteristics of power metering data and industry standards, the feature segmentation range is set; Anomaly score calculation: The anomaly score is calculated based on the weighted result of the sample's path length in the tree and the principal component variance contribution rate. Samples with scores higher than the set threshold are considered anomalies.

2. The method for identifying anomalies in power metering data based on the PCA-IF model according to claim 1, characterized in that: According to the variance contribution rate of the principal component, a higher segmentation priority is assigned to the principal component with high contribution rate, and the segmentation depth of the principal component with low contribution rate is limited.

3. The method for identifying anomalies in power metering data based on the PCA-IF model according to claim 1, characterized in that: The industry standards mentioned include: The voltage characteristic segmentation range is set based on the power system voltage tolerance standard; The split range of the power factor characteristic is set based on the reactive power compensation threshold.

4. The method for identifying anomalies in power metering data based on the PCA-IF model according to claim 1, characterized in that: The anomaly score threshold is determined through experimental verification to balance detection sensitivity and false alarm rate.

5. The method for identifying anomalies in power metering data based on the PCA-IF model according to claim 1, characterized in that: The interpolation filling includes: Cubic spline interpolation is used for voltage and current time series data; The ambient temperature data is filled with the average data of the adjacent metering devices in the same power supply area during the same period.

6. A system for identifying abnormalities in power metering data based on the PCA-IF model, characterized in that: include: a data acquisition module configured to collect multi-dimensional raw data of voltage, current, power factor, ambient temperature and operating time of the power metering device; A preprocessing module is configured to perform interpolation to fill missing values ​​based on temporal characteristics and spatial correlation; The feature dimensionality reduction module is configured to extract orthogonal principal components using principal component analysis (PCA) and select the number of principal components after dimensionality reduction based on the cumulative variance contribution rate; Anomaly detection model module, including: An isolation forest construction unit, configured to dynamically adjust the split point of each random tree based on the principal component features after dimensionality reduction; a segmentation range setting unit configured to set a feature segmentation range in combination with physical characteristics of the power metering data and industry standards; The anomaly determination module is configured to calculate an anomaly score based on a weighted result of the path length of the sample in the tree and the principal component variance contribution rate, and determine that the sample with a score higher than a set threshold is an anomaly.

7. The system for identifying anomalies in power metering data based on the PCA-IF model according to claim 6, characterized in that: The pre-processing module comprises: A time series interpolation submodule is configured to fill missing values ​​in voltage and current time series data using cubic spline interpolation; A spatial correlation filling submodule is configured to fill in the missing values ​​of ambient temperature based on the mean data of adjacent metering devices in the same power supply area during the same period; The regression prediction submodule is configured to use a random forest regression model to fill in the missing values ​​of power factor and running time.

8. The system for identifying anomalies in power metering data based on the PCA-IF model according to claim 6, characterized in that: The cumulative variance contribution rate in the feature dimension reduction module is ≥85%, and the principal components that are strongly correlated with voltage and power factor are preferentially retained.

9. The system for identifying anomalies in power metering data based on the PCA-IF model according to claim 6, characterized in that: In the isolation forest construction unit: The number of random trees is 100; The maximum depth of each tree is limited to 10 levels; In the abnormality determination module: When the principal component variance contribution rate is ≥10%, the weight is 1.2; When the variance contribution rate is 5%-10%, the weight is 1.0; When the variance contribution rate is less than 5%, the weight is 0.8; The anomaly score threshold is 0.

8.

10. The system for identifying abnormalities in power metering data based on the PCA-IF model according to claim 6, characterized in that: Also includes: The visualization module is configured to generate an anomaly detection report and mark the main component contribution ratios of voltage and power factor anomalies.

Images