A method and system for abnormal analysis of measurement data

Abstract

The present invention provides a method and system for abnormal analysis of metering data. The method comprises the following steps: obtaining power grid system component data and power grid metering data; performing component operation constraint analysis on the power grid system component data to generate component operation constraint data; constructing a dynamic topology network on the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network; performing dynamic load characteristic analysis on the power grid metering data based on the component dynamic topology detection network to generate dynamic load characteristic data; performing latent variable characterization drive analysis on the dynamic load characteristic data to generate latent variable time series data; performing multi-scale frequency decomposition processing on the latent variable time series data to generate frequency band energy distribution data. The present invention realizes efficient and accurate abnormal analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular to a method and system for analyzing abnormality of metering data. Background Art

[0002] In the power industry, the operational stability and reliability of the power grid system are crucial to ensuring the continuity and quality of power supply. Abnormal analysis of power metering data is one of the key tasks to ensure the normal operation of the power grid system. In the power grid system, power metering data will be affected by many factors, such as power load fluctuations, equipment failures, power theft, abnormal weather, etc. Traditional power metering data abnormal analysis methods mainly rely on manual observation and offline analysis, and have problems such as low analysis efficiency and limited accuracy. In order to meet the needs of abnormal analysis of power metering data in modern power grid systems, an intelligent method and system are needed. Summary of the invention

[0003] The present invention provides a method and system for analyzing abnormality of measurement data to solve at least one of the above technical problems.

[0004] To achieve the above object, the present invention provides a method for analyzing abnormality of measurement data, comprising the following steps:

[0005] Step S1: acquiring power grid system component data and power grid power metering data; performing component operation constraint analysis on the power grid system component data to generate component operation constraint data; constructing a dynamic topology network for the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network;

[0006] Step S2: Performing dynamic load characteristic analysis on the power metering data of the power grid based on the component dynamic topology detection network to generate dynamic load characteristic data; performing latent variable characterization drive analysis on the dynamic load characteristic data to generate latent variable time series data;

[0007] Step S3: performing multi-scale frequency decomposition processing on the latent variable time series data to generate frequency segment energy distribution data; performing frequency change characteristic analysis on the frequency segment energy distribution data to generate sub-frequency change characteristic data;

[0008] Step S4: Perform frequency evolution trajectory analysis on the latent variable time series data based on the sub-frequency change characteristic data to generate load frequency evolution trajectory data; compare the load frequency evolution trajectory data with the preset component load frequency evolution trajectory threshold to generate abnormal load frequency evolution trajectory data; perform abnormal source tracing on the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data;

[0009] Step S5: performing abnormal situation evolution analysis on the load frequency abnormality spatial data to generate frequency abnormality trend data; performing abnormal trend change prediction on the frequency abnormality trend data to obtain abnormal diffusion path data;

[0010] Step S6: Use a deep neural network to perform abnormal optimization decision on the abnormal diffusion path data to generate power abnormal optimization decision data; perform data-driven modeling on the power abnormal optimization decision data to construct a power metering data abnormal decision model to perform abnormal analysis operations on the power metering data.

[0011] The present invention obtains comprehensive power system information by acquiring power grid system component data and power metering data, providing basic data for abnormal analysis. Component operation constraint analysis identifies the normal range of component operation and helps determine the basis for determining component abnormalities. The construction of a dynamic topology network reflects the association between components in the power grid system and provides the real-time status of the power grid topology, providing a basis for subsequent abnormal analysis. Dynamic load feature analysis extracts load feature information such as power consumption and power fluctuation from power metering data to describe the load condition of the power grid system. Hidden variable characterization driven analysis converts dynamic load feature data into implicit key features to provide higher-dimensional data representation. Multi-scale frequency decomposition processing decomposes hidden variable time series data at different frequency scales, extracts frequency domain information, and reveals frequency features in power metering data. Frequency change feature analysis further analyzes frequency segment energy distribution data, identifies the pattern and trend of frequency changes, and provides a basis for subsequent abnormal analysis. Frequency evolution trajectory analysis reveals the change trajectory of load frequency in power metering data, helping to discover the pattern and trend of abnormal changes. Abnormal load frequency evolution trajectory data is compared with a preset threshold value for identification and Mark the abnormal load frequency evolution trajectory and provide accurate positioning of abnormal data. The load frequency abnormal spatial data spatially locates the abnormal load frequency evolution trajectory in the component dynamic topology detection network to help determine the specific location of the abnormality. The abnormal situation evolution analysis performs trend analysis on the load frequency abnormal spatial data to reveal the evolution process and trend of the abnormal situation. The abnormal change prediction of the frequency abnormal trend data predicts the future development direction of the abnormal situation based on historical data and trend analysis, and provides early warning and decision support. The abnormal diffusion path data describes the propagation path and impact range of the abnormal situation in the power grid system, helping to understand the propagation mechanism and impact degree of the abnormality. The use of deep neural networks for abnormal optimization decision-making provides optimized decision-making solutions for abnormal situations by learning abnormal diffusion path data and historical experience. The power abnormal optimization decision data is generated according to the deep neural network model, which includes specific handling measures and optimization suggestions for abnormal situations. Data-driven modeling associates and models the power abnormal optimization decision data with the power metering data, and constructs a power metering data abnormal decision model to support subsequent abnormal analysis work and provide guidance and decision support for the operation and management of the power system.

[0012] Preferably, step S1 comprises the following steps:

[0013] Step S11: Acquire power grid system component data and power grid power metering data;

[0014] Step S12: monitoring the operation status of the power grid system component data, thereby obtaining the power grid component operation data;

[0015] Step S13: performing component operation constraint analysis on the power grid system component data to generate component operation constraint data;

[0016] Step S14: performing topological structure distribution analysis on the power grid system component data based on the component operation constraint data to obtain component topological structure data;

[0017] Step S15: constructing a dynamic topology network for the component topology structure data through the power grid component operation data to generate a component dynamic topology detection network.

[0018] The present invention provides comprehensive power system information by obtaining power grid system component data and power metering data, including component configuration, parameters, location, etc., as well as the actual operation of the power system. The operation status monitoring monitors the operation status of each component in the power grid system in real time, including the actual measured values ​​of parameters such as current, voltage, frequency, etc., and provides real-time operation data of the components. The component operation constraint analysis identifies the normal range and restriction conditions of component operation, such as the rated power and voltage range of the components, and generates component operation constraint data as a reference for subsequent abnormal analysis. The topological structure distribution analysis reveals the association relationship and topological structure between components in the power grid system, including the connection mode and power flow direction between components, and generates component topological structure data to provide topological information of the power grid system. The construction of the dynamic topological network is based on the actual component operation data, reflecting the real-time relationship between components in the power grid system, including power flow direction, connection status, etc. The generated component dynamic topology detection network provides the real-time status of the power grid topology, providing a basis for subsequent abnormal analysis.

[0019] Preferably, the specific steps of step S2 are:

[0020] Step S21: performing dynamic load characteristic analysis on power grid power metering data based on the component dynamic topology detection network to generate dynamic load characteristic data;

[0021] Step S22: performing time series fluctuation analysis on the dynamic load characteristic data to generate time series load fluctuation data;

[0022] Step S23: performing fluctuation curve fitting on the time series load fluctuation data to construct a load fluctuation curve;

[0023] Step S24: performing latent variable characterization drive analysis on the load fluctuation curve to generate latent variable time series data.

[0024] The present invention uses the dynamic topology detection network and power metering data of the component through dynamic load feature analysis to extract and analyze the features of the load in the power grid. By analyzing the change trend, amplitude, frequency and other features of the power metering data, the dynamic characteristics of the load are obtained, and the corresponding dynamic load feature data is generated for subsequent abnormal detection and analysis to help find abnormal load conditions. The time series fluctuation analysis performs statistics and analysis on the dynamic load feature data to identify the time series fluctuation of the load. By calculating the fluctuation degree, frequency and other indicators of the load feature data, the time series load fluctuation data is generated, and the detailed information of the load fluctuation is provided, which provides a basis for subsequent abnormal detection and analysis. Dynamic curve fitting uses time-series load fluctuation data to construct a mathematical model of load fluctuation through mathematical modeling and curve fitting technology, describing the trend, periodicity, amplitude and other characteristics of load fluctuation, providing a more accurate characterization of load fluctuation behavior. The construction of the fluctuation curve provides a more accurate reference for anomaly detection and analysis. Hidden variable characterization-driven analysis uses load fluctuation curves to explore the potential variables and factors behind load fluctuations through hidden variable analysis methods. By extracting implicit information from the load fluctuation curve, hidden variable time series data is generated, providing an understanding and revelation of the mechanism behind the load fluctuation. The generation of hidden variable time series data provides deeper insights for anomaly detection and analysis.

[0025] Preferably, the specific steps of step S21 are:

[0026] Step S211: Calculate the load power distribution of the power grid metering data to obtain the load power distribution data of the power grid;

[0027] Step S212: Performing dynamic load harmonic analysis on the power grid metering data to generate dynamic complex harmonic data of the power grid;

[0028] Step S213: Perform dynamic load characteristic analysis on the grid load power distribution data and the grid dynamic complex harmonic data to generate dynamic load characteristic data.

[0029] The present invention analyzes and calculates the power of each load in the power grid based on the power metering data through load power distribution calculation, obtains the power distribution data of the power grid load by calculating the power proportion, power density and other indicators of different loads, provides the overall situation of the load power distribution and the relationship between different loads, and provides a basis for abnormal analysis. The dynamic load harmonic analysis identifies the harmonic components in the load by performing spectrum analysis and harmonic analysis on the power metering data, generates dynamic complex harmonic data of the power grid, including the amplitude, phase and other information of different harmonic components, reveals the harmonic problems and nonlinear characteristics in the load, and provides an important basis for abnormal analysis and power grid quality assessment. The dynamic load feature analysis uses the power distribution data of the power grid load and the dynamic complex harmonic data to extract and analyze the comprehensive features of the load, analyze the power distribution characteristics, harmonic characteristics and the relationship between power and harmonics of the load, and generates dynamic load feature data to more comprehensively describe the characteristics and behavior of the load, so as to provide support for abnormal analysis and load management.

[0030] Preferably, the specific steps of step S3 are:

[0031] Step S31: performing multi-scale frequency decomposition processing on the latent variable time series data to generate a multi-level frequency signal domain;

[0032] Step S32: performing energy intensity distribution analysis on the multi-level frequency signal domain to generate frequency segment energy distribution data;

[0033] Step S33: performing same-frequency signal domain recognition on the frequency segment energy distribution data to extract the same-frequency signal domain;

[0034] Step S34: reconstructing the scale frequency of the same frequency signal domain to generate a sub-frequency signal domain;

[0035] Step S35: performing frequency variation characteristic analysis on the sub-frequency signal domain to generate sub-frequency variation characteristic data.

[0036] The present invention decomposes latent variable time series data into signal domains of different frequency ranges through multi-scale frequency decomposition processing. By decomposing the data, frequency information at different scales is obtained, thereby obtaining a multi-level frequency signal domain, better capturing the frequency characteristics in the data, and providing a basis for subsequent frequency domain analysis. The energy intensity distribution analysis targets the multi-level frequency signal domain, calculates the energy intensity of each frequency segment, and obtains frequency segment energy distribution data by analyzing the energy distribution of different frequency segments, reflecting the energy changes of different frequency segments, helping to identify and analyze the abnormalities or abnormal changes of frequency segments. The same frequency signal domain recognition utilizes the frequency segment energy distribution data to identify signal domains with the same frequency characteristics, and extracts the same frequency signals. Domain, aggregates signals with similar frequency characteristics, divides them into the same group, better understands and analyzes the same frequency signal domains in different frequency bands, and provides a basis for subsequent analysis and processing. Scale frequency reconstruction targets the same frequency signal domain, performs scale transformation and frequency reconstruction, and obtains signals with finer frequency characteristics by reconstructing the sub-frequency signal domain. It further refines the frequency information and provides more details and accurate descriptions of frequency changes. Frequency change feature analysis uses the sub-frequency signal domain to analyze the frequency change characteristics therein, and generates sub-frequency change feature data by calculating indicators such as the frequency change rate and amplitude, providing frequency change conditions in the sub-frequency signal domain, helping to identify and analyze abnormal or abnormal frequency changes.

[0037] Preferably, the specific steps of step S31 are:

[0038] Step S311: performing frequency domain conversion on the latent variable time series data to generate dynamic load frequency domain data;

[0039] Step S312: extracting low-frequency components from the dynamic load frequency domain data to generate low-frequency load frequency domain component data;

[0040] Step S313: performing high-frequency component refinement processing on the dynamic load frequency domain data to generate high-frequency load frequency domain component data;

[0041] Step S314: performing frequency domain signal fusion on the low-frequency load frequency domain component data and the high-frequency load frequency domain component data to generate a multi-level frequency signal domain.

[0042] The present invention converts time domain data into frequency domain data through frequency domain conversion, and converts latent variable time series data into frequency domain representation by applying Fourier transform or other frequency domain conversion methods to obtain dynamic load frequency domain data. The frequency domain data provides the energy distribution and frequency characteristics of the load at different frequencies, and reveals the frequency components and frequency changes of the load. Low-frequency component extraction is aimed at dynamic load frequency domain data, and the low-frequency components therein are extracted. By extracting the low-frequency components, the energy distribution and frequency characteristics in the lower frequency range of the load are obtained, so as to better analyze and understand the behavior of the load in the low-frequency range, and provide a basis for abnormal analysis. The high-frequency component is refined. The processing extracts and processes the high-frequency details of the dynamic load frequency domain data. By processing the high-frequency components, the frequency characteristics and behaviors of the load in the high-frequency range are analyzed and described more specifically, providing more accurate and detailed high-frequency frequency information to help discover and identify abnormal conditions in the high-frequency range. The frequency domain signal fusion integrates and fuses the low-frequency load frequency domain component data and the high-frequency load frequency domain component data. By fusing the frequency domain component data in different frequency ranges, a multi-level frequency signal domain is generated. At the same time, the frequency characteristics in different frequency ranges are considered to provide more comprehensive and integrated frequency domain information, which is helpful for abnormal analysis and explanation of load behavior.

[0043] Preferably, the specific steps of step S4 are:

[0044] Step S41: performing frequency evolution trajectory analysis on latent variable time series data based on sub-frequency variation characteristic data to generate load frequency evolution trajectory data;

[0045] Step S42: comparing the preset element load frequency evolution trajectory threshold with the load frequency evolution trajectory data, when the preset element load frequency evolution trajectory threshold is less than or equal to the load frequency evolution trajectory data, it is determined that the power load is abnormal, and abnormal load frequency evolution trajectory data is generated;

[0046] Step S43: when the preset component load frequency evolution trajectory threshold is greater than the load frequency evolution trajectory data, it is determined that the power load is in a normal state;

[0047] Step S44: Based on the abnormal load frequency evolution trajectory data, the component dynamic topology detection network is traced and located to generate load frequency abnormality spatial data.

[0048] The present invention analyzes the latent variable time series data based on the sub-frequency change characteristic data through frequency evolution trajectory analysis, generates load frequency evolution trajectory data by analyzing the frequency change of the load in different time periods, observes and records the change pattern of the load frequency over time, and provides an overall understanding of the evolution of load behavior. By comparing the preset component load frequency evolution trajectory threshold with the actual load frequency evolution trajectory data, it is determined whether the load is abnormal. If the actual load frequency evolution trajectory data exceeds or equals the preset threshold, it is determined that the power load is abnormal, and abnormal load frequency evolution trajectory data is generated, and abnormal load conditions are automatically detected and identified. If the preset component load frequency evolution trajectory threshold is greater than the actual load frequency evolution trajectory data, the load is judged to be in a normal state, and the normal load situation is quickly judged to avoid false alarms or unnecessary processing of normal loads. Based on the abnormal load frequency evolution trajectory data, the component dynamic topology detection network is traced and located for abnormalities. By analyzing the propagation path and impact range of the abnormal load frequency evolution trajectory data in the topological network, the specific component or area causing the abnormality is determined, and the load frequency abnormality spatial data is generated. The spatial positioning information of the abnormal load is provided, which is helpful for further analysis and resolution of abnormal situations.

[0049] Preferably, the specific steps of step S5 are:

[0050] Step S51: performing principal component analysis on the load frequency abnormality spatial data to generate low-dimensional frequency abnormality feature data;

[0051] Step S52: performing abnormal situation evolution analysis on the low-dimensional frequency abnormal feature data to generate frequency abnormal trend data;

[0052] Step S53: predicting abnormal trend changes of the frequency abnormal trend data to generate load frequency abnormal prediction data;

[0053] Step S54: performing quantitative diagnosis of abnormal diffusion paths on the load frequency abnormality prediction data, thereby obtaining abnormal diffusion path data.

[0054] The present invention is a dimensionality reduction technology through principal component analysis. By performing principal component analysis on load frequency anomaly spatial data, high-dimensional data is converted into low-dimensional data, and the most representative principal component is extracted. The generated low-dimensional frequency anomaly feature data helps to simplify data representation and analysis, reduce data redundancy, and extract the core features of load frequency anomaly. By performing abnormal situation evolution analysis on low-dimensional frequency anomaly feature data, the evolution trend of load frequency anomaly is revealed. The generated frequency anomaly trend data shows the change pattern and trend of load frequency anomaly, which helps to better understand and predict the development trend of load frequency anomaly. By performing abnormal trend change prediction on frequency anomaly trend data, the future development of load frequency anomaly is predicted. The generated load frequency anomaly prediction data provides prediction and early warning of future abnormal situations, helps to take appropriate measures to deal with potential abnormal situations. By performing abnormal diffusion path quantitative diagnosis on load frequency anomaly prediction data, the propagation path and influence range of abnormal situations in the power system are determined. The generated abnormal diffusion path data provides quantitative information on the spatial positioning and diffusion degree of abnormal situations, which helps to accurately locate the abnormal source and evaluate its impact on the system.

[0055] Preferably, the specific steps of step S6 are:

[0056] Step S61: using a deep neural network to perform abnormal feature space learning processing on the abnormal diffusion path data to generate power abnormal feature data;

[0057] Step S62: performing abnormal optimization decision on the power abnormality feature data to generate power abnormality optimization decision data;

[0058] Step S63: Perform data-driven modeling on the power anomaly optimization decision data, and construct a power metering data anomaly decision model to perform an anomaly analysis operation on the power metering data.

[0059] The present invention uses a deep neural network to perform abnormal feature space learning processing on abnormal diffusion path data to extract key features of power anomalies. The deep neural network has powerful learning and representation capabilities and learns effective feature representation from complex abnormal diffusion path data. The generated power anomaly feature data better describes and represents power anomalies, providing more accurate input for subsequent abnormal optimization decisions. After obtaining the power anomaly feature data, an abnormal optimization decision is made to formulate corresponding optimization strategies for different abnormal situations. By analyzing and making decisions on the power anomaly feature data, power anomaly optimization decision data is generated, which includes specific decision measures for abnormal situations. These decision data help operators take targeted measures to solve power anomaly problems and improve the reliability and efficiency of the power system. By performing data-driven modeling on the power anomaly optimization decision data, a power metering data anomaly decision model is constructed. Based on historical data and decision data, machine learning or other modeling technologies are used for training and optimization. The constructed power metering data anomaly decision model is used to perform abnormal analysis operations of power metering data, automatically detect, diagnose and process power anomalies, improve the efficiency and accuracy of anomaly analysis, and reduce the need for manual intervention.

[0060] In this specification, a measurement data abnormality analysis system is provided, which is used to execute the measurement data abnormality analysis method as described above, including:

[0061] A topology network module is used to obtain power grid system component data and power grid power metering data; perform component operation constraint analysis on the power grid system component data to generate component operation constraint data; and perform dynamic topology network construction on the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network;

[0062] The load characteristic analysis module is used to perform dynamic load characteristic analysis on the power metering data of the power grid based on the component dynamic topology detection network to generate dynamic load characteristic data; perform latent variable characterization driving analysis on the dynamic load characteristic data to generate latent variable time series data;

[0063] The frequency decomposition module is used to perform multi-scale frequency decomposition processing on the latent variable time series data to generate frequency band energy distribution data; perform frequency change characteristic analysis on the frequency band energy distribution data to generate sub-frequency change characteristic data;

[0064] The frequency evolution trajectory module is used to perform frequency evolution trajectory analysis on the hidden variable time series data based on the sub-frequency change characteristic data to generate load frequency evolution trajectory data; compare the preset component load frequency evolution trajectory threshold with the load frequency evolution trajectory data to generate abnormal load frequency evolution trajectory data; and perform abnormal source tracing and positioning on the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data;

[0065] The trend prediction module is used to analyze the abnormal situation evolution of the load frequency abnormality spatial data and generate frequency abnormality trend data; predict the abnormal trend change of the frequency abnormality trend data to obtain abnormal diffusion path data;

[0066] The abnormal optimization decision module is used to use deep neural networks to make abnormal optimization decisions on abnormal diffusion path data to generate power abnormal optimization decision data; perform data-driven modeling on the power abnormal optimization decision data, and build a power metering data abnormal decision model to perform abnormal analysis operations on the power metering data.

[0067] The present invention obtains relevant data of various components (such as generators, transformers, lines, etc.) in the power grid system and data related to power metering, including load, voltage, current and other information, through the topological network module. By analyzing the data of the power grid system components, the operating constraints of the components are identified, including rated capacity, voltage range, frequency range, etc. The generated component operating constraint data helps the subsequent dynamic topological network construction and load feature analysis. The load feature analysis module analyzes the power metering data of the power grid based on the component dynamic topological detection network, and extracts the dynamic characteristics of the load, including load power change, power factor, harmonic conditions, etc. The generated dynamic load The characteristic data reflects the real-time changes of the load in the power grid. The dynamic load characteristic data is analyzed to extract potential hidden variables, that is, higher-level information hidden behind the load characteristic data. The generated hidden variable time series data helps to understand the inherent laws and evolution trends of load behavior. The frequency decomposition module performs frequency decomposition processing on the hidden variable time series data, converts the time domain data into frequency domain data, and considers multiple frequency scales at the same time. The generated frequency band energy distribution data provides energy distribution in different frequency ranges. The frequency band energy distribution data is analyzed to extract frequency change characteristics, including frequency offset, frequency disturbance, etc., and the generated sub-frequency change characteristic data Reveal the change pattern and trend of load frequency. The frequency evolution trajectory module analyzes the frequency evolution trajectory of hidden variable time series data based on sub-frequency change feature data to reveal the evolution trajectory of load frequency. The generated load frequency evolution trajectory data helps to understand the change law and trend of load frequency. Based on the preset component load frequency evolution trajectory threshold and load frequency evolution trajectory data, the component dynamic topology detection network is traced and located to determine the propagation path and impact range of abnormal conditions in the power system. The generated abnormal load frequency evolution trajectory data helps to accurately locate the abnormal source and evaluate its impact on the system. The abnormal optimization decision module uses deep neural networks to make abnormal optimization decisions on abnormal diffusion path data and formulate optimization strategies for abnormal conditions. The generated power abnormal optimization decision data helps operators take targeted measures to solve power abnormal problems and improve the reliability and efficiency of the power system. Data-driven modeling is performed on power abnormal optimization decision data to build a power metering data abnormal decision model. Based on historical data and decision data, a prediction model is established using machine learning or statistical methods to perform abnormal analysis of power metering data. The model identifies potential abnormal patterns by learning the correlation between abnormal conditions and corresponding decision data, and provides abnormal analysis results and suggestions. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 A schematic diagram of the steps of a method for analyzing abnormality of measurement data according to the present invention;

[0069] Figure 2Detailed implementation flow chart of step S1;

[0070] Figure 3 Detailed implementation flow chart of step S2;

[0071] Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0072] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0073] The present application example provides a method and system for abnormal analysis of metering data. The execution subject of the abnormal analysis method and system for metering data includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of an audio and image management system, an information management system, and a cloud data management system.

[0074] See also Figures 1 to 4 The present invention provides a method for analyzing abnormality of measurement data, comprising the following steps:

[0075] Step S1: acquiring power grid system component data and power grid power metering data; performing component operation constraint analysis on the power grid system component data to generate component operation constraint data; constructing a dynamic topology network for the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network;

[0076] Step S2: Performing dynamic load characteristic analysis on the power metering data of the power grid based on the component dynamic topology detection network to generate dynamic load characteristic data; performing latent variable characterization drive analysis on the dynamic load characteristic data to generate latent variable time series data;

[0077] Step S3: performing multi-scale frequency decomposition processing on the latent variable time series data to generate frequency segment energy distribution data; performing frequency change characteristic analysis on the frequency segment energy distribution data to generate sub-frequency change characteristic data;

[0078] Step S4: Perform frequency evolution trajectory analysis on the latent variable time series data based on the sub-frequency change characteristic data to generate load frequency evolution trajectory data; compare the load frequency evolution trajectory data with the preset component load frequency evolution trajectory threshold to generate abnormal load frequency evolution trajectory data; perform abnormal source tracing on the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data;

[0079] Step S5: performing abnormal situation evolution analysis on the load frequency abnormality spatial data to generate frequency abnormality trend data; performing abnormal trend change prediction on the frequency abnormality trend data to obtain abnormal diffusion path data;

[0080] Step S6: Use a deep neural network to perform abnormal optimization decision on the abnormal diffusion path data to generate power abnormal optimization decision data; perform data-driven modeling on the power abnormal optimization decision data to construct a power metering data abnormal decision model to perform abnormal analysis operations on the power metering data.

[0081] The present invention obtains comprehensive power system information by acquiring power grid system component data and power metering data, providing basic data for abnormal analysis. Component operation constraint analysis identifies the normal range of component operation and helps determine the basis for determining component abnormalities. The construction of a dynamic topology network reflects the association between components in the power grid system and provides the real-time status of the power grid topology, providing a basis for subsequent abnormal analysis. Dynamic load feature analysis extracts load feature information such as power consumption and power fluctuation from power metering data to describe the load condition of the power grid system. Hidden variable characterization driven analysis converts dynamic load feature data into implicit key features to provide higher-dimensional data representation. Multi-scale frequency decomposition processing decomposes hidden variable time series data at different frequency scales, extracts frequency domain information, and reveals frequency features in power metering data. Frequency change feature analysis further analyzes frequency segment energy distribution data, identifies the pattern and trend of frequency changes, and provides a basis for subsequent abnormal analysis. Frequency evolution trajectory analysis reveals the change trajectory of load frequency in power metering data, helping to discover the pattern and trend of abnormal changes. Abnormal load frequency evolution trajectory data is compared with a preset threshold value for identification and Mark the abnormal load frequency evolution trajectory and provide accurate positioning of abnormal data. The load frequency abnormal spatial data spatially locates the abnormal load frequency evolution trajectory in the component dynamic topology detection network to help determine the specific location of the abnormality. The abnormal situation evolution analysis performs trend analysis on the load frequency abnormal spatial data to reveal the evolution process and trend of the abnormal situation. The abnormal change prediction of the frequency abnormal trend data predicts the future development direction of the abnormal situation based on historical data and trend analysis, and provides early warning and decision support. The abnormal diffusion path data describes the propagation path and impact range of the abnormal situation in the power grid system, helping to understand the propagation mechanism and impact degree of the abnormality. The use of deep neural networks for abnormal optimization decision-making provides optimized decision-making solutions for abnormal situations by learning abnormal diffusion path data and historical experience. The power abnormal optimization decision data is generated according to the deep neural network model, which includes specific handling measures and optimization suggestions for abnormal situations. Data-driven modeling associates and models the power abnormal optimization decision data with the power metering data, and constructs a power metering data abnormal decision model to support subsequent abnormal analysis work and provide guidance and decision support for the operation and management of the power system.

[0082] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of the steps of a method for analyzing abnormality of measurement data of the present invention. In this example, the steps of the method for analyzing abnormality of measurement data include:

[0083] Step S1: acquiring power grid system component data and power grid power metering data; performing component operation constraint analysis on the power grid system component data to generate component operation constraint data; constructing a dynamic topology network for the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network;

[0084] In this embodiment, power grid system component data is collected and obtained, including information on various devices, lines, transformers, generators and other components of the power grid. At the same time, power grid power metering data is obtained, including power supply and demand data, load data, etc. The power grid system component data is analyzed to determine the operating constraints of each component, including the rated power, voltage range, current range, temperature limit, etc. of the component. The operating constraint data of the component is obtained by analyzing the technical specifications and operating requirements of the component. According to the power grid system component data, an initial topological network is established, including various components and the connection relationship between them. According to the component operation constraint data, the initial topological network is adjusted and optimized to ensure that the components in the network meet the operating constraints. Operations such as scheduling of components, adjustment of connection relationships, and load balancing are involved. In order to achieve safe and stable operation of the power grid, it is necessary to monitor and detect topological changes of the power grid in real time. Based on the component operation constraint data, a dynamic topological detection network is constructed to monitor the dynamic changes of components in the power grid. This network detects the connection status, operation status and association relationship of components in the power grid with other components in real time.

[0085] Step S2: Performing dynamic load characteristic analysis on the power metering data of the power grid based on the component dynamic topology detection network to generate dynamic load characteristic data; performing latent variable characterization drive analysis on the dynamic load characteristic data to generate latent variable time series data;

[0086] In this embodiment, dynamic load characteristic analysis is performed based on the component dynamic topology detection network and power metering data, including analyzing the load status of each component in the power grid, such as current changes, power fluctuations, etc., and using statistical analysis, data mining, and machine learning methods to extract characteristic information of the components and generate dynamic load characteristic data, including load characteristic vectors of the components, such as current change curves, power fluctuation characteristics, etc., which are used to describe the dynamic changes of components in the power grid system. The data is reduced in dimension or feature extracted to capture implicit variables and association patterns in the data. Commonly used methods include principal component analysis (PCA), independent component analysis (ICA), etc.

[0087] Step S3: performing multi-scale frequency decomposition processing on the latent variable time series data to generate frequency segment energy distribution data; performing frequency change characteristic analysis on the frequency segment energy distribution data to generate sub-frequency change characteristic data;

[0088] In this embodiment, multi-scale frequency decomposition processing is performed on the latent variable time series data. This processing method can convert the time series data in the time domain into frequency data in the frequency domain, and perform frequency decomposition at different scales. Common multi-scale frequency decomposition methods include wavelet transform, Fourier transform, etc. According to the results of multi-scale frequency decomposition, the energy distribution data of each frequency band is calculated. For each frequency band, the proportion of its energy to the total energy or the statistical indicators of energy, such as mean, variance, etc., are calculated. Based on the frequency band energy distribution data, the frequency change characteristics are analyzed. These characteristics include the changing trend of frequency band energy, the volatility of frequency band energy, the relationship between frequency band energies, etc. According to the results of the frequency change characteristic analysis, the frequency change characteristics are converted into sub-frequency change characteristic data, which is a description of the frequency band energy change trend, the statistical indicators of the frequency band energy volatility, etc.

[0089] Step S4: Perform frequency evolution trajectory analysis on the latent variable time series data based on the sub-frequency change characteristic data to generate load frequency evolution trajectory data; compare the load frequency evolution trajectory data with the preset component load frequency evolution trajectory threshold to generate abnormal load frequency evolution trajectory data; perform abnormal source tracing on the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data;

[0090] In this embodiment, based on the sub-frequency change characteristic data, the frequency evolution trajectory analysis is performed on the hidden variable time series data, the frequency change of the hidden variable in different time periods is tracked, and the load frequency evolution trajectory data is generated. According to the preset element load frequency evolution trajectory threshold, the generated load frequency evolution trajectory data is compared with the threshold to determine whether the load frequency evolution trajectory is abnormal. According to the comparison result, the load frequency evolution trajectory data exceeding the preset threshold is marked as abnormal. These abnormal load frequency evolution trajectory data indicate that there are abnormal changes in the frequency evolution of the load. Based on the abnormal load frequency evolution trajectory data, the component dynamic topology detection network is traced and located for abnormality. By analyzing the connection relationship between the abnormal load frequency evolution trajectory data and the components, the component or component combination that causes the load abnormality is determined, and the load frequency abnormality spatial data is generated.

[0091] Step S5: performing abnormal situation evolution analysis on the load frequency abnormality spatial data to generate frequency abnormality trend data; performing abnormal trend change prediction on the frequency abnormality trend data to obtain abnormal diffusion path data;

[0092] In this embodiment, the load frequency anomaly spatial data is analyzed for abnormal situation evolution, the changing trend of abnormal load frequency is tracked, and frequency anomaly trend data is generated. Statistical methods, time series analysis and other techniques are used to analyze the evolution of the abnormal situation. Based on the frequency anomaly trend data, the abnormal trend change is predicted. This prediction process uses machine learning, time series prediction and other methods to analyze the evolution law of the abnormal trend and predict the future abnormal diffusion path. The prediction result is expressed as abnormal diffusion path data.

[0093] Step S6: Use a deep neural network to perform abnormal optimization decision on the abnormal diffusion path data to generate power abnormal optimization decision data; perform data-driven modeling on the power abnormal optimization decision data to construct a power metering data abnormal decision model to perform abnormal analysis operations on the power metering data.

[0094] In this embodiment, a deep neural network is used to perform abnormal optimization decisions on abnormal diffusion path data. The deep neural network learns the patterns and laws in the abnormal diffusion path data and generates power abnormality optimization decision data. These decision data include optimization plans for abnormal diffusion paths, abnormality handling strategies, etc. Based on the power abnormality optimization decision data, data-driven modeling is performed. This modeling process uses machine learning, statistical modeling and other methods to construct a power metering data abnormality decision model. The model performs abnormal analysis operations based on the input power metering data to identify and report abnormal situations.

[0095] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0096] Step S11: Acquire power grid system component data and power grid power metering data;

[0097] Step S12: monitoring the operation status of the power grid system component data, thereby obtaining the power grid component operation data;

[0098] Step S13: performing component operation constraint analysis on the power grid system component data to generate component operation constraint data;

[0099] Step S14: performing topological structure distribution analysis on the power grid system component data based on the component operation constraint data to obtain component topological structure data;

[0100] Step S15: constructing a dynamic topology network for the component topology structure data through the power grid component operation data to generate a component dynamic topology detection network.

[0101] In this embodiment, the data of power grid system components in the power grid system are obtained, and the power grid system components include transformers, switches, lines, generators, etc. The power grid system component data include information such as the type, location, parameters, and status of the components. The power metering data of the power grid is obtained, including current, voltage, power factor, frequency, and other data, which are collected through electric energy metering equipment, sensors, etc. The operating status of the components in the power grid system is monitored, and the operating data of the components are obtained by real-time monitoring of the parameters and status of the components, such as current, voltage, temperature, etc. Based on the operating status monitoring, the operating data of the power grid components are obtained, including the real-time parameter values, status information, operating status changes, etc. of the components, and the operating constraint analysis is performed on the component data in the power grid system. The parameter range and performance requirements of the grid components are used to determine the operating constraints of the components. According to the results of the operation constraint analysis, the component operation constraint data is generated, which describes the operating constraints of each component, such as the upper and lower limits of current, the voltage stability range, etc. Based on the component operation constraint data, the component data in the power grid system is subjected to topological structure distribution analysis to determine the connection relationship, topological structure and other information between the components. Through the topological structure distribution analysis, the component topological structure data is generated, which describes the relationship between the components in the power grid system, such as the connection mode, topological diagram, etc. Based on the grid component operation data and the component topological structure data, the component dynamic topological network is constructed, and the operating status and topological structure of the component are combined to form a dynamic component topology detection network.

[0102] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0103] Step S21: performing dynamic load characteristic analysis on power grid power metering data based on the component dynamic topology detection network to generate dynamic load characteristic data;

[0104] Step S22: performing time series fluctuation analysis on the dynamic load characteristic data to generate time series load fluctuation data;

[0105] Step S23: performing fluctuation curve fitting on the time series load fluctuation data to construct a load fluctuation curve;

[0106] Step S24: performing latent variable characterization drive analysis on the load fluctuation curve to generate latent variable time series data.

[0107] In this embodiment, dynamic load characteristic analysis is performed based on the component dynamic topology detection network and power metering data, including analyzing the load status of each component in the power grid, such as current change, power fluctuation, etc., using statistical analysis, data mining and machine learning methods to extract characteristic information of the component, generate dynamic load characteristic data, including load characteristic vectors of the component, such as current change curve, power fluctuation characteristics, etc., which are used to describe the dynamic changes of the components in the power grid system, perform time series fluctuation analysis on the dynamic load characteristic data, including calculating statistical indicators such as time series difference, standard deviation, and root mean square of the load characteristic data to reveal the time series fluctuation of the load data, fit the fluctuation curve of the time series load fluctuation data, use a suitable mathematical model or curve fitting algorithm, such as polynomial fitting, spline interpolation, etc., to construct a load fluctuation curve, perform hidden variable characterization driving analysis on the load fluctuation curve, use dimensionality reduction technology, such as principal component analysis (PCA) or independent component analysis (ICA), convert the load fluctuation curve into a hidden variable representation, generate hidden variable time series data, which is used to describe the key features and implicit information in the load fluctuation curve, thereby providing a higher dimensional load data representation.

[0108] In this embodiment, the specific steps of step S21 are:

[0109] Step S211: Calculate the load power distribution of the power grid metering data to obtain the load power distribution data of the power grid;

[0110] Step S212: Performing dynamic load harmonic analysis on the power grid metering data to generate dynamic complex harmonic data of the power grid;

[0111] Step S213: Perform dynamic load characteristic analysis on the grid load power distribution data and the grid dynamic complex harmonic data to generate dynamic load characteristic data.

[0112] In this embodiment, the load power distribution calculation is performed on the power metering data, and the power value of each component is obtained by multiplying the current and voltage data. Then, statistical analysis is performed to calculate the load proportion or power density distribution in different power intervals. Dynamic load harmonic analysis is performed on the power metering data. By using harmonic analysis technology, such as fast Fourier transform (FFT), spectrum analysis is performed on the load power data to extract harmonic components. Based on the power grid load power distribution data and the power grid dynamic complex harmonic data, dynamic load feature analysis is performed, including statistical analysis of load power distribution, such as mean, variance, kurtosis and other indicators, and spectrum feature analysis of harmonic data, such as main harmonic frequency, harmonic distortion degree, etc. According to the analysis results, dynamic load feature data is generated to describe the characteristics of the power grid load, including power distribution characteristics, harmonic characteristics, etc., and provide information about the changes and complexity of the load at different time points.

[0113] In this embodiment, reference Figure 4 The above is a schematic flow chart of the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0114] Step S31: performing multi-scale frequency decomposition processing on the latent variable time series data to generate a multi-level frequency signal domain;

[0115] Step S32: performing energy intensity distribution analysis on the multi-level frequency signal domain to generate frequency segment energy distribution data;

[0116] Step S33: performing same-frequency signal domain recognition on the frequency segment energy distribution data to extract the same-frequency signal domain;

[0117] Step S34: reconstructing the scale frequency of the same frequency signal domain to generate a sub-frequency signal domain;

[0118] Step S35: performing frequency variation characteristic analysis on the sub-frequency signal domain to generate sub-frequency variation characteristic data.

[0119] In this embodiment, multi-scale frequency decomposition processing is performed on the latent variable time series data. A signal processing technology, such as wavelet transform or Fourier transform, is used to convert the time domain signal into a frequency domain signal. According to the result of the multi-scale frequency decomposition processing, a multi-level frequency signal domain is generated to represent the characteristic components of the latent variable data in different frequency ranges, including low frequency, medium frequency and high frequency. An energy intensity distribution analysis is performed on the multi-level frequency signal domain. The energy intensity of each frequency segment is calculated, for example, the square of the amplitude is used to represent the energy of the signal. According to the result of the energy intensity distribution analysis, frequency segment energy distribution data is generated to represent the energy distribution in different frequency segments, including information such as the range and energy intensity of the frequency segment. The frequency segment energy distribution data is identified in the same frequency signal domain. A threshold is set to determine which frequency segments have energy intensities exceeding the threshold, thereby determining which frequency segments have energy intensities exceeding the threshold. Determine the same frequency signal domain, extract the same frequency signal domain according to the result of the same frequency signal domain identification, and take it as the object of further analysis, reconstruct the scale frequency of the same frequency signal domain, and convert the frequency domain signal back to the time domain signal by using inverse wavelet transform or other inverse transform technology, and generate sub-frequency signal domain according to the result of scale frequency reconstruction. These signal domains represent the sub-frequency characteristics of the same frequency signal domain, which helps to further analyze the frequency change of latent variable data, and perform frequency change characteristic analysis on the sub-frequency signal domain, including calculating the frequency change rate, frequency change amplitude, frequency change trend and other indicators to describe the frequency change characteristics of the sub-frequency signal domain. According to the results of the frequency change characteristic analysis, generate sub-frequency change characteristic data to describe the frequency change characteristics of the sub-frequency signal domain, so as to further understand and analyze the frequency change behavior of the latent variable data.

[0120] In this embodiment, the specific steps of step S31 are:

[0121] Step S311: performing frequency domain conversion on the latent variable time series data to generate dynamic load frequency domain data;

[0122] Step S312: extracting low-frequency components from the dynamic load frequency domain data to generate low-frequency load frequency domain component data;

[0123] Step S313: performing high-frequency component refinement processing on the dynamic load frequency domain data to generate high-frequency load frequency domain component data;

[0124] Step S314: performing frequency domain signal fusion on the low-frequency load frequency domain component data and the high-frequency load frequency domain component data to generate a multi-level frequency signal domain.

[0125] In this embodiment, the latent variable time series data is frequency-domain converted. Common frequency-domain conversion methods include Fourier transform (FFT) and fast Fourier transform (FFT), etc. This converts the time-domain data into frequency-domain data, and obtains the energy distribution of the signal at different frequencies. According to the result of the frequency-domain conversion, dynamic load frequency-domain data is generated, which represents the energy distribution of the latent variable data at different frequencies, and can be used to further analyze the load characteristics, extract low-frequency components from the dynamic load frequency-domain data, and select the frequency components in the low-frequency range by setting a threshold, or filter high-frequency noise by using techniques such as filters, and generate low-frequency load frequency-domain component data according to the result of the low-frequency component extraction, which represents the load energy distribution in the low-frequency range, which is helpful Capture the long-term change characteristics of the load, refine the high-frequency components of the dynamic load frequency domain data, extract the frequency components in the high-frequency range through filters, downsampling or other technologies, so as to describe the short-term change characteristics of the load in a finer granularity. According to the results of the high-frequency component refinement processing, generate high-frequency load frequency domain component data, which represents the load energy distribution in the high-frequency range, and helps to capture the instantaneous change characteristics of the load. Perform frequency domain signal fusion on the low-frequency load frequency domain component data and the high-frequency load frequency domain component data, and merge the low-frequency and high-frequency components into a multi-level frequency signal domain through weighted averaging, linear combination or other fusion technologies. The fused data will contain the load energy distribution in different frequency ranges, providing more comprehensive load characteristic information.

[0126] In this embodiment, the specific steps of step S4 are:

[0127] Step S41: performing frequency evolution trajectory analysis on latent variable time series data based on sub-frequency variation characteristic data to generate load frequency evolution trajectory data;

[0128] Step S42: comparing the preset element load frequency evolution trajectory threshold with the load frequency evolution trajectory data, when the preset element load frequency evolution trajectory threshold is less than or equal to the load frequency evolution trajectory data, it is determined that the power load is abnormal, and abnormal load frequency evolution trajectory data is generated;

[0129] Step S43: when the preset component load frequency evolution trajectory threshold is greater than the load frequency evolution trajectory data, it is determined that the power load is in a normal state;

[0130] Step S44: Based on the abnormal load frequency evolution trajectory data, the component dynamic topology detection network is traced and located to generate load frequency abnormality spatial data.

[0131] In this embodiment, sub-frequency change characteristics are extracted from the latent variable time series data, and the change trend of the load frequency is captured by calculating the frequency change rate, the frequency evolution curve, etc. Based on the sub-frequency change characteristic data, the frequency evolution trajectory analysis is performed on the latent variable time series data, and the evolution of the load frequency is displayed by drawing a curve or trajectory of the frequency change over time. According to the result of the frequency evolution trajectory analysis, the load frequency evolution trajectory data is generated, which represents the change pattern of the load frequency over time and can be used for subsequent abnormality detection and positioning. The preset element load frequency evolution trajectory threshold is compared with the generated load frequency evolution trajectory data. If the preset element load frequency evolution trajectory threshold is less than or equal to the load frequency The evolution trajectory data indicates that the load is abnormal and needs further processing. Otherwise, it means that the load is in a normal state. If the load is judged to be abnormal, the abnormal load frequency evolution trajectory data is generated according to the load frequency evolution trajectory data, which records the time period of the load abnormality and the specific frequency evolution. Based on the abnormal load frequency evolution trajectory data, the component dynamic topology detection network is traced and located for abnormalities. By analyzing the propagation path, impact range and other information of the load frequency abnormality, the source of the abnormality and the affected components are determined. According to the results of the abnormality tracing and positioning, the load frequency abnormality spatial data is generated to record the spatial distribution of the load frequency abnormality, including the location, connection relationship and other information of the abnormality source and the affected components.

[0132] In this embodiment, the specific steps of step S5 are:

[0133] Step S51: performing principal component analysis on the load frequency abnormality spatial data to generate low-dimensional frequency abnormality feature data;

[0134] Step S52: performing abnormal situation evolution analysis on the low-dimensional frequency abnormal feature data to generate frequency abnormal trend data;

[0135] Step S53: predicting abnormal trend changes of the frequency abnormal trend data to generate load frequency abnormal prediction data;

[0136] Step S54: performing quantitative diagnosis of abnormal diffusion paths on the load frequency abnormality prediction data, thereby obtaining abnormal diffusion path data.

[0137] In this embodiment, principal component analysis is performed on the load frequency anomaly spatial data. Principal component analysis is a dimensionality reduction technology that converts high-dimensional data into low-dimensional data, retains as much variability of the original data as possible, and calculates the principal components of the load frequency anomaly spatial data. These principal components are the most representative features of the load frequency anomaly spatial data. According to the variance contribution rate or other evaluation indicators of the principal components, the principal component with the highest variance contribution rate is selected as the low-dimensional frequency anomaly feature data. An abnormal situation evolution analysis is performed on the low-dimensional frequency anomaly feature data, including using time series analysis methods, machine learning algorithms or other related technologies to identify and analyze the evolution trend of frequency anomalies, extracting trend data of frequency anomalies from the abnormal situation evolution analysis, and describing the evolution pattern of frequency anomalies, such as rise, fall, fluctuation, etc. , organize and record the extracted frequency anomaly trend data, generate frequency anomaly trend data for subsequent anomaly trend prediction and analysis, predict the abnormal trend change of the frequency anomaly trend data, use time series analysis, regression analysis, neural network and other methods to predict the future change trend of frequency anomalies, and generate load frequency anomaly prediction data based on the results of the abnormal trend change prediction, describing the predicted frequency anomaly trend for further quantitative diagnosis of the anomaly diffusion path, and perform quantitative diagnosis of the anomaly diffusion path on the load frequency anomaly prediction data, including using network analysis methods, propagation models or other related technologies to determine the path and degree of anomaly diffusion, extract the path data of anomaly diffusion from the quantitative diagnosis of anomaly diffusion path, and describe the propagation path and impact range of the load frequency anomaly in the system.

[0138] In this embodiment, the specific steps of step S6 are:

[0139] Step S61: using a deep neural network to perform abnormal feature space learning processing on the abnormal diffusion path data to generate power abnormal feature data;

[0140] Step S62: performing abnormal optimization decision on the power abnormality feature data to generate power abnormality optimization decision data;

[0141] Step S63: Perform data-driven modeling on the power anomaly optimization decision data, and construct a power metering data anomaly decision model to perform an anomaly analysis operation on the power metering data.

[0142] In this embodiment, the constructed deep neural network model is used to perform feature learning on the abnormal diffusion path data. Through the forward propagation process of the network, the abnormal diffusion path data is mapped to a low-dimensional power anomaly feature space. The power anomaly feature space data learned by the deep neural network is used as output to generate power anomaly feature data, which contains key features in the abnormal diffusion path data and is used for subsequent abnormal optimization decisions. The power anomaly feature data is input into the abnormal optimization decision algorithm to execute the abnormal optimization decision. The algorithm analyzes the power anomaly feature data and generates corresponding optimization decision results. The results obtained by executing the abnormal optimization decision are sorted and recorded to generate power anomaly optimization decision data, which describes the optimization measures and decision suggestions for power anomalies and the power anomaly optimization decision. Data preparation work is carried out, including data cleaning, feature selection, data conversion and other operations to meet the needs of modeling. Based on the power anomaly optimization decision data, a data-driven modeling method is used to build a power metering data anomaly decision model, including the use of machine learning algorithms, statistical analysis methods or other related technologies to establish the model, and the power anomaly optimization decision data is used to train the established model, and the performance and accuracy of the model are evaluated through cross-validation, indicator evaluation and other methods. The trained power metering data anomaly decision model is used to perform anomaly analysis on the actual power metering data, and the power metering data is input into the model. The model will determine whether there is an abnormality in the input data based on the learned abnormal decision rules and features, and provide corresponding abnormal analysis results.

[0143] In this embodiment, a measurement data abnormality analysis system is provided, which is used to execute the measurement data abnormality analysis method as described above, including:

[0144] A topology network module is used to obtain power grid system component data and power grid power metering data; perform component operation constraint analysis on the power grid system component data to generate component operation constraint data; and perform dynamic topology network construction on the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network;

[0145] The load characteristic analysis module is used to perform dynamic load characteristic analysis on the power metering data of the power grid based on the component dynamic topology detection network to generate dynamic load characteristic data; perform latent variable characterization driving analysis on the dynamic load characteristic data to generate latent variable time series data;

[0146] The frequency decomposition module is used to perform multi-scale frequency decomposition processing on the latent variable time series data to generate frequency band energy distribution data; perform frequency change characteristic analysis on the frequency band energy distribution data to generate sub-frequency change characteristic data;

[0147] The frequency evolution trajectory module is used to perform frequency evolution trajectory analysis on the hidden variable time series data based on the sub-frequency change characteristic data to generate load frequency evolution trajectory data; compare the preset component load frequency evolution trajectory threshold with the load frequency evolution trajectory data to generate abnormal load frequency evolution trajectory data; and perform abnormal source tracing and positioning on the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data;

[0148] The trend prediction module is used to analyze the abnormal situation evolution of the load frequency abnormality spatial data and generate frequency abnormality trend data; predict the abnormal trend change of the frequency abnormality trend data to obtain abnormal diffusion path data;

[0149] The abnormal optimization decision module is used to use deep neural networks to make abnormal optimization decisions on abnormal diffusion path data to generate power abnormal optimization decision data; perform data-driven modeling on the power abnormal optimization decision data, and build a power metering data abnormal decision model to perform abnormal analysis operations on the power metering data.

[0150] The present invention obtains relevant data of various components (such as generators, transformers, lines, etc.) in the power grid system and data related to power metering, including load, voltage, current and other information, through the topological network module. By analyzing the data of the power grid system components, the operating constraints of the components are identified, including rated capacity, voltage range, frequency range, etc. The generated component operating constraint data helps the subsequent dynamic topological network construction and load feature analysis. The load feature analysis module analyzes the power metering data of the power grid based on the component dynamic topological detection network, and extracts the dynamic characteristics of the load, including load power change, power factor, harmonic conditions, etc. The generated dynamic load The characteristic data reflects the real-time changes of the load in the power grid. The dynamic load characteristic data is analyzed to extract potential hidden variables, that is, higher-level information hidden behind the load characteristic data. The generated hidden variable time series data helps to understand the inherent laws and evolution trends of load behavior. The frequency decomposition module performs frequency decomposition processing on the hidden variable time series data, converts the time domain data into frequency domain data, and considers multiple frequency scales at the same time. The generated frequency band energy distribution data provides energy distribution in different frequency ranges. The frequency band energy distribution data is analyzed to extract frequency change characteristics, including frequency offset, frequency disturbance, etc., and the generated sub-frequency change characteristic data Reveal the change pattern and trend of load frequency. The frequency evolution trajectory module analyzes the frequency evolution trajectory of hidden variable time series data based on sub-frequency change feature data to reveal the evolution trajectory of load frequency. The generated load frequency evolution trajectory data helps to understand the change law and trend of load frequency. Based on the preset component load frequency evolution trajectory threshold and load frequency evolution trajectory data, the component dynamic topology detection network is traced and located to determine the propagation path and impact range of abnormal conditions in the power system. The generated abnormal load frequency evolution trajectory data helps to accurately locate the abnormal source and evaluate its impact on the system. The abnormal optimization decision module uses deep neural networks to make abnormal optimization decisions on abnormal diffusion path data and formulate optimization strategies for abnormal conditions. The generated power abnormal optimization decision data helps operators take targeted measures to solve power abnormal problems and improve the reliability and efficiency of the power system. Data-driven modeling is performed on power abnormal optimization decision data to build a power metering data abnormal decision model. Based on historical data and decision data, a prediction model is established using machine learning or statistical methods to perform abnormal analysis of power metering data. The model identifies potential abnormal patterns by learning the correlation between abnormal conditions and corresponding decision data, and provides abnormal analysis results and suggestions.

[0151] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0152] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for analyzing abnormality of measurement data, characterized in that: The following steps are involved: Step S1: Acquire power grid system component data and power grid power metering data; Performing component operation constraint analysis on power grid system component data to generate component operation constraint data; Based on the component operation constraint data, a dynamic topology network is constructed for the power grid system component data to generate a component dynamic topology detection network; Step S2; Performing dynamic load characteristic analysis on power grid power metering data based on a component dynamic topology detection network to generate dynamic load characteristic data; Perform latent variable characterization driven analysis on dynamic load characteristic data to generate latent variable time series data; Step S3: performing multi-scale frequency decomposition processing on the latent variable time series data to generate frequency band energy distribution data; Performing frequency variation characteristic analysis on the frequency band energy distribution data to generate sub-frequency variation characteristic data; Step S4: performing frequency evolution trajectory analysis on the latent variable time series data based on the sub-frequency variation characteristic data to generate load frequency evolution trajectory data; Based on the preset component load frequency evolution trajectory threshold, the load frequency evolution trajectory data is compared to generate abnormal load frequency evolution trajectory data; based on the abnormal load frequency evolution trajectory data, the component dynamic topology detection network is abnormally traced and located to generate load frequency abnormality spatial data; Step S5: performing abnormal situation evolution analysis on the load frequency abnormality spatial data to generate frequency abnormality trend data; Predict the abnormal trend change of the frequency abnormal trend data to obtain the abnormal diffusion path data; Step S6: using a deep neural network to perform abnormal optimization decision on the abnormal diffusion path data to generate power abnormal optimization decision data; Carry out data-driven modeling of power anomaly optimization decision data and build a power metering data anomaly decision model to perform power metering data anomaly analysis operations; The specific steps of step S4 are: Step S41: performing frequency evolution trajectory analysis on latent variable time series data based on sub-frequency variation characteristic data to generate load frequency evolution trajectory data; Step S42: comparing the preset element load frequency evolution trajectory threshold with the load frequency evolution trajectory data, when the preset element load frequency evolution trajectory threshold is less than or equal to the load frequency evolution trajectory data, it is determined that the power load is abnormal, and abnormal load frequency evolution trajectory data is generated; Step S43: when the preset component load frequency evolution trajectory threshold is greater than the load frequency evolution trajectory data, it is determined that the power load is in a normal state; Step S44: tracing and locating the abnormality of the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data; The specific steps of step S5 are: Step S51: performing principal component analysis on the load frequency abnormality spatial data to generate low-dimensional frequency abnormality feature data; Step S52: performing abnormal situation evolution analysis on the low-dimensional frequency abnormal feature data to generate frequency abnormal trend data; Step S53: predicting abnormal trend changes of the frequency abnormal trend data to generate load frequency abnormal prediction data; Step S54: performing quantitative diagnosis of abnormal diffusion path on the load frequency abnormality prediction data, thereby obtaining abnormal diffusion path data; The specific steps of step S6 are: Step S61: using a deep neural network to perform abnormal feature space learning processing on the abnormal diffusion path data to generate power abnormal feature data; Step S62: performing abnormal optimization decision on the power abnormality feature data to generate power abnormality optimization decision data; Step S63: Perform data-driven modeling on the power anomaly optimization decision data, and construct a power metering data anomaly decision model to perform an anomaly analysis operation on the power metering data.

2. The method according to claim 1, characterized in that The specific steps of step S1 are: Step S11: Acquire power grid system component data and power grid power metering data; Step S12: monitoring the operation status of the power grid system component data, thereby obtaining the power grid component operation data; Step S13: performing component operation constraint analysis on the power grid system component data to generate component operation constraint data; Step S14: performing topological structure distribution analysis on the power grid system component data based on the component operation constraint data to obtain component topological structure data; Step S15: constructing a dynamic topology network for the component topology structure data through the power grid component operation data to generate a component dynamic topology detection network.

3. The method according to claim 1, characterized in that The specific steps of step S2 are: Step S21: performing dynamic load characteristic analysis on power grid power metering data based on the component dynamic topology detection network to generate dynamic load characteristic data; Step S22: performing time series fluctuation analysis on the dynamic load characteristic data to generate time series load fluctuation data; Step S23: performing fluctuation curve fitting on the time series load fluctuation data to construct a load fluctuation curve; Step S24: performing latent variable characterization drive analysis on the load fluctuation curve to generate latent variable time series data.

4. The method according to claim 1, characterized in that: The specific steps of step S21 are: Step S211: Calculate the load power distribution of the power grid metering data to obtain the load power distribution data of the power grid; Step S212: Performing dynamic load harmonic analysis on the power grid metering data to generate dynamic complex harmonic data of the power grid; Step S213: Perform dynamic load characteristic analysis on the grid load power distribution data and the grid dynamic complex harmonic data to generate dynamic load characteristic data.

5. The method according to claim 1, characterized in that The specific steps of step S3 are: Step S31: performing multi-scale frequency decomposition processing on the latent variable time series data to generate a multi-level frequency signal domain; Step S32: performing energy intensity distribution analysis on the multi-level frequency signal domain to generate frequency segment energy distribution data; Step S33: performing same-frequency signal domain recognition on the frequency segment energy distribution data to extract the same-frequency signal domain; Step S34: reconstructing the scale frequency of the same frequency signal domain to generate a sub-frequency signal domain; Step S35: performing frequency variation characteristic analysis on the sub-frequency signal domain to generate sub-frequency variation characteristic data.

6. The method according to claim 5, characterized in that The specific steps of step S31 are: Step S311: performing frequency domain conversion on the latent variable time series data to generate dynamic load frequency domain data; Step S312: extracting low-frequency components from the dynamic load frequency domain data to generate low-frequency load frequency domain component data; Step S313: performing high-frequency component refinement processing on the dynamic load frequency domain data to generate high-frequency load frequency domain component data; Step S314: performing frequency domain signal fusion on the low-frequency load frequency domain component data and the high-frequency load frequency domain component data to generate a multi-level frequency signal domain.

7. A measurement data abnormality analysis system, characterized in that: The method for performing abnormal analysis of measurement data according to claim 1 comprises: A topology network module is used to obtain power grid system component data and power grid power metering data; perform component operation constraint analysis on the power grid system component data to generate component operation constraint data; and perform dynamic topology network construction on the power grid system component data based on the component operation constraint data to generate a component dynamic topology detection network; The load characteristic analysis module is used to perform dynamic load characteristic analysis on the power metering data of the power grid based on the component dynamic topology detection network to generate dynamic load characteristic data; perform latent variable characterization driving analysis on the dynamic load characteristic data to generate latent variable time series data; The frequency decomposition module is used to perform multi-scale frequency decomposition processing on the latent variable time series data to generate frequency band energy distribution data; perform frequency change characteristic analysis on the frequency band energy distribution data to generate sub-frequency change characteristic data; The frequency evolution trajectory module is used to perform frequency evolution trajectory analysis on the hidden variable time series data based on the sub-frequency change characteristic data to generate load frequency evolution trajectory data; compare the preset component load frequency evolution trajectory threshold with the load frequency evolution trajectory data to generate abnormal load frequency evolution trajectory data; and perform abnormal source tracing and positioning on the component dynamic topology detection network based on the abnormal load frequency evolution trajectory data to generate load frequency abnormality spatial data; The trend prediction module is used to analyze the abnormal situation evolution of the load frequency abnormality spatial data and generate frequency abnormality trend data; predict the abnormal trend change of the frequency abnormality trend data to obtain abnormal diffusion path data; The abnormal optimization decision module is used to use deep neural networks to make abnormal optimization decisions on abnormal diffusion path data to generate power abnormal optimization decision data; perform data-driven modeling on the power abnormal optimization decision data, and build a power metering data abnormal decision model to perform abnormal analysis operations on the power metering data.

Images