Abnormal data detection method and system based on data mining

Abstract

The invention provides an abnormal data detection method and system based on data mining, and the method comprises the steps: carrying out the abnormal feature extraction of obtained power utilization data in a power system, and obtaining the abnormal feature data corresponding to the power utilization data; performing classification processing on the abnormal feature data by using a clustering algorithm to obtain a center membership degree of each data point in the abnormal feature data; according to the center membership degree of each data point in the abnormal feature data, outputting an abnormal score corresponding to each data point in the abnormal feature data by using a pre-constructed abnormal detection model; obtaining abnormal data in the power consumption data according to the abnormal score corresponding to each data point in the abnormal feature data; according to the method, the core features related to the anomaly can be rapidly extracted from the large-scale power consumption data, and the structured classification of different anomaly modes is constructed, so that the anomaly detection efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and analysis, and in particular to an abnormal data detection method and system based on data mining. Background Art

[0002] At present, with the advent of the big data era, indicator reports in various business systems have become important tools for measuring business health and monitoring business operation status. However, due to the diversity of data sources and complex processing procedures, indicator reports often have problems such as missing data and anomalies, which affect data quality and the accuracy of business decisions. In existing technologies, most rely on manual review or simple rule judgment to identify abnormal data, which is inefficient and difficult to cover all abnormal situations. This method has many technical problems in actual operation:

[0003] First, the recognition efficiency is low. Manual review of abnormal data requires a lot of manpower and time, especially when the amount of data is huge, the efficiency is extremely low. In addition, manual review is easily affected by personal experience and subjective judgment, resulting in inconsistent recognition results.

[0004] Second, the coverage is not comprehensive. Although automated judgment based on simple rules can improve recognition efficiency to a certain extent, it is often difficult to cover all potential abnormal situations due to the limitations of the rules. Some complex or hidden abnormal data may be missed, thus affecting the overall improvement of data quality.

[0005] Third, the false alarm rate is high. Due to the mechanical and inflexible nature of rule judgment, when there is a slight fluctuation or change in the data, the rule may be triggered, resulting in a false alarm. This not only increases the workload of data processing, but may also cause unnecessary interference and misleading to business decisions.

[0006] In summary, the existing manual review or simple rule-based judgment methods to identify abnormal data have technical problems such as low recognition efficiency, incomplete coverage, and high false alarm rate. Summary of the invention

[0007] In order to solve the technical problems in the prior art of identifying abnormal data by manual review or simple rule judgment due to the diversification and complexity of electricity consumption data in the power system, such as low recognition efficiency, incomplete coverage and high false alarm rate, the present invention proposes an abnormal data detection method based on data mining, including:

[0008] Extracting abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data;

[0009] Using a clustering algorithm to classify the abnormal feature data to obtain the central membership of each data point in the abnormal feature data;

[0010] Outputting the anomaly score corresponding to each data point in the abnormal feature data using a pre-built anomaly detection model according to the central membership of each data point in the abnormal feature data;

[0011] Obtaining abnormal data in the power consumption data according to an abnormal score corresponding to each data point in the abnormal feature data;

[0012] Among them, the anomaly detection model is obtained by training the graph neural network using the historical anomaly feature data in the power system as training data.

[0013] Optionally, the extracting abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data includes:

[0014] Extracting features of the voltage deviation of the acquired power consumption data in the power system to obtain deviation feature data of the power consumption data;

[0015] Extracting the power consumption characteristics of the power consumption data to obtain power consumption characteristic data of the power consumption data;

[0016] Extracting features of the load anomaly of the power consumption data to obtain load feature data of the power consumption data;

[0017] Data integration is performed based on the deviation characteristic data, the power consumption characteristic data and the load characteristic data to obtain abnormal characteristic data corresponding to the power consumption data.

[0018] Optionally, the using a clustering algorithm to classify the abnormal feature data to obtain the central membership of each data point in the abnormal feature data includes:

[0019] Using a clustering algorithm to classify the abnormal feature data, and obtain a cluster center corresponding to each data point in the abnormal feature data;

[0020] According to the cluster center corresponding to each data point in the abnormal feature data, the central membership between each data point and its corresponding cluster center is calculated using the basic probability assignment method.

[0021] Optionally, the anomaly detection model includes the following training process:

[0022] Obtain historical abnormal feature data;

[0023] Performing cluster analysis on the historical abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the historical abnormal feature data;

[0024] Using the central membership of each data point in the historical abnormal feature data as input of training data;

[0025] Using the anomaly score of each data point in the historical anomaly feature data as output of training data;

[0026] Based on the input and output of the training data, the graph neural network is trained to obtain an anomaly detection model.

[0027] Optionally, the power consumption data in the power system includes the following acquisition methods:

[0028] Extracting power consumption index data from a preset database of the power system;

[0029] Performing data preprocessing on the electricity consumption index data to obtain processed index data;

[0030] The scale space mapping technology is used to perform spatial mapping on the processing index data to obtain the processing index data in a unified scale space, and the processing index data in the unified scale space is obtained as the power consumption data in the power system.

[0031] Optionally, the electricity consumption index data includes one or more of the following: voltage data, current data, power factor data, electricity load data, electricity consumption data and electricity consumption period data.

[0032] Optionally, obtaining abnormal data in the power consumption data according to an abnormal score corresponding to each data point in the abnormal feature data includes:

[0033] Using a quantile-based dynamic threshold method, the abnormal score of each data point in the abnormal feature data is classified to generate a classification label corresponding to each data point;

[0034] According to the classification label corresponding to each data point, a time series anomaly detection algorithm is used to perform sequence analysis on the abnormal feature data to obtain an abnormal sequence structure in the abnormal feature data;

[0035] A spatial correlation analysis algorithm based on geographic space is used to perform spatial correlation analysis on the abnormal sequence structure, and the abnormal data in the electricity consumption data is output.

[0036] Optionally, performing sequence analysis on the abnormal feature data using a time series anomaly detection algorithm according to the classification label corresponding to each data point to obtain an abnormal sequence structure in the abnormal feature data includes:

[0037] According to the classification label corresponding to each data point, each data point in the abnormal feature data is reintegrated in time series to obtain integrated abnormal feature data;

[0038] A time series anomaly detection algorithm is used to perform sequence analysis on the integrated abnormal feature data to obtain an abnormal sequence structure in the abnormal feature data.

[0039] Based on the same inventive concept, the present invention also provides an abnormal data detection system based on data mining, comprising:

[0040] A feature extraction module is used to extract abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data;

[0041] A clustering analysis module, used to classify the abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the abnormal feature data;

[0042] An anomaly detection module, used to output an anomaly score corresponding to each data point in the abnormal feature data according to the central membership of each data point in the abnormal feature data and using a pre-built anomaly detection model;

[0043] an abnormality determination module, used to obtain abnormal data in the power consumption data according to the abnormality score corresponding to each data point in the abnormal characteristic data;

[0044] Among them, the anomaly detection model is obtained by training the graph neural network using the historical anomaly feature data in the power system as training data.

[0045] Optionally, the feature extraction module includes:

[0046] The deviation feature extraction submodule is used to extract the features of the voltage deviation of the power consumption data acquired in the power system to obtain the deviation feature data of the power consumption data;

[0047] An electricity usage feature extraction submodule, used to extract the electricity usage features of the electricity usage data to obtain electricity usage feature data of the electricity usage data;

[0048] A load anomaly feature extraction submodule is used to extract features of the load anomaly of the power consumption data to obtain load feature data of the power consumption data;

[0049] The feature integration submodule is used to integrate the deviation feature data, the power consumption feature data and the load feature data to obtain abnormal feature data corresponding to the power consumption data.

[0050] Optionally, the cluster analysis module includes:

[0051] A classification processing submodule is used to classify the abnormal feature data using a clustering algorithm to obtain a cluster center corresponding to each data point in the abnormal feature data;

[0052] The membership calculation submodule is used to calculate the central membership between each data point and its corresponding cluster center using the basic probability assignment method according to the cluster center corresponding to each data point in the abnormal feature data.

[0053] Optionally, the detection system further includes: a model building module, which is used to:

[0054] Obtain historical abnormal feature data;

[0055] Performing cluster analysis on the historical abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the historical abnormal feature data;

[0056] Using the central membership of each data point in the historical abnormal feature data as input of training data;

[0057] Using the anomaly score of each data point in the historical anomaly feature data as output of training data;

[0058] Based on the input and output of the training data, the graph neural network is trained to obtain an anomaly detection model.

[0059] Optionally, the detection system further includes: a data acquisition module, configured to:

[0060] Extracting power consumption index data from a preset database of the power system;

[0061] Performing data preprocessing on the electricity consumption index data to obtain processed index data;

[0062] The scale space mapping technology is used to perform spatial mapping on the processing index data to obtain the processing index data in a unified scale space, and the processing index data in the unified scale space is obtained as the power consumption data in the power system.

[0063] Optionally, the electricity consumption index data includes one or more of the following: voltage data, current data, power factor data, electricity load data, electricity consumption data and electricity consumption period data.

[0064] Optionally, the abnormality determination module includes:

[0065] An anomaly classification submodule, used to classify the anomaly score of each data point in the abnormal feature data using a dynamic threshold method based on quantiles, and generate a classification label corresponding to each data point;

[0066] A time anomaly detection submodule is used to perform sequence analysis on the abnormal feature data using a time series anomaly detection algorithm according to the classification label corresponding to each data point, so as to obtain an abnormal sequence structure in the abnormal feature data;

[0067] The spatial correlation analysis submodule is used to perform spatial correlation analysis on the abnormal sequence structure by using a correlation analysis algorithm based on geographic space, and output abnormal data in the power consumption data.

[0068] Optionally, the time anomaly detection submodule includes:

[0069] A time series integration unit, configured to reintegrate each data point in the abnormal feature data according to a time series according to a classification label corresponding to each data point, to obtain integrated abnormal feature data;

[0070] The time series analysis unit is used to perform sequence analysis on the integrated abnormal feature data using a time series anomaly detection algorithm to obtain an abnormal sequence structure in the abnormal feature data.

[0071] In another aspect, the present invention further provides an electronic device, comprising: at least one processor and a memory; the memory and the processor are connected via a bus;

[0072] The memory is used to store one or more programs;

[0073] When the one or more programs are executed by the at least one processor, the above-mentioned abnormal data detection method based on data mining is implemented.

[0074] On the other hand, the present invention further provides a computer-readable storage medium having an execution program stored thereon, and when the execution program is executed, the above-mentioned abnormal data detection method based on data mining is implemented.

[0075] Compared with the prior art, the present invention has the following beneficial effects:

[0076] The present invention provides an abnormal data detection method and system based on data mining, comprising: extracting abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data; classifying the abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the abnormal feature data; outputting the abnormal score corresponding to each data point in the abnormal feature data according to the central membership of each data point in the abnormal feature data using a pre-built abnormal detection model; obtaining the abnormal data in the power consumption data according to the abnormal score corresponding to each data point in the abnormal feature data; wherein the abnormal detection model is based on the historical abnormality in the power system. The normal feature data is the training data, which is obtained by training the graph neural network; the present application can identify potential abnormal patterns and key features in the data by extracting abnormal features of the power data; by using the classification processing of the clustering algorithm, similar abnormal data can be classified into one category to reveal different types of abnormal patterns; finally, by using the anomaly detection model constructed based on the graph neural network, the complex relationships and dependencies between the nodes in the power data can be captured, and the understanding and recognition capabilities of abnormal data can be improved; therefore, the method of the present invention can quickly extract core features related to anomalies in large-scale electricity consumption data, and construct structured classifications of different abnormal patterns, thereby improving the efficiency and accuracy of anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 A schematic diagram of a flow chart of an abnormal data detection method based on data mining provided by the present invention;

[0078] Figure 2 A schematic diagram showing the comparison results of anomaly detection between an abnormal data detection method based on data mining provided by the present invention and an anomaly detection algorithm in the prior art;

[0079] Figure 3 A schematic diagram of the structure of an abnormal data detection system based on data mining provided by the present invention;

[0080] Figure 4 The present invention provides a schematic structural diagram of an electronic device. DETAILED DESCRIPTION

[0081] The present invention proposes an abnormal data detection method, system, device and medium based on data mining. The specific implementation manner of the present invention is further described in detail below with reference to the accompanying drawings.

[0082] Embodiment 1:

[0083] The present invention provides a method for detecting abnormal data based on data mining, the flow chart of which is as follows: Figure 1 As shown, including:

[0084] Step 1: extract abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data;

[0085] Step 2: Use clustering algorithm to classify the abnormal feature data and obtain the central membership of each data point in the abnormal feature data;

[0086] Step 3: According to the central membership of each data point in the abnormal feature data, use the pre-built anomaly detection model to output the anomaly score corresponding to each data point in the abnormal feature data;

[0087] Step 4: Obtain abnormal data in the electricity consumption data according to the abnormal scores corresponding to each data point in the abnormal feature data;

[0088] Among them, the anomaly detection model is obtained by training the Graph Neural Network (GNN) with the historical anomaly feature data in the power system as training data;

[0089] In modern power systems, with the continuous growth of electricity demand and the increasing complexity of power grid operation, it is particularly important to accurately detect and identify abnormal electricity consumption behaviors. Traditional anomaly detection methods mainly rely on manually set rules or are based on simple statistical models. These methods are often powerless when faced with large-scale, multi-dimensional, and dynamically changing electricity consumption data, and are prone to high false alarm rates or omissions of important anomalies. To this end, in recent years, data mining and machine learning technologies have gradually been introduced into the field of anomaly detection in power systems. Through feature extraction, clustering analysis, and deep learning modeling of massive electricity consumption data, abnormal features can be captured more accurately and potential risks can be predicted. In this context, this embodiment proposes an anomaly detection method based on a graph neural network, which combines a clustering algorithm and anomaly score evaluation to achieve efficient and accurate anomaly detection of electricity consumption data. Specifically:

[0090] In one implementation, the process of extracting abnormal features from the power consumption data acquired in the power system in step 1 to obtain abnormal feature data corresponding to the power consumption data may include:

[0091] Extracting features of the voltage deviation of the acquired power consumption data in the power system to obtain deviation feature data of the power consumption data;

[0092] Extracting the electricity consumption characteristics of the electricity consumption data to obtain electricity consumption characteristic data of the electricity consumption data;

[0093] Extracting features of load anomaly of power consumption data to obtain load feature data of the power consumption data;

[0094] Perform data integration based on the deviation characteristic data, power consumption characteristic data and load characteristic data to obtain abnormal characteristic data corresponding to the power consumption data;

[0095] In this implementation, the multi-dimensional features of the power system power consumption data are refined and effectively integrated, achieving a significant improvement in the technical path of anomaly detection. According to the characteristics of power consumption data and the needs of anomaly detection, key features such as voltage deviation, power consumption characteristics and load anomaly are extracted, capturing the three core dimensions of voltage stability, power consumption regularity and load change dynamics in the operation of the power system, and fully reflecting the pattern changes of power consumption data; in the feature extraction stage, not only the traditional single feature of voltage deviation is considered, but also multi-dimensional features such as power consumption, power consumption period, and power load are innovatively integrated. For example, the voltage deviation feature is used to reflect the voltage fluctuation. It is an important indicator to measure the difference between the actual voltage and the rated voltage, and plays a core role in the detection of abnormal power consumption data. The power consumption features include peak power consumption period, valley power consumption period and average power consumption, which are used to describe the regularity and stability of industrial and commercial power consumption behavior in different periods. The load abnormality feature is used to evaluate the load fluctuation and abnormality. It can identify abnormally high load rates (used to predict excessive power consumption or equipment failure) and capture abnormally low load rates (used to indicate low power efficiency or underutilization of equipment). Through the multi-dimensional fusion and deep mining of these features, this implementation method breaks through the limitations of single feature analysis, so that the extracted abnormal feature data has higher characterization ability and detection accuracy, and extracts key features from the preprocessed data, standardizes and normalizes them to improve the effect and accuracy of subsequent model training. This processing method not only improves the ability to identify complex abnormal power consumption behavior, but also enhances the robustness of feature data, and effectively copes with the noise and errors that may exist in large-scale, multi-source data.

[0096] In order to achieve efficient and accurate abnormal feature extraction, the above implementation method relies on high-quality electricity consumption data input. By comprehensively and meticulously acquiring and preprocessing the electricity consumption data in the power system, it is possible to ensure that the extracted data is consistent and complete, thereby providing a solid foundation for subsequent abnormal feature analysis. Based on this, before extracting abnormal features from electricity consumption data, it is first necessary to design a set of scientific data acquisition and processing methods to fully reflect the multi-dimensional information in the power system, eliminate the differences between different indicators, and improve the availability and analysis efficiency of electricity consumption data. The following will specifically describe the acquisition method and standardized processing process of electricity consumption data to lay a data foundation for the implementation of abnormal feature extraction, specifically:

[0097] In one implementation, the power consumption data in the above power system may include the following acquisition methods:

[0098] Extracting power consumption index data from a preset database of the power system;

[0099] Performing data preprocessing on the electricity consumption index data to obtain processed index data;

[0100] The scale space mapping technology is used to perform spatial mapping on the processing index data to obtain the processing index data in a unified scale space, and the processing index data in the unified scale space is obtained as the electricity consumption data in the power system.

[0101] For example, as shown in Table 1, the above-mentioned power consumption index data may include one or more of the following: user ID, date, time, total active power, total reactive power, maximum active power, minimum active power, power factor, harmonic distortion rate, abnormal flag, voltage data, current data, power load data, power consumption data and power consumption period data, etc.;

[0102] Table 1 Description of electricity consumption indicator dataset

[0103]

[0104] In this implementation, multi-dimensional power consumption index data, such as voltage, current, power factor, etc., are first extracted from the preset database of the power system, providing a comprehensive data basis for subsequent analysis. Compared with the traditional method that is limited to a single data source or simple preprocessing, this implementation expands the data acquisition channels and significantly improves the comprehensiveness and diversity of the data. In the data preprocessing stage, the scale space mapping technology and Lagrange interpolation method are creatively introduced to clean, deduplicate and fill the missing data of the extracted power consumption index data. The Lagrange interpolation method plays an important role in the data cleaning process, effectively improving the data integrity and ensuring the quality and consistency of the data. This stage not only removes the noise and outliers in the data through cleaning and sorting, but also solves the deviation and heterogeneity of the power consumption index data in different dimensions through scale space mapping. Scale space mapping maps the original power consumption index data to a unified scale space, eliminating the dimensional differences and heterogeneity between different indicators. At the same time, through decentralized processing, the characteristic mean of the data is close to zero, which improves the stability and comparability of the data. Through this process, the implementation method successfully simplifies the complex power consumption index data into a consistent data representation, significantly reduces the complexity of data processing, and provides a more accurate and reliable data foundation for subsequent feature extraction and model training. This standardized processing method enhances the efficiency of data analysis, and also provides high-quality input data for clustering algorithms and anomaly detection models, ensuring the generalization ability and robustness of the model in different scenarios. Therefore, this implementation method not only has significant advantages in data integrity and consistency, but also provides an efficient and intelligent solution for the integration and analysis of complex data by innovatively combining scale space mapping technology. By simplifying data representation and optimizing data structure, it lays a solid foundation for large-scale and dynamic power consumption data processing of power systems, and improves the adaptability, accuracy and efficiency of anomaly detection models. This technical path provides a new benchmark for the operation and management of intelligent power systems, and also promotes the development of multi-source heterogeneous data fusion and preprocessing technology, providing strong support for intelligent monitoring and anomaly detection of power data.

[0105] In order to further explore the potential patterns in electricity consumption data, after acquiring and standardizing the electricity consumption data in the power system, it is necessary to perform higher-level classification processing on the abnormal feature data. Through cluster analysis of abnormal feature data, data points with similar characteristics can be grouped to reveal the potential structure and abnormal patterns in the data. This classification processing can not only improve the efficiency of data analysis, but also provide a basis for subsequent central membership calculation to achieve more accurate anomaly detection. The following will specifically describe the application process of clustering algorithm in classification processing and the calculation method of central membership, specifically:

[0106] In one implementation, the process of classifying the abnormal feature data using a clustering algorithm in step 2 to obtain the central membership of each data point in the abnormal feature data may include:

[0107] The abnormal feature data are classified using a clustering algorithm to obtain the cluster center corresponding to each data point in the abnormal feature data;

[0108] According to the cluster center corresponding to each data point in the abnormal feature data, the central membership between each data point and its corresponding cluster center is calculated using the basic probability assignment method;

[0109] In this implementation, the abnormal feature data is classified and processed by clustering algorithms (for example, K-Means clustering algorithm, fuzzy C-means clustering algorithm, density-based clustering algorithm, hierarchical clustering algorithm, spectral clustering algorithm or Gaussian mixture model, etc.), and the central membership is calculated by combining the basic probability assignment method, so as to achieve accurate capture and effective expression of abnormal patterns in complex electricity consumption data. By selecting a suitable clustering algorithm for grouping analysis of abnormal feature data, data points with similar features can be classified, thereby revealing the inherent association and potential structure between data. At the same time, the classification results of data points are further refined by the basic probability assignment method, so that the model can capture potential abnormal features more sensitively. This process not only improves the efficiency of data organization, but also eliminates redundant information, making the feature data more representative and hierarchical. By calculating the central membership between each data point and the corresponding cluster center, this implementation further quantifies the similarity between the data point and its classification group. This quantification method based on the basic probability assignment method not only solves the ambiguity problem of traditional clustering methods in boundary point judgment, but also provides high-precision basic support for subsequent abnormal score evaluation. In addition, the introduction of central membership enables anomaly detection to capture subtle differences between data in a more detailed manner, thereby improving the robustness and adaptability of the detection model. Therefore, through the organic combination of cluster analysis and probability assignment, this implementation method realizes the multi-dimensional expression and dynamic adaptation of abnormal feature data, setting a new benchmark for the recognition and classification of abnormal patterns. This method not only breaks through the reliance on fixed rules or single models in traditional methods, but also significantly improves detection efficiency and accuracy, providing a cutting-edge solution for the intelligent identification of abnormal data in complex power systems. Through accurate feature classification and high-quality membership calculation, it provides strong data support for the training and optimization of subsequent anomaly detection models, greatly promoting the development of anomaly detection technology in intelligent power systems.

[0110] In order to further improve the accuracy of anomaly detection and the generalization ability of the model, after using the clustering algorithm to classify the abnormal feature data and calculate the central membership of the data points, these calculation results need to be used as a basis to support the training of the anomaly detection model. By associating the central membership in the historical abnormal feature data with the anomaly score, more abundant training data can be provided for the model to learn abnormal patterns. This processing method can not only effectively utilize the information in the classification process, but also lay a solid data foundation for the subsequent graph neural network model training. The following will introduce in detail the training process of the anomaly detection model and how to optimize the model performance based on the above classification results, specifically:

[0111] In one implementation, the anomaly detection model in step 3 above includes the following training process:

[0112] Obtain historical abnormal feature data;

[0113] Use clustering algorithm to perform cluster analysis on historical abnormal feature data to obtain the central membership of each data point in the historical abnormal feature data;

[0114] The central membership of each data point in the historical abnormal feature data is used as the input of training data;

[0115] The anomaly score of each data point in the historical anomaly feature data is used as the output of the training data;

[0116] Based on the input and output of the training data, the graph neural network is trained to obtain an anomaly detection model;

[0117] In this implementation, by introducing the combination of graph neural network and cluster analysis, an efficient and accurate anomaly detection model is innovatively constructed, which significantly improves the overall ability of anomaly detection of power consumption index data. The clustering algorithm is used to analyze the historical abnormal feature data, and the central membership of the data point is used as the model input, which effectively quantifies the correlation between the data point and its cluster center, providing a solid foundation for capturing complex data dependencies. In the process of model construction and training, the graph neural network algorithm is applied to the field of power consumption data anomaly detection for the first time. By fully considering the physical connection and logical relationship between power consumption equipment, an accurate graph structure is constructed, which can fully reflect the dynamic changes and complex associations between data points. The model takes the central membership of historical abnormal feature data as input and the anomaly score as output. Through cross-validation and parameter optimization, the detection accuracy and generalization ability of the model are continuously improved. After training, the anomaly detection model based on graph neural network can not only accurately identify known abnormal patterns, but also has adaptive learning capabilities, which can effectively deal with new or unknown anomalies and achieve efficient detection in dynamic environments. In practical applications, the model can monitor electricity consumption data in real time, calculate the anomaly score of each data point, and trigger alarms or exception processing based on set thresholds. By setting reasonable anomaly thresholds, the model achieves efficient classification and accurate evaluation of industrial and commercial electricity consumption data, ensuring the convenience and accuracy of the anomaly evaluation process. For example, the method of the present invention is compared with the method in the prior art (for example, represented as method 1 and method 2; method 1 can be an anomaly detection method based on statistics, and method 2 can be an anomaly detection method based on machine learning), and the comparison results are shown in Table 2:

[0118] Table 2 Anomaly detection of electricity consumption data

[0119]

[0120] It can be seen from the data in the table that the method of the present invention performs significantly better than the traditional statistical method (method 1) and machine learning method (method 2) in actual anomaly detection. For the actual abnormal data under each experimental number, the method of the present invention can always be completely consistent with the detection results of the actual abnormal data, showing extremely high detection accuracy and reliability. However, there are certain deviations in the detection results of methods 1 and 2. In particular, as the amount of abnormal data increases, the detection accuracy of the traditional method gradually decreases, indicating that its adaptability to large-scale complex data is limited. Compared with method 1, the method of the present invention overcomes the limitations of statistical methods that rely on fixed rules and thresholds, and can accurately capture complex nonlinear associations and abnormal patterns in data; compared with method 2, the method of the present invention fully combines technologies such as time series analysis and dynamic classification of abnormal scores, so that the learning ability and detection ability of the model are significantly enhanced, especially in multi-dimensional and complex scenarios. It shows stronger robustness and accuracy. This performance advantage not only proves the technical advancement of the method of the present invention, but also shows that it has a high application value in actual power systems, which can effectively improve the efficiency and accuracy of abnormal detection of power consumption data, and provide important guarantees for the stable operation of power systems.

[0121] In addition, in this implementation, the powerful feature learning ability of the graph neural network enables the model to effectively capture the multi-dimensional dependencies in the electricity consumption data, thereby showing excellent robustness and adaptability in complex power systems. Therefore, through the implementation of this implementation, not only the whole process of anomaly detection of electricity consumption data is optimized, but also the accuracy and efficiency of detection are fundamentally improved. By combining clustering algorithms and graph neural networks, data preprocessing, feature quantization and deep modeling are organically integrated, setting a new benchmark for anomaly detection technology. While providing a scientific basis for industrial and commercial electricity management, this implementation also provides innovative technical support for the intelligent development of the power industry, laying the foundation for building a more stable and efficient power system.

[0122] In this implementation, the process of constructing the above graph structure may specifically include:

[0123] Based on the information of physical connection and logical relationship between power consumption equipment, the graph structure required by the graph neural network is constructed. The nodes in the graph structure represent power consumption equipment or data points, and the edges represent the connection relationship between equipment or the correlation between data. In order to improve the generalization ability of the model, the constructed graph structure needs to be optimized: redundant edges and nodes are removed to improve the compactness and effectiveness of the graph structure; and a weight mechanism is introduced to assign weights to edges according to the connection strength between equipment and the correlation between data to improve the expression ability of the graph structure. In addition, when constructing anomaly detection model of power consumption index data based on graph neural network, the autonomous learning ability of graph neural network for the representation of graph structure data nodes is fully utilized to capture the complex dependencies in the data. Specifically, the model uses power consumption equipment and power grid nodes as constituent elements to construct a graph structure reflecting the correlation between these elements, collects historical power consumption data from power systems and smart meters, and uses the extracted abnormal features as nodes of the graph structure, and the topological structure of the power grid as the edges of the graph structure. Then, the model transmits the neighborhood information of the feature nodes through an effective activation function to obtain the weight matrix of the graph structure edge. Furthermore, we normalize the weights of all nodes and edges to obtain the attribute vector of the final layer graph node, which accurately describes the attributes of the abnormal characteristics of the power consumption index data. On this basis, combined with the time series information of power consumption, the graph neural network algorithm is used to solve the characteristic attribute vector, and finally a power consumption index data anomaly detection model is constructed, which provides strong support for the subsequent implementation of power consumption index data anomaly detection.

[0124] After completing the training of the anomaly detection model, it is necessary to use the anomaly score output by the model to further analyze and process the anomalies in the electricity consumption data. Through the quantile-based dynamic threshold method, the anomaly score of each data point can be classified, thereby more intuitively revealing the distribution of abnormal feature data. At the same time, in order to improve the comprehensiveness and accuracy of anomaly identification, it is also necessary to combine time series analysis and geospatial association analysis to deeply explore the spatiotemporal characteristics of abnormal data. The following will describe in detail how to locate and output abnormal information in electricity consumption data based on the classification results of anomaly scores, combined with time series and geospatial characteristics, specifically:

[0125] In one implementation, the process of obtaining abnormal data in the power consumption data according to the abnormal score corresponding to each data point in the abnormal feature data in the above step 4 may include:

[0126] Using the quantile-based dynamic threshold method, the anomaly score of each data point in the abnormal feature data is classified to generate a classification label corresponding to each data point;

[0127] According to the classification label corresponding to each data point, the time series anomaly detection algorithm is used to perform sequence analysis on the abnormal feature data to obtain the abnormal sequence structure in the abnormal feature data;

[0128] Using the association analysis algorithm based on geographic space, the abnormal sequence structure is spatially analyzed to output the abnormal data in the electricity consumption data;

[0129] In this implementation, by combining the quantile-based dynamic threshold method, the time series anomaly detection algorithm, and the geospatial association analysis algorithm, the efficient extraction and multi-dimensional analysis of abnormal information in electricity consumption data are achieved, which significantly improves the accuracy and comprehensiveness of anomaly detection. First, the dynamic threshold method based on quantiles innovatively adjusts the classification threshold dynamically according to the distribution characteristics of the anomaly score, avoiding the limitations of the traditional fixed threshold method in different scenarios. This method can flexibly adapt to different electricity consumption scenarios, ensure the accuracy and robustness of the classification results, and provide accurate classification labels for subsequent analysis. On the basis of the classification labels, the time series anomaly detection algorithm is used to perform sequence analysis on the abnormal feature data to capture the hidden time series patterns and dynamic change trends in the electricity consumption data. The algorithm can effectively identify complex patterns such as periodic anomalies and sudden anomalies in the data, thereby providing in-depth information for further anomaly evaluation. In addition, the introduction of the geospatial association analysis algorithm enables anomaly detection to break through the limitations of a single dimension and make full use of spatial characteristics to analyze the regional distribution and propagation laws of anomalies. Through the spatial correlation analysis of the abnormal sequence structure, not only can the specific location of the abnormality be located, but also the possible geographical correlation between abnormal events can be revealed, thereby achieving a comprehensive characterization of abnormal information in time and space. Therefore, through the overall design of this implementation method from dynamic threshold classification to spatiotemporal multidimensional analysis, it not only improves the detection ability of complex power consumption anomalies, but also provides a scientific basis for the operation monitoring and risk management of power systems. Its innovation lies in the organic combination of quantile dynamic thresholds, time series analysis and spatial correlation analysis to form a new anomaly detection method system, which sets a new technical benchmark for intelligent anomaly detection and in-depth analysis of power consumption data. The implementation of this method not only provides an efficient solution for the precise monitoring of complex power systems, but also provides important technical support for intelligent power management and decision support.

[0130] In order to further explore the potential rules of abnormal feature data, after classifying the abnormal score of each data point and generating classification labels, it is necessary to combine the time series analysis method to reorganize the data according to the time dimension to capture the changing trend of the abnormal feature data in time series. This process can not only reveal the dynamic characteristics of abnormal data, but also provide structured time series input data for subsequent geospatial association analysis, thereby realizing multi-dimensional analysis of abnormal information. The following will specifically explain how to analyze the reintegrated abnormal feature data through the time series anomaly detection algorithm, generate an abnormal sequence structure, and lay the foundation for the spatiotemporal positioning of anomalies. Specifically:

[0131] In this implementation, the process of performing sequence analysis on the abnormal feature data using a time series anomaly detection algorithm according to the classification label corresponding to each data point to obtain an abnormal sequence structure in the abnormal feature data may include:

[0132] According to the classification label corresponding to each data point, each data point in the abnormal feature data is reintegrated according to the time series to obtain the integrated abnormal feature data;

[0133] The time series anomaly detection algorithm is used to perform sequence analysis on the integrated abnormal feature data to obtain the abnormal sequence structure in the abnormal feature data;

[0134] In this implementation, the classification labels of abnormal feature data are reintegrated in time series, and combined with the in-depth analysis of the time series anomaly detection algorithm, the abnormal dynamic characteristics of power consumption data are innovatively captured and deeply understood. By integrating the abnormal feature data according to the time series, the scattered classification information is effectively connected, so that the data can reflect the continuity and change trend in the time dimension. This integration process not only eliminates the analysis limitations brought by isolated data points, but also lays the foundation for the logical consistency and structural integrity of the subsequent sequence analysis, so that anomaly detection has a stronger analysis depth and breadth. By analyzing the integrated data using the time series anomaly detection algorithm, the time series pattern of abnormal feature data can be deeply mined to capture complex time dynamic characteristics, such as periodic fluctuations, sudden anomalies, and persistent deviations. By fully considering the time dependence of the data, this implementation method not only realizes the detection of static anomalies, but also can identify potential abnormal behavior patterns related to time, providing scientific support for the dynamic monitoring of the power system. In addition, this modeling and analysis process of time series features can also provide predictive information, such as identifying impending abnormal trends, thereby providing a reliable basis for early warning and fault prevention and control of the system. In actual deployment, the trained graph neural network model is used to monitor electricity consumption data in real time, and the monitored indicators are processed and analyzed by calculating the anomaly scores, such as Figure 2As shown, the method of the present invention is compared with the above-mentioned method 1 and method 2 by the Matthews Correlation Coefficient (MCC). It can be seen from the figure that with the increase in the number of input data samples, the method of the present invention always performs better than method 1 and method 2 in the Matthews Correlation Coefficient (MCC), and can stably maintain a high correlation coefficient value, reflecting the superiority and robustness of the method of the present invention under different data scales. Compared with method 1 and method 2, the method of the present invention can still show a higher MCC value when the data samples are small, which shows that it has a strong anomaly detection ability and generalization performance under small sample conditions. With the increase in the number of data samples, although the MCC values ​​of method 1 and method 2 are improved, the overall fluctuation is large and lower than that of the method of the present invention, indicating that the traditional method is prone to a decrease in detection performance due to insufficient model adaptability when facing large-scale data. Through the above comparative analysis, it can be seen that the performance stability and accuracy of the method of the present invention under different data scales are significantly better than those of the traditional method, which not only reflects its technical advantages in anomaly detection capabilities, but also lays a solid foundation for intelligent management in practical applications. In actual power systems, the ultimate goal of anomaly detection is to quickly respond to and accurately handle potential problems. Therefore, combining the anomaly detection results with real-time alarm and fault handling mechanisms can further leverage the advantages of the method of the present invention and ensure the efficiency of the entire process from data analysis to problem solving.

[0135] According to the preset anomaly score threshold, when the data exceeds the threshold, an alarm is triggered and detailed anomaly information is provided, including anomaly type, occurrence time, and impact range, etc., to provide efficient decision-making support for operation and maintenance personnel. This overall solution forms a complete technical chain from data monitoring to fault handling through closed-loop feedback of dynamic monitoring, anomaly identification, and real-time alarms. It not only improves the depth and accuracy of anomaly detection, but also provides a new idea for the stable operation and intelligent management of power systems. This innovative method that combines time series analysis with real-time anomaly detection has set a technical benchmark for dynamic monitoring and efficient management in the power industry, and ensures the safety and reliability of system operation.

[0136] In summary, the present invention aims at the problems of low recognition efficiency, incomplete coverage and high false alarm rate in the existing manual review or simple rule judgment method for identifying abnormal data due to the diversification and complexity of power consumption data in the power system. A more efficient, accurate and comprehensive abnormal data detection method is proposed. By mapping the collected power consumption data to a new scale space and performing numerical decentralized processing, the method ensures the consistency of the data sample dimension and provides a unified standard for subsequent analysis. On this basis, the key features that can accurately characterize the abnormality of power consumption data, such as voltage deviation, power consumption characteristics and load rate abnormality, are extracted, and an efficient abnormality detection model is established in combination with the graph neural network algorithm, realizing the accurate detection of abnormal power consumption data. The method comprehensively uses advanced automation technologies such as data mining, machine learning and neural networks, which not only improves the monitoring efficiency and accuracy of the quality of power consumption data in the power system, but also enhances the quality monitoring and analysis capabilities of indicator report data, and strongly supports the scientificity and accuracy of business decision-making. In key links such as data collection and preprocessing, feature engineering, graph structure construction, model training and verification, and abnormality detection and feedback, the method of the present invention shows significant improvement effects. Through in-depth mining and modeling of power system power consumption data, comprehensive technical support is provided for the construction of an intelligent power monitoring system, and through comparative experiments and practical application verification, the accuracy of power data anomaly detection is significantly improved. By effectively identifying abnormal events that are difficult to detect with traditional methods, it is beneficial to improve the monitoring capability and operational reliability of the power system, and provide strong support for the intelligent development of the power industry and the construction of smart cities. In addition, the method of the present invention performs well in real-time and robustness, and can efficiently process and analyze large-scale power consumption data in a short time, adapt to power systems of different sizes and power consumption data of varying complexity, and show good adaptability and scalability. By introducing automated and intelligent anomaly detection and feedback mechanisms, the work efficiency and response speed of operation and maintenance personnel are greatly improved, and the operation and maintenance costs and failure risks are reduced, thereby providing reliable guarantees for abnormal monitoring of data quality. In general, the method of the present invention plays an important role in improving power data quality monitoring, optimizing business decisions, and promoting the development of intelligent power systems.

[0137] Embodiment 2:

[0138] The present invention based on the same inventive concept also provides an abnormal data detection system based on data mining, the structural composition diagram is as follows: Figure 3 As shown, including:

[0139] A feature extraction module is used to extract abnormal features from the power consumption data acquired in the power system to obtain abnormal feature data corresponding to the power consumption data;

[0140] The clustering analysis module is used to classify the abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the abnormal feature data;

[0141] The anomaly detection module is used to output the anomaly score corresponding to each data point in the abnormal feature data according to the central membership of each data point in the abnormal feature data and using a pre-built anomaly detection model;

[0142] An abnormality determination module, used to obtain abnormal data in the power consumption data according to the abnormality score corresponding to each data point in the abnormal feature data;

[0143] Among them, the anomaly detection model is obtained by training the graph neural network with historical abnormal feature data in the power system as training data.

[0144] In one implementation, the feature extraction module may include:

[0145] The deviation feature extraction submodule is used to extract the features of the voltage deviation of the power consumption data acquired in the power system to obtain the deviation feature data of the power consumption data;

[0146] The power consumption feature extraction submodule is used to extract the power consumption features of the power consumption data to obtain the power consumption feature data of the power consumption data;

[0147] The load anomaly feature extraction submodule is used to extract features of the load anomaly of the power consumption data to obtain load feature data of the power consumption data;

[0148] The feature integration submodule is used to integrate the deviation feature data, the power consumption feature data and the load feature data to obtain the abnormal feature data corresponding to the power consumption data.

[0149] In one implementation, the cluster analysis module may include:

[0150] The classification processing submodule is used to classify the abnormal feature data using a clustering algorithm to obtain the cluster center corresponding to each data point in the abnormal feature data;

[0151] The membership calculation submodule is used to calculate the central membership between each data point and its corresponding cluster center using the basic probability assignment method according to the cluster center corresponding to each data point in the abnormal feature data.

[0152] In one implementation, the above detection system may further include: a model building module, specifically used for:

[0153] Obtain historical abnormal feature data;

[0154] Use clustering algorithm to perform cluster analysis on historical abnormal feature data to obtain the central membership of each data point in the historical abnormal feature data;

[0155] The central membership of each data point in the historical abnormal feature data is used as the input of training data;

[0156] The anomaly score of each data point in the historical anomaly feature data is used as the output of the training data;

[0157] Based on the input and output of the training data, the graph neural network is trained to obtain the anomaly detection model.

[0158] In one implementation, the above detection system may further include: a data acquisition module, specifically used for:

[0159] Extracting power consumption index data from a preset database of the power system;

[0160] Performing data preprocessing on the electricity consumption index data to obtain processed index data;

[0161] The scale space mapping technology is used to perform spatial mapping on the processing index data to obtain the processing index data in a unified scale space, and the processing index data in the unified scale space is obtained as the electricity consumption data in the power system.

[0162] For example, the above-mentioned power consumption index data may include one or more of the following: voltage data, current data, power factor data, power load data, power consumption data and power consumption period data.

[0163] In one implementation, the above-mentioned abnormality determination module may include:

[0164] The anomaly classification submodule is used to classify the anomaly score of each data point in the abnormal feature data using a dynamic threshold method based on quantiles, and generate a classification label corresponding to each data point;

[0165] The time anomaly detection submodule is used to perform sequence analysis on the abnormal feature data using the time series anomaly detection algorithm according to the classification label corresponding to each data point, and obtain the abnormal sequence structure in the abnormal feature data;

[0166] The spatial correlation analysis submodule is used to perform spatial correlation analysis on abnormal sequence structures using a correlation analysis algorithm based on geographic space, and output abnormal data in the electricity consumption data.

[0167] In this implementation, the above-mentioned time anomaly detection submodule may include:

[0168] A time series integration unit is used to reintegrate each data point in the abnormal feature data according to the time series according to the classification label corresponding to each data point, so as to obtain integrated abnormal feature data;

[0169] The time series analysis unit is used to perform sequence analysis on the integrated abnormal feature data using a time series anomaly detection algorithm to obtain an abnormal sequence structure in the abnormal feature data.

[0170] Embodiment 3:

[0171] like Figure 4 As shown, the present invention also provides an electronic device, which may be a computer device, a single-chip device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected via a bus; the memory may be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory may also be used to store data, which may be called and / or modified when the instructions are executed.

[0172] 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 gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in a storage medium to implement a corresponding method flow or a corresponding function, so as to implement the steps of an abnormal data detection method based on data mining in the above-mentioned embodiment.

[0173] Embodiment 4:

[0174] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory), which is a memory device in an electronic device for storing programs and data. It can be understood that the storage medium here can include both built-in storage media in the electronic device and, of course, extended storage media supported by the electronic device. The storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage. The processor loads and executes one or more instructions stored in the storage medium, which can implement the steps of an abnormal data detection method based on data mining in the above embodiment.

[0175] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0176] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0177] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0178] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0179] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A method for detecting abnormal data based on data mining, characterized in that: include: Extracting abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data; Using a clustering algorithm to classify the abnormal feature data to obtain the central membership of each data point in the abnormal feature data; Outputting the anomaly score corresponding to each data point in the abnormal feature data using a pre-built anomaly detection model according to the central membership of each data point in the abnormal feature data; Obtaining abnormal data in the power consumption data according to an abnormal score corresponding to each data point in the abnormal feature data; Among them, the anomaly detection model is obtained by training the graph neural network using the historical anomaly feature data in the power system as training data.

2. The method according to claim 1, characterized in that The extracting abnormal features of the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data includes: Extracting features of the voltage deviation of the acquired power consumption data in the power system to obtain deviation feature data of the power consumption data; Extracting the power consumption characteristics of the power consumption data to obtain power consumption characteristic data of the power consumption data; Extracting features of the load anomaly of the power consumption data to obtain load feature data of the power consumption data; Data integration is performed based on the deviation characteristic data, the power consumption characteristic data and the load characteristic data to obtain abnormal characteristic data corresponding to the power consumption data.

3. The method according to claim 1, characterized in that The method of using a clustering algorithm to classify the abnormal feature data to obtain the central membership of each data point in the abnormal feature data includes: Using a clustering algorithm to classify the abnormal feature data, and obtain a cluster center corresponding to each data point in the abnormal feature data; According to the cluster center corresponding to each data point in the abnormal feature data, the central membership between each data point and its corresponding cluster center is calculated using the basic probability assignment method.

4. The method according to claim 1, characterized in that The anomaly detection model includes the following training process: Obtain historical abnormal feature data; Performing cluster analysis on the historical abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the historical abnormal feature data; Using the central membership of each data point in the historical abnormal feature data as input of training data; Using the anomaly score of each data point in the historical anomaly feature data as output of training data; Based on the input and output of the training data, the graph neural network is trained to obtain an anomaly detection model.

5. The method according to claim 1, characterized in that The power consumption data in the power system includes the following acquisition methods: Extracting power consumption index data from a preset database of the power system; Performing data preprocessing on the electricity consumption index data to obtain processed index data; The scale space mapping technology is used to perform spatial mapping on the processing index data to obtain the processing index data in a unified scale space, and the processing index data in the unified scale space is obtained as the power consumption data in the power system.

6. The method according to claim 5, characterized in that The power consumption index data includes one or more of the following: voltage data, current data, power factor data, power load data, power consumption data and power consumption period data.

7. The method according to claim 1, characterized in that The obtaining the abnormal data in the power consumption data according to the abnormal score corresponding to each data point in the abnormal feature data includes: Using a quantile-based dynamic threshold method, the abnormal score of each data point in the abnormal feature data is classified to generate a classification label corresponding to each data point; According to the classification label corresponding to each data point, a time series anomaly detection algorithm is used to perform sequence analysis on the abnormal feature data to obtain an abnormal sequence structure in the abnormal feature data; A spatial correlation analysis algorithm based on geographic space is used to perform spatial correlation analysis on the abnormal sequence structure, and the abnormal data in the electricity consumption data is output.

8. The method according to claim 7, characterized in that The method of performing sequence analysis on the abnormal feature data using a time series anomaly detection algorithm according to the classification label corresponding to each data point to obtain an abnormal sequence structure in the abnormal feature data includes: According to the classification label corresponding to each data point, each data point in the abnormal feature data is reintegrated in time series to obtain integrated abnormal feature data; A time series anomaly detection algorithm is used to perform sequence analysis on the integrated abnormal feature data to obtain an abnormal sequence structure in the abnormal feature data.

9. An abnormal data detection system based on data mining, characterized in that: include: A feature extraction module is used to extract abnormal features from the acquired power consumption data in the power system to obtain abnormal feature data corresponding to the power consumption data; A clustering analysis module, used to classify the abnormal feature data using a clustering algorithm to obtain the central membership of each data point in the abnormal feature data; An anomaly detection module, used to output an anomaly score corresponding to each data point in the abnormal feature data according to the central membership of each data point in the abnormal feature data and using a pre-built anomaly detection model; an abnormality determination module, used to obtain abnormal data in the power consumption data according to the abnormality score corresponding to each data point in the abnormal characteristic data; Among them, the anomaly detection model is obtained by training the graph neural network using the historical anomaly feature data in the power system as training data.

10. The system according to claim 9, characterized in that The feature extraction module comprises: The deviation feature extraction submodule is used to extract the features of the voltage deviation of the power consumption data acquired in the power system to obtain the deviation feature data of the power consumption data; An electricity usage feature extraction submodule, used to extract the electricity usage features of the electricity usage data to obtain electricity usage feature data of the electricity usage data; A load anomaly feature extraction submodule is used to extract features of the load anomaly of the power consumption data to obtain load feature data of the power consumption data; The feature integration submodule is used to integrate the deviation feature data, the power consumption feature data and the load feature data to obtain abnormal feature data corresponding to the power consumption data.

Images