Database anomaly detection method for power plant statistical management system

Abstract

The invention relates to the technical field of database anomaly detection, and provides a database anomaly detection method for a power plant statistical management system, which can perform collaborative anomaly identification and analysis on power plant statistical data. The method comprises the steps that effective clustering of statistical data of a power plant is achieved in combination with a language embedding-VAE clustering model, so that high-quality data input is provided for a subsequent model, the analysis capacity of a data fluctuation source is improved through a multi-decomposition-LSTM model, and accurate recognition and explanation of key abnormal points are achieved; according to the invention, efficient analysis and early warning of power plant statistics can be realized, and important data support and early warning capability are provided for intelligent operation and maintenance and decision support of a power plant.

Description

Technical Field

[0001] The present invention relates to the technical field of database anomaly detection, in particular to a database anomaly detection method for a power plant statistical management system. Background Art

[0002] With the continuous improvement of the informatization level of power plants, statistical management systems have been widely used in various aspects of power plant production operations, equipment management, and business decision-making. The database in the statistical management system carries a large amount of data, including power generation plans, equipment operating parameters, fuel quantity and quality data, major and minor production indicators, etc. The integrity and accuracy of this data are directly related to the operating efficiency and safety of the power plant. However, in actual operation, due to factors such as system complexity, data source diversity, and high data update frequency, various anomalies often exist in the system database. If these anomalies are not discovered and handled in a timely manner, they will have a negative impact on decision support and safety management. However, existing database anomaly detection methods still have many shortcomings: First, existing methods have difficulty capturing such multi-scale and nonlinear time series characteristics, resulting in insensitivity to potential equipment anomalies such as frequent starts and stops and abnormal shutdowns. Second, existing methods are based on traditional detection methods such as single-variable analysis or simple indicator monitoring, which makes it difficult to timely reveal hidden systematic deviations or plan execution anomalies, reducing the intelligent response capability of scheduling. Summary of the Invention

[0003] The present invention provides a database anomaly detection method for a power plant statistical management system. The method can simultaneously perform in-depth anomaly analysis on power plant statistical data, improving the system's ability to automatically perceive, locate, and interpret potential risks. The method combines a semantic embedding-VAE clustering model with a multiple decomposition-LSTM model, possessing strong temporal feature learning capabilities and semantic anomaly recognition capabilities, making it suitable for intelligent anomaly detection scenarios in the complex business data environment of power plants. On the one hand, the present invention introduces a clustering strategy that combines semantic embedding with an anomaly-aware VAE model. By compressing and expressing high-dimensional embedding vectors and combining them with cluster boundary modeling, it effectively clusters power plant statistical data, thereby providing high-quality data input for subsequent models. On the other hand, the present invention constructs an LSTM model that integrates a multiple decomposition method with a masked prediction mechanism to decompose raw data into multiple conceptual terms, such as trends, cycles, and residuals. The Shapley value analysis mechanism driven by masked samples and residuals is introduced, effectively improving the ability to analyze data fluctuation sources and accurately identifying and interpreting key anomalies. Through the above-mentioned collaborative modeling and analysis, the present invention can efficiently output power plant anomaly detection results, providing important data support and early warning capabilities for intelligent operation and maintenance and decision support of power plants.

[0004] The present invention provides a database anomaly detection method for a power plant statistical management system, the method comprising the following steps:

[0005] Step S1: Collecting power plant statistical data from the power plant statistical management system database; the power plant statistical data includes power generation plans, equipment operating parameters, fuel quantity and quality data, and major and minor production indicators;

[0006] Step S2: Perform data cleaning, standardization, and feature engineering on the power plant statistical data to generate preprocessed statistical data;

[0007] Feature engineering processing includes extracting moving average features, differential processing features and time lag features from power plant statistical data;

[0008] Step S3: constructing a language embedding-VAE clustering model, clustering the pre-processed statistical data using the language embedding-VAE clustering model to obtain statistical clustering feature data;

[0009] Step S4: Establish an LSTM model, adopt the time series data multiple decomposition method and the fusion masking prediction mechanism to optimize the LSTM model, thereby constructing a multiple decomposition-LSTM model, and use the multiple decomposition-LSTM model to process the preprocessed statistical data and statistical clustering feature data to obtain the power plant anomaly detection results;

[0010] Step S5: Generate an abnormality report and early warning notification based on the abnormality detection results of the power plant to assist decision-making.

[0011] Furthermore, in step S3, the process of obtaining statistical clustering feature data specifically includes the following steps:

[0012] Step S31: performing word segmentation, stop word removal, and stem extraction on the pre-processed statistical data to generate a standardized text sequence;

[0013] Step S32: Use word embedding technology to convert the standardized text sequence into a dense vector to generate a high-dimensional embedding vector;

[0014] Step S33: Constructing an anomaly-aware VAE model, using a two-stage strategy to compress the high-dimensional embedding vector into a low-dimensional space to generate a low-dimensional text vector; the two-stage strategy includes unsupervised pre-training and anomaly-aware fine-tuning; in the anomaly-aware fine-tuning stage, a small number of labeled anomaly samples are introduced, and an additional penalty mechanism is introduced for the anomaly samples, which causes the anomaly-aware VAE model to perform poorly when reconstructing anomaly data, thereby enhancing the separability of anomaly and normal samples;

[0015] Step S34: Construct an enhanced HDBSCAN clustering model. The enhanced HDBSCAN clustering model includes an online stage and an offline stage. The online stage uses Bubble-tree to perform online data summary on low-dimensional text vectors, dynamically maintains the compressed representation of text semantics, and obtains Bubble-tree low-dimensional embedding vectors; the offline stage reconstructs the main cluster structure of the Bubble-tree low-dimensional embedding vectors, learns the semantic distribution boundaries, and generates statistical clustering feature data; Bubble-tree is an online data summary structure for dynamic data, which uses a tree-like hierarchical organization to cluster and compress low-dimensional text vectors; each bubble node in the Bubble-tree represents a local semantic cluster, which contains a center point, density information and embedding range, and finally forms a text semantic compression tree, in which each leaf node represents a Bubble-tree low-dimensional embedding vector.

[0016] Furthermore, in step S4, the process of obtaining the abnormality detection result of the power plant specifically includes the following steps:

[0017] Step S41: inputting the preprocessed statistical data and the statistical clustering feature data into the multi-decomposition-LSTM model for training to obtain a trained LSTM model;

[0018] Step S42: Decomposing the pre-processed structured data into multiple concept items, namely, trend concept items, cycle concept items, and residual concept items. When decomposing the cycle concept item, it is further refined into two sub-concepts, namely, seasonal fluctuation term and short-term cyclical fluctuation term, to construct a more refined decomposition concept set, thereby obtaining a decomposed concept set.

[0019] Step S43: performing a masking operation on each concept in the decomposed concept set to generate masked samples; calculating the prediction result after each masking based on the masked samples and the trained LSTM model to obtain a masked prediction result;

[0020] Step S44: Introduce a residual sensitivity factor to dynamically adjust the weights of multiple concept items and construct a residual-driven-Shapley enhancement mechanism; based on the masking prediction results, use the residual-driven-Shapley enhancement mechanism to evaluate the influence of multiple concept items on the prediction results of the trained LSTM model to obtain SHAP contribution data;

[0021] Step S45: Based on the trained LSTM model, determine whether the current time point is an anomaly point. For the detected anomaly point, combine the SHAP contribution data to determine the root cause of the anomaly point and generate the power plant anomaly detection result.

[0022] By adopting the above scheme, the beneficial effects achieved by the present invention are as follows:

[0023] This paper introduces a semantic embedding and anomaly-aware VAE clustering model. By converting statistical data such as scheduling records and operation reports into high-dimensional semantic vectors and combining them with a small number of labeled anomaly samples for semantic reconstruction and cluster identification under an unsupervised learning framework, the system can effectively identify abnormal expressions, potential risk signals, and operational behaviors that deviate from routine in the text, thereby significantly improving the perception accuracy and input quality of the subsequent anomaly detection model for statistical data.

[0024] By constructing an LSTM model that integrates multiple decomposition methods and a masked prediction mechanism, this paper achieves in-depth modeling of structured time series data from power plants and fine-grained detection of abnormal behaviors. This model not only captures both medium- and long-term trends and short-term fluctuations in the data, but also improves the interpretability of anomaly detection through a residual-driven Shapley contribution mechanism. Compared with traditional methods that rely solely on error judgment, this paper can clearly identify the dominant factors behind anomalies, achieve intelligent location of the root causes of abnormal data, improve the accuracy of anomaly detection and the reliability of decision-making, and solve the problem that existing statistical systems lack the ability to deeply analyze complex dynamic data.

[0025] Through the above-mentioned collaborative analysis mechanism, the present invention effectively breaks through data silos and significantly enhances the power plant statistical management system's ability to integrate and process multi-source information; the organic combination of the language embedding-VAE clustering model and the multiple decomposition-LSTM model in the present invention not only realizes the comprehensive coverage of intelligent detection of various data types in the database, but also improves the response speed and visual interpretation ability of data warnings, enhances the system's auxiliary decision-making level in key business links such as operation and maintenance scheduling, anomaly investigation, and risk prevention, and provides key technical support for building a highly reliable and intelligent power information management platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 The present invention provides a flow chart of a method for detecting anomalies in a database of a power plant statistical management system;

[0027] Figure 2 This is a diagram showing the influence of multiple concept terms on the prediction results of the trained LSTM model in Example 5. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0029] Example 1, according to Figure 1The present invention provides a database anomaly detection method for a power plant statistical management system, the method comprising the following steps:

[0030] Step S1: Data collection: Collecting power plant statistical data from the power plant statistical management system database; the power plant statistical data includes power generation plans, equipment operating parameters, fuel quantity and quality data, and major and minor production indicators;

[0031] Step S2: Data preprocessing: Perform data cleaning, standardization, and feature engineering on the power plant statistical data to generate preprocessed statistical data;

[0032] Feature engineering processing includes extracting moving average features, differential processing features and time lag features from power plant statistical data;

[0033] Step S3: Cluster feature extraction: construct a language embedding-VAE clustering model, cluster the pre-processed statistical data through the language embedding-VAE clustering model to obtain statistical cluster feature data;

[0034] Step S4: Anomaly detection: Establish an LSTM model, use the time series data multiple decomposition method and the fusion masking prediction mechanism to optimize the LSTM model, thereby constructing a multiple decomposition-LSTM model. The preprocessed statistical data and statistical clustering feature data are processed by the multiple decomposition-LSTM model to obtain the power plant anomaly detection results;

[0035] Step S5: Early warning decision-making: Generate an abnormality report and early warning notification based on the abnormality detection results of the power plant to assist in decision-making.

[0036] Example 2: This example is based on Example 1. In this example, in step S3, the process of obtaining statistical clustering feature data specifically includes the following steps:

[0037] Step S31: performing word segmentation, stop word removal, and stem extraction on the pre-processed statistical data to generate a standardized text sequence;

[0038] Step S32: Use word embedding technology to convert the standardized text sequence into a dense vector to generate a high-dimensional embedding vector;

[0039] Step S33: Construct an anomaly-aware VAE model and use a two-stage strategy to compress the high-dimensional embedding vector into a low-dimensional space to generate a low-dimensional text vector. The two-stage strategy includes unsupervised pre-training and anomaly-aware fine-tuning. In the anomaly-aware fine-tuning stage, a small number of labeled anomaly samples are introduced, and an additional penalty mechanism is introduced for the anomaly samples, which makes the anomaly-aware VAE model perform poorly when reconstructing anomaly data, thereby enhancing the separability of anomaly and normal samples. The formula used is as follows:

[0040] Joint loss function in the anomaly perception fine-tuning stage:

[0041] ;

[0042] in, represents the total loss function, represents the VAE loss of normal samples, represents the abnormal perception penalty coefficient, represents the input data of abnormal samples, represents the reconstruction output of the abnormal sample; represents the reconstruction error of abnormal samples, Represents the expected value of abnormal samples;

[0043] Step S34: Construct an enhanced HDBSCAN clustering. The enhanced HDBSCAN clustering includes an online stage and an offline stage. The online stage uses Bubble-tree to perform online data summary on low-dimensional text vectors, dynamically maintains the compressed representation of text semantics, and obtains Bubble-tree low-dimensional embedding vectors; the offline stage reconstructs the main cluster structure of the Bubble-tree low-dimensional embedding vectors, learns the semantic distribution boundaries, and generates statistical clustering feature data; Bubble-tree is an online data summary structure for dynamic data, which uses a tree-like hierarchical organization to cluster and compress low-dimensional text vectors; each bubble node in the Bubble-tree represents a local semantic cluster, which contains a center point, density information and embedding range, and finally forms a text semantic compression tree, in which each leaf node represents a Bubble-tree low-dimensional embedding vector.

[0044] Example 3: This example is based on Example 1. In this example, in step S3, the process of obtaining statistical clustering feature data specifically includes the following steps:

[0045] Step E1: Perform word segmentation, stop word removal, and stem extraction on the preprocessed statistical data to generate a standardized text sequence;

[0046] Step E2: Use word embedding technology to convert the normalized text sequence into a dense vector to generate a high-dimensional embedding vector;

[0047] Step E3: Build an anomaly-aware VAE model. This model uses a two-stage strategy of unsupervised pre-training and anomaly-aware fine-tuning to compress the high-dimensional embedding vector into a low-dimensional space, generating a low-dimensional text vector. During the anomaly-aware fine-tuning stage, a small number of labeled anomaly samples are introduced, and an additional penalty mechanism is introduced for these anomaly samples. This causes the anomaly-aware VAE model to perform poorly when reconstructing anomaly data, thereby enhancing the separability of anomaly and normal samples.

[0048] Step E4: Use HDBSCAN clustering to process the low-dimensional text vectors and generate statistical clustering feature data.

[0049] Embodiment 4: This embodiment is based on embodiment 1. In this embodiment, in step S3, the process of obtaining statistical clustering feature data specifically includes the following steps:

[0050] Step Q1: Perform word segmentation, stop word removal, and stem extraction on the preprocessed statistical data to generate a standardized text sequence;

[0051] Step Q2: Use word embedding technology to convert the standardized text sequence into a dense vector to generate a high-dimensional embedding vector;

[0052] Step Q3: Use the VAE model to compress the high-dimensional embedding vector into a low-dimensional space to generate a low-dimensional text vector;

[0053] Step Q4: Use HDBSCAN clustering to process low-dimensional text vectors and generate statistical clustering feature data.

[0054] Example 5, according to Figure 2 This embodiment is based on the second embodiment. In this embodiment, in step S4, the process of obtaining the abnormality detection result of the power plant specifically includes the following steps:

[0055] Step S41: inputting the preprocessed statistical data and the statistical clustering feature data into the multi-decomposition-LSTM model for training to obtain a trained LSTM model;

[0056] Step S42: Decomposing the pre-processed structured data into multiple concept items, namely, trend concept items, cycle concept items, and residual concept items. When decomposing the cycle concept item, it is further refined into two sub-concepts, namely, seasonal fluctuation term and short-term cyclical fluctuation term, to construct a more refined decomposition concept set, thereby obtaining a decomposed concept set.

[0057] Step S43: performing a masking operation on each concept in the decomposed concept set to generate masked samples; calculating the prediction result after each masking based on the masked samples and the trained LSTM model to obtain a masked prediction result;

[0058] Step S44: Introduce the residual sensitivity factor to dynamically adjust the weights of multiple concept items and construct a residual-driven-Shapley enhancement mechanism. Based on the masking prediction results, the residual-driven-Shapley enhancement mechanism is used to evaluate the influence of multiple concept items on the prediction results of the trained LSTM model to obtain SHAP contribution data. The formula used is as follows:

[0059] ;

[0060] in, represents the time step, represents the concept item index, Indicates the conceptual items, including trend conceptual item, residual conceptual item, seasonal fluctuation item and short-term cyclical item; represents the residual sensitivity factor, At the time point The actual observed value of represents the concept set after decomposition, Indicates removal of The combination of concept items, namely, masking , Indicates a time point The input sample, Indicates blocking right Make predictions using the trained LSTM model;

[0061] ;

[0062] in, represents SHAP contribution data, Representation subset In the cover On the premise of A subset selected from Representation subset The number of elements, express The factorial of Represents the total number of elements in the decomposed concept set, express The factorial of represents the union operation, represents marginal contribution;

[0063] Step S45: Based on the trained LSTM model, determine whether the current time point is an anomaly point. For the detected anomaly point, combine the SHAP contribution data to determine the root cause of the anomaly point and generate the power plant anomaly detection result.

[0064] Example 6. This example is based on Example 5. In this example, step S5: early warning decision-making: generating an abnormality report and early warning notification based on the abnormality detection results of the power plant to assist in decision-making;

[0065] Power plant anomaly detection results:

[0066] Abnormal time: 2025-04-28 14:00;

[0067] Abnormal indicators: Boiler coal consumption rate (actual value 356g / kWh, predicted value 312g / kWh, deviation +14%);

[0068] [Root causes of abnormality]: short-term cyclical fluctuations (contributing 48%), residual noise (contributing 30%);

[0069] [Cluster Analysis]: The current data is an isolated point in the bubble-tree clustering, and there is an atypical operating mode;

[0070]

Suggestion

[0071] The present invention and its embodiments are described above. Such description is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in this field are inspired by it and do not depart from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical solution, which should all fall within the scope of protection of the present invention.

Claims

1. A database anomaly detection method for a power plant statistical management system, characterized by: The method comprises the following steps: Step S1: Collecting power plant statistical data from the power plant statistical management system database; Step S2: Perform feature engineering on the statistical data of the power plant to generate preprocessed statistical data; Step S3: Constructing a language embedding-VAE clustering model, clustering the preprocessed statistical data using the language embedding-VAE clustering model to obtain statistical clustering feature data; Step S4: Construct a multiple decomposition-LSTM model, and process the preprocessed structured data and statistical clustering feature data through the multiple decomposition-LSTM model to obtain the power plant anomaly detection results.

2. The database anomaly detection method for a power plant statistical management system according to claim 1, characterized in that: Feature engineering processing includes extracting moving average features, differential processing features and time lag features from power plant statistical data.

3. The database anomaly detection method for a power plant statistical management system according to claim 1, characterized in that: In step S4, constructing the multiple decomposition-LSTM model specifically includes: establishing an LSTM model, adopting a multiple decomposition method of time series data and a fusion masking prediction mechanism, optimizing the LSTM model, and thus constructing the multiple decomposition-LSTM model.

4. The database anomaly detection method for a power plant statistical management system according to claim 1, characterized in that: In step S3, the process of obtaining statistical clustering feature data specifically includes the following steps: Step S31: Processing the statistical clustering feature data to generate a standardized text sequence; Step S32: Use word embedding technology to convert the standardized text sequence into a dense vector to generate a high-dimensional embedding vector; Step S33: Build an anomaly perception-VAE model, and use a two-stage strategy to compress the high-dimensional embedding vector into a low-dimensional space to generate a low-dimensional text vector; Step S34: Construct an enhanced HDBSCAN clustering. In the online stage, Bubble-tree is used to process the low-dimensional text vector to obtain the Bubble-tree low-dimensional embedding vector; in the offline stage, the main cluster structure of the Bubble-tree low-dimensional embedding vector is reconstructed to generate statistical clustering feature data.

5. The database anomaly detection method for a power plant statistical management system according to claim 4, characterized in that: The two-stage strategy includes unsupervised pre-training and anomaly-aware fine-tuning.

6. The database anomaly detection method for a power plant statistical management system according to claim 3, characterized in that: In step S4, the process of obtaining the abnormality detection result of the power plant specifically includes the following steps: Step S41: inputting the preprocessed statistical data and the statistical clustering feature data into the multi-decomposition-LSTM model for training to obtain a trained LSTM model; Step S42: Decomposing the pre-processed statistical data into multiple concept items to obtain a decomposed concept set; Step S43: performing a masking operation on the decomposed concept set and combining it with the trained LSTM model to generate a masking prediction result; Step S44: constructing a residual-driven-Shapley enhancement mechanism, combining the occlusion prediction results, and obtaining SHAP contribution data; Step S45: Generate time series anomaly detection results based on the trained LSTM model and SHAP contribution data.

7. The database anomaly detection method for a power plant statistical management system according to claim 6, characterized in that: Step S44 specifically includes: introducing a residual sensitivity factor and constructing a residual-driven-Shapley enhancement mechanism; based on the masking prediction results, evaluating the influence of multiple concept items on the prediction results of the trained LSTM model through the residual-driven-Shapley enhancement mechanism to obtain SHAP contribution data.

Images