A knowledge graph-based method and system for processing comprehensive power grid evaluation indicators

Abstract

This invention discloses a method and system for processing comprehensive power grid evaluation indicators based on knowledge graphs. The method includes the following steps: classifying the power grid evaluation indicators using their influencing factors as a benchmark to obtain initialized power grid evaluation indicator data, and constructing a knowledge graph from this data; normalizing the knowledge graph to convert the initialized power grid evaluation indicator data into normalized power grid evaluation indicator data; performing fault-tolerant processing on the data indicators within the normalized knowledge graph; performing secure transcoding on the waveform data in the power grid evaluation indicators; performing data cleaning by filling in the data of each power grid evaluation indicator using a linear regression model; and performing data verification on the comprehensive power grid evaluation indicators. This invention combines multiple power grid evaluation indicators, provides two different cleaning methods, and employs fault-tolerant processing and secure transcoding, thereby improving overall robustness.

Description

Technical Field

[0001] This invention belongs to the field of knowledge graph data cleaning technology, specifically relating to a method and system for processing comprehensive power grid evaluation indicators based on knowledge graphs. Technical Background

[0002] The five performance characteristics of a regional power grid—abundance, coordination, reliability, security, and economy—are a comprehensive reflection of the level of power grid construction and operation control. The influence of factors such as the grid structure, geographical conditions, operation management, and equipment configuration of the regional power grid can be accurately controlled from these five performance characteristics. Therefore, a scientific, complete, and comprehensive indicator system should be formed for the comprehensive performance evaluation of a regional power grid.

[0003] A knowledge graph is a series of different graphs that display the development process and structural relationships of knowledge. It uses visualization technology to describe knowledge resources and their carriers, and to mine, analyze, construct, draw and display knowledge and the interrelationships between them. It solves the problem that traditional databases have insufficient information expression capabilities and cannot effectively support the back-end data service needs of most power systems.

[0004] Due to the large volume of data, wide range of data sources, and inconsistent data types in the comprehensive evaluation indicators of the power grid, and because the process of constructing knowledge graphs is often automated to avoid human supervision, various errors inevitably arise, including ambiguity, conflicts, erroneous information, and redundancy. Quality assurance of the knowledge graph is a prerequisite for various knowledge-driven applications. Therefore, designing a general and efficient knowledge graph-based method and system for processing comprehensive evaluation indicators of the power grid, enabling standardized processing, fault-tolerant processing, secure transcoding, indicator cleaning, and data verification of various comprehensive evaluation indicators within the knowledge graph architecture, and improving the stability, reliability, and robustness of the knowledge graph in the comprehensive evaluation system of the power grid, has become an urgent need for power companies. Summary of the Invention

[0005] This invention provides a method and system for processing power grid comprehensive evaluation indicators based on knowledge graphs, which facilitates staff learning and mastering the knowledge system of power grid comprehensive evaluation, enabling them to perceive the regional power grid operation status in all aspects, and combining traditional power grid comprehensive evaluation methods with knowledge graph technology, which has interconnectivity, openness, structure, and visualization of knowledge.

[0006] This invention provides a method for processing comprehensive power grid evaluation indicators based on knowledge graphs, comprising the following steps:

[0007] Using the influencing factors of various power grid evaluation indicators as a benchmark, the power grid evaluation indicators are classified to achieve differentiated processing of the power grid evaluation indicators, obtain initialized power grid evaluation indicator data, and construct a knowledge graph from it.

[0008] The knowledge graph is normalized to convert the initialized power grid evaluation index data into normalized power grid evaluation index data.

[0009] The standardized power grid evaluation index data is subjected to fault tolerance processing, and then missing values ​​are filled by regression using an imputation function;

[0010] The waveform data in the power grid evaluation index data after regression imputation is subjected to secure transcoding processing;

[0011] Data cleaning is performed on the power grid evaluation index data after the security transcoding process, and then the data cleaning results are verified until the training model and triplet classification model of the knowledge graph are fully converged. Then, data is filled in using a linear regression model.

[0012] Data verification was performed on the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph;

[0013] The knowledge graph of the comprehensive evaluation indicators of the power grid after the data verification is completed is stored in a graph database.

[0014] Furthermore, the power grid evaluation indicators include at least one of the following: power grid adequacy, coordination, reliability, security, and economy.

[0015] Furthermore, after the fault-tolerant processing is performed, the missing data is automatically backfilled using a filling function, thereby obtaining relatively complete data in the power grid comprehensive evaluation system.

[0016] Furthermore, the secure transcoding process includes:

[0017] The waveform data of the comprehensive evaluation index of the power grid are smoothed and filtered using Hilbert-Huang transform, and the correlation between waveform data is found using configuration files. The waveforms are then drawn in parallel to form a graphical waveform file.

[0018] Furthermore, the data cleaning and verification of the data cleaning results include:

[0019] Obtain an uncleaned knowledge graph to be cleaned, the knowledge graph to be cleaned includes entities composed of power grid comprehensive evaluation index data, and their corresponding entity attributes and relation attributes, wherein each group of entities, entity attributes and their relationships form a triple.

[0020] The power grid comprehensive evaluation index data to be processed is cleaned for the first time, and invalid and outlier values ​​are removed using the Schauville criterion;

[0021] The knowledge graph-based power grid comprehensive evaluation index processing system is equipped with a pre-set knowledge graph training model and a triplet classification model.

[0022] The system uses a pre-set knowledge graph training model and a triplet classification model for training and analysis, and then performs a second cleaning process to filter out and remove erroneous triplets.

[0023] Using a linear regression model, missing values ​​were imputed in the removed data.

[0024] The cleaned power grid comprehensive evaluation index is tested by double cumulative analysis. If the test result meets the predetermined value, the data cleaning can be ended; otherwise, the data is iterated and the first cleaning, second cleaning and missing value filling processes are repeated until the test result meets the predetermined value.

[0025] Furthermore, the first cleaning includes:

[0026] Using the Schauville criterion, all power grid evaluation indicators are used as samples to construct a dataset, and a probability band centered on the mean of the normal distribution is determined. Any sample data value that is not within the probability band will be judged as an outlier and removed from the dataset.

[0027] The formula for calculating the Schauville criterion is as follows:

[0028]

[0029] Among them, D max δ is the maximum set deviation value, x is the suspected outlier, μ is the sample mean, and δ is the sample standard deviation.

[0030] Furthermore, the second cleaning includes:

[0031] Input all triples into the Trans-E training model and use random negative sampling to train the noise-aware knowledge graph model.

[0032] The Trans-E model has a pre-set triplet scoring formula. Based on the triplet scores after training, all triplets are input into the triplet classification model for training. After training, the confidence of each triplet is refreshed, and the confidence of different triplets are different.

[0033] A preset confidence threshold is set: if the confidence of a triple is greater than the preset threshold, the triple is considered correct and retained; if the confidence of a triple is less than the preset threshold, the triple is considered incorrect and discarded.

[0034] The triplet classification model uses the Sigmoid function to constrain the classifier's output to 0-1.

[0035] The Trans-E model and the triplet classification model are combined and trained iteratively until both the knowledge graph training model and the triplet classification model converge completely.

[0036] Furthermore, the filling in using a linear regression model includes:

[0037] Convert the mathematical expression of the linear regression model into a vector expression:

[0038] The mathematical expression for the linear regression model is:

[0039]

[0040] The above mathematical expression can be transformed into the following vector expression:

[0041]

[0042] Where x represents a power grid evaluation index with missing values; y represents a power grid evaluation index without missing values; k represents the total number of power grid evaluation indices; n represents the number of power grid comprehensive evaluation performance categories; a k is the influence factor of each evaluation index on a certain power grid performance evaluation item; b is a constant term, which serves as the threshold for judging missing values ​​of power grid evaluation indicators.

[0043] Furthermore, the data verification of the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph includes the following steps:

[0044] Verify the accuracy of the data for each power grid evaluation indicator;

[0045] Check that the data format and type are correct;

[0046] The correlation between the different power grid evaluation indicators is used to determine whether the relationship attributes of each power grid evaluation indicator correspond.

[0047] Check for duplicate or redundant indicators under a specific power grid performance evaluation item;

[0048] Verify the standardization of data to prevent situations where the names of power grid evaluation indicators are not standardized, resulting in the inability to summarize relevant power grid evaluation indicators under a certain power grid performance assessment item in the knowledge graph.

[0049] This invention also provides a knowledge graph-based power grid comprehensive evaluation index processing system, which includes:

[0050] The knowledge graph construction module is used to classify the power grid evaluation indicators based on the influence factors of each power grid evaluation indicator, realize the differentiated processing of the power grid evaluation indicators, obtain the initialized power grid evaluation indicator data, and construct a knowledge graph from it.

[0051] The normalization processing module is used to normalize the knowledge graph and convert the initialized power grid evaluation index data into normalized power grid evaluation index data.

[0052] The fault-tolerant processing module is used to perform fault-tolerant processing on the standardized power grid evaluation index data, and then use the imputation function to perform regression imputation on the missing values;

[0053] The secure transcoding processing module is used to perform secure transcoding processing on the waveform data in the regression-filled power grid evaluation index data;

[0054] The data cleaning module is used to clean the power grid evaluation index data after the security transcoding process, and then verify the data cleaning results until the training model and triplet classification model of the knowledge graph are fully converged. Then, the data is filled in by the linear regression model.

[0055] The data verification module is used to verify the data of the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph;

[0056] The data storage module is used to store the knowledge graph of the power grid comprehensive evaluation indicators after the data verification is completed into a graph database.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] 1. This invention provides an effective and complete data cleaning process for the entity correlation of different power grid comprehensive evaluation indicators by using two different cleaning methods, missing value imputation, and cleaning result verification. This enables the constructed power grid comprehensive evaluation knowledge graph to more accurately and accurately reflect the relationship between different evaluation indicator data on different power grid performances. Furthermore, multi-level data verification operations are conducted to improve the construction efficiency of the power grid comprehensive evaluation knowledge graph, the authenticity of the evaluation indicator correlation, and the data traceability and reliability of the power grid comprehensive evaluation knowledge graph.

[0059] 2. By using the influencing factors of various power grid evaluation indicators as a benchmark to classify all power grid evaluation indicators, it is possible to differentiate the processing of power grid evaluation indicators. This avoids deviations in subsequent analysis and processing due to differences in the weight of different power grid evaluation indicators in power grid performance evaluation, and improves the pertinence and effectiveness of processing data of different power grid evaluation indicators. This reduces the complexity of subsequent fault-tolerant processing, secure transcoding, data cleaning and data verification steps, and improves the robustness of data when constructing the map.

[0060] 3. It can provide important technical support and research platform for the dispatching, planning and related decision-making of the regional power grid, and has high application value. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the structure of the knowledge graph ontology model of the power grid comprehensive evaluation system provided by the present invention;

[0062] Figure 2 A flowchart illustrating a knowledge graph-based method for processing comprehensive power grid performance evaluation indicators provided by this invention;

[0063] Figure 3 This is a flowchart illustrating a data cleaning method for a comprehensive power grid evaluation index provided by the present invention. Detailed Implementation

[0064] The following is in conjunction with the appendix Figures 1-3 The present invention will be further described in detail with reference to specific embodiments.

[0065] like Figure 1 As shown in the figure, this embodiment discloses a method for processing comprehensive power grid evaluation indicators based on knowledge graphs, including the following steps:

[0066] Using the influencing factors of various power grid evaluation indicators as a benchmark, the power grid evaluation indicators are classified to achieve differentiated processing of the power grid evaluation indicators, obtain initialized power grid evaluation indicator data, and construct a knowledge graph from it.

[0067] The knowledge graph is normalized to convert the initialized power grid evaluation index data into normalized power grid evaluation index data.

[0068] The standardized power grid evaluation index data is subjected to fault tolerance processing, and then missing values ​​are filled by regression using an imputation function;

[0069] The waveform data in the power grid evaluation index data after regression imputation is subjected to secure transcoding processing;

[0070] Data cleaning is performed on the power grid evaluation index data after the security transcoding process, and then the data cleaning results are verified until the training model and triplet classification model of the knowledge graph are fully converged. Then, data is filled in using a linear regression model.

[0071] Data verification was performed on the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph;

[0072] The knowledge graph of the comprehensive evaluation indicators of the power grid after the data verification is completed is stored in a graph database.

[0073] In this embodiment, the power grid evaluation indicators include the adequacy, coordination, reliability, security, and economy of the power grid. However, it can also adopt a form that includes at least one of these five power grid evaluation indicators. For example... Figure 1 As shown, by analyzing the characteristics of the digital model of the regional power grid system, and based on the knowledge graph, a comprehensive performance evaluation index system for the power grid is established in five aspects: sufficiency, coordination, reliability, security, and economy.

[0074] The adequacy assessment indicators consist of node voltage qualification rate, grid expansion margin, power supply capacity margin, high-loss transformer rate, and resource margin. The coordination assessment indicators consist of node voltage qualification rate, DG capacity grid connection rate, power supply capacity matching degree, line interconnection rate, DG penetration rate, and load balancing degree. The reliability assessment indicators consist of voltage fluctuation rate, system average outage frequency, user average outage frequency, and capacity-to-load ratio. The safety assessment indicators consist of capacity-to-load ratio, N-1 pass rate, post-accident overload risk, post-accident energy loss rate, and failure rate. The economic assessment indicators consist of failure rate, EV charging station electricity consumption ratio, investment payback period, and net present value.

[0075] The grid connection rate of distributed generation (DG) capacity is an important indicator for assessing the coordination of the regional power grid. DG represents distributed generation. The impact of the grid connection rate of DG capacity on the security assessment of the regional power grid is very small and can be ignored. Therefore, it can be considered that changes in the grid connection rate of DG capacity do not affect the assessment of the security performance of the regional power grid.

[0076] By using the influence factors of various power grid evaluation indicators as a benchmark to classify all power grid evaluation indicators, it is possible to differentiate the processing of power grid evaluation indicators. This avoids deviations in subsequent analysis and processing due to differences in the weighting of different power grid evaluation indicators in power grid performance assessment, and improves the pertinence and effectiveness of processing data of different power grid evaluation indicators. This reduces the complexity of subsequent processing steps such as fault tolerance processing, secure transcoding, data cleaning, and data verification, and improves the data robustness when constructing the map.

[0077] To eliminate the dimensional influence between different power grid evaluation indicators within the same power grid performance evaluation item, as well as the coupling effect caused by different power grid performance evaluation items referencing the same power grid evaluation indicator, it is necessary to standardize the raw data. After standardization, all power grid evaluation indicators are at the same order of magnitude, which facilitates subsequent comprehensive evaluation work.

[0078] In this embodiment, the normalization processing of the knowledge graph composed of initial power grid evaluation index data includes at least one of weight normalization, logical normalization, and relational attribute normalization, thereby converting the initial power grid evaluation index data into normalized power grid evaluation index data. The purpose of normalization is to eliminate the dimensional influence between different power grid evaluation indicators within the same power grid performance evaluation item, as well as the coupling effect caused by different power grid performance evaluation items referencing the same power grid evaluation indicator. Normalization is necessary to ensure that all power grid evaluation indicators are of the same magnitude after normalization, facilitating subsequent comprehensive evaluation. Preprocessing the indicators constituting the knowledge graph in the early stages avoids errors due to data loss, format incompatibility, etc., during graph construction.

[0079] In this embodiment, after the fault-tolerance processing, the missing data is automatically filled in using an imputation function, thereby obtaining relatively complete data in the power grid comprehensive evaluation system. The imputation function is the MICE algorithm. The MICE algorithm is a commonly used method in R language for regression imputation of missing values.

[0080] The specific implementation steps of the MICE algorithm include:

[0081] Enter commands in the command-line window of the system's built-in Python IDE to check for missing values ​​of various power grid comprehensive evaluation indicators:

[0082] data_full.isnull().sum().sum()

[0083] The percentage of missing values ​​is usually calculated by dividing the number of missing values ​​by the length of the data.

[0084] data_mv.isnull().sum() / len(data_mv)

[0085] In this embodiment, the regional power grid adequacy is missing approximately 16% of the data value, specifically, the power grid expansion margin in the regional power grid adequacy evaluation index is missing approximately 30% of the data value; the power supply capacity margin is missing approximately 9% of the data value; and the resource margin is missing approximately 16% of the data value.

[0086] The regional power grid coordination is missing approximately 15% of the data values. Specifically, among the regional power grid coordination evaluation indicators, the DG capacity grid connection rate is missing approximately 7% of the data values; the line interconnection rate is missing approximately 14% of the data values; the DG penetration rate is missing approximately 29% of the data values; and the load balance is missing approximately 31% of the data values.

[0087] The data on the reliability of the regional power grid and its various evaluation indicators are complete, and no missing values ​​were found in the MICE algorithm check.

[0088] The regional power grid security data is missing by approximately 7%, including approximately 17% for post-accident overload risk, approximately 19% for post-accident energy loss rate, and approximately 6% for failure rate among the regional power grid security evaluation indicators.

[0089] The data on the regional power grid economy is missing by approximately 17%, including approximately 6% missing data on the failure rate, approximately 15% missing data on the investment payback period, and approximately 21% missing data on the net present value.

[0090] The MICE algorithm is implemented by calling the system's built-in Python IDE and the third-party plugin fancyimpute library. In fancyimpute, the MICE algorithm is named laterativelmputer. The metric filling operation is performed by entering commands in the command-line window of the system's built-in Python IDE. The specific MICE filling operation is as follows:

[0091] Comprehensive_evaluation_index_imputed=data_mv.copy()

[0092] # Copy the data to obesity_mice_imputed

[0093] mice_imputer=IterativeImputer()

[0094] #Initialize TerativeImputer

[0095] Comprehensive_evaluation_index_imputed.iloc[:,:]=mice_imputer.fit_transform(Comprehensive_evaluation_index_mice_imputed)

[0096] #Use fit_tranform to fill in the data

[0097] To check for missing values ​​after imputation, the specific command is:

[0098] Comprehensive_evaluation_index_mice_imputed.isnull().sum()

[0099] The percentage calculation results after filling in the missing values ​​of various power grid comprehensive evaluation indicators using the MICE algorithm show that all missing values ​​were successfully filled.

[0100] In this embodiment, the secure transcoding of waveform data in power grid evaluation indicators includes: securely transcoding waveform data obtained by power grid personnel during relevant data collection or experiments. Specifically, the waveform data of high-loss distribution transformer rate, line interconnection rate, distributed generation (DG) penetration rate, system average outage frequency, and user average outage frequency in the comprehensive power grid evaluation indicators are smoothed and filtered using Hilbert-Huang transform to minimize noise interference. Furthermore, the correlation between waveform data is found using configuration files, and waveforms are drawn in parallel to form a graphical waveform file that can be directly used by power grid personnel, thus achieving secure transcoding of waveform data. Data types in power grid evaluation indicators also include common data types such as integers, characters, and floating-point numbers.

[0101] In this embodiment, as Figure 3 As shown, the data cleaning and verification results include:

[0102] Obtain an uncleaned knowledge graph to be cleaned, the knowledge graph to be cleaned includes entities composed of power grid comprehensive evaluation index data, and their corresponding entity attributes and relation attributes, wherein each group of entities, entity attributes and their relationships form a triple.

[0103] The power grid comprehensive evaluation index data to be processed is cleaned for the first time, and invalid and outlier values ​​are removed using the Schauville criterion;

[0104] The knowledge graph-based power grid comprehensive evaluation index processing system is equipped with a pre-set knowledge graph training model and a triplet classification model.

[0105] The system uses a pre-set knowledge graph training model and a triplet classification model for training and analysis, and then performs a second cleaning process to filter out and remove erroneous triplets.

[0106] Using a linear regression model, missing values ​​were imputed in the removed data.

[0107] The cleaned power grid comprehensive evaluation index is tested by double cumulative analysis. If the test result meets the predetermined value, the data cleaning can be ended; otherwise, the data is iterated and the first cleaning, second cleaning and missing value filling processes are repeated until the test result meets the predetermined value.

[0108] In this embodiment, the first cleaning includes: using the Schauville criterion to process outlier data and removing data that is determined to be outlier.

[0109] Based on the causal relationship between different power grid evaluation indicators and the five aspects of power grid performance, we eliminate those evaluation indicators that, without any artificial settings, establish a causal relationship between a certain aspect of power grid performance and a certain evaluation indicator, and whose influence factor on the power grid performance evaluation of that aspect is less than a preset threshold.

[0110] The knowledge graph-based power grid comprehensive evaluation index processing system incorporates the Chauvenet criterion calculation model and utilizes the Chauvenet criterion to handle outlier data. By calculating and analyzing the influence factors of each evaluation index on a specific power grid performance evaluation item, the weight of that index in that item is determined based on the influence factors. Unreasonable, redundant, or ineffective evaluation indicators are then eliminated.

[0111] The Chauville criterion works as follows: First, a dataset is constructed from all the power grid evaluation indicators that constitute the five performance metrics of the regional power grid. Then, a probability band centered on the mean of a normal distribution is defined, encompassing all k indicators of the comprehensive power grid evaluation. In this embodiment, the power grid evaluation indicators are defined as samples, and k is set to 21. After the above settings, any sample data value not falling within the probability band will be judged as an outlier and removed from the dataset.

[0112] The outlier is determined by finding the value that is the same as the average value D. max The number of standard deviations corresponding to the boundaries of the surrounding probability bands is calculated, and this value is compared with the absolute value of the difference between the suspected outlier and the mean. If it is greater than the set maximum deviation value, it can be judged as an outlier.

[0113] The formula for calculating the Chauville criterion is as follows:

[0114]

[0115] Among them, D max δ is the maximum set deviation value, x is the suspected outlier, μ is the sample mean, and δ is the sample standard deviation.

[0116] The second cleaning process involves training and analyzing a pre-set knowledge graph training model and a triplet classification model to filter out and remove erroneous triplets.

[0117] The power grid comprehensive evaluation knowledge graph includes entities composed of the five aspects of sufficiency, coordination, reliability, security, and economy mentioned above, as well as their corresponding entity attributes and relationship attributes. The power grid comprehensive evaluation index processing system forms a triplet for each corresponding group of entities, entity attributes, and their relationships, and initializes the confidence of all triplets in the knowledge graph.

[0118] The knowledge graph-based power grid comprehensive evaluation index processing system incorporates a pre-set knowledge graph training model and a triplet classification model. The pre-set training model, TransE, is integrated and packaged; users can download the relevant plugins and configure the calling commands in the backend to use it.

[0119] The knowledge graph training model used is the Trans-E model, which is a pre-built, plug-and-play model that can be trained and tuned to its optimal parameters for different datasets. All triples are input into the Trans-E model, and a noise-aware knowledge graph model is trained using random negative sampling.

[0120] In one alternative implementation, a Trans-E model is used to map each entity and relation in the knowledge graph into a vector, and confidence is added to the loss function of the Trans-E model to reduce the impact of noisy data on the embedded vector.

[0121] The Trans-E model has a pre-defined triplet scoring formula. Based on the triplet scores after training, all triplets are input into the triplet classification model for training. After training, the confidence of each triplet is refreshed, and the confidence of different triplets are different.

[0122] The triplet classification model uses the Sigmoid function to constrain the classifier's output to 0-1, meaning the confidence level of each triplet is at most 1 and at least 0. The Sigmoid function, also called the Logistic function, is used for the output of hidden layer neurons and has a value range of (0,1). It can map a real number to the interval (0,1) and can be used for binary classification.

[0123] In one optional implementation, the Trans-E model and the triplet classification model can be combined for iterative training until both the knowledge graph training model and the triplet classification model have fully converged.

[0124] The number of iterations can be set according to the actual situation.

[0125] In one optional implementation, according to the results of multiple parameter tuning experiments, setting the number of training iterations to 6 times achieves the best cleaning effect; when the number of cleaning iterations is less than 6 times, the cleaning effect is poor and a small part of the "dirty" data is not cleaned out; when the number of cleaning iterations is more than 6 times, the cleaning effect is overfitted and a small part of the clean data will be accidentally removed.

[0126] In one optional implementation, the preset confidence threshold is 0.75. That is, when the confidence of a triple is greater than or equal to 0.75, the triple is determined to be correct and is retained; when the confidence of a triple is less than 0.75, the triple is determined to be incorrect and is discarded.

[0127] Once the data is cleaned, clean data can be extracted, which can improve the data quality of the power grid comprehensive evaluation knowledge graph system and avoid repeated cleaning work due to unqualified data verification in the future.

[0128] The aforementioned data cleaning method combines the characteristics of the power grid comprehensive evaluation indicators with the advantages of machine learning in data cleaning, and utilizes the Python language to write the relevant algorithms. By maintaining the corresponding feature vector structure in the linear regression model and removing qualitative data labels, the data cleaning task can be completed efficiently and conveniently, effectively reducing data variability.

[0129] Therefore, the multivariate secondary cleaning and subsequent filling and verification of power grid comprehensive evaluation indicators are used to eliminate outdated or invalid useless evaluation indicators, retain core data and backbone network topology, and provide a reasonable and consistent data foundation for the further construction of a knowledge graph for the power grid comprehensive evaluation system.

[0130] In this embodiment, the filling using a linear regression model includes:

[0131] Convert the mathematical expression of the linear regression model into a vector expression:

[0132] The mathematical expression for the linear regression model is:

[0133]

[0134] Where k is the total number of evaluation indicators in the power grid comprehensive evaluation knowledge graph system, and x = (x1, x2, ..., x...). k ) T Let be a k-dimensional random variable, existing in the form of a column vector; b is a constant term, serving as the threshold for judging missing values ​​of power grid evaluation indicators.

[0135] The missing data imputation method based on linear regression models involves using variables with missing values ​​as the prediction target, other variables or subsets of the dataset as input variables, constructing a training set using the non-missing values ​​of the variables, training the regression model, and using the constructed linear regression model to predict the missing values ​​of the corresponding variables.

[0136] In one optional implementation, the complete triples in the given power grid comprehensive evaluation knowledge graph system can be set as dataset S:

[0137]

[0138] To facilitate the use of a linear regression model for missing value imputation within the knowledge graph-based power grid comprehensive evaluation index processing system, the mathematical expression of the linear regression model is transformed into a vector expression that is easy to program and construct.

[0139] The vector expression for the linear regression model is:

[0140]

[0141] In the above formula, let the power grid evaluation index with missing values ​​be the dependent variable x; the remaining power grid evaluation indexes be the multidimensional independent variables y; k is the total number of evaluation indicators in the power grid comprehensive evaluation knowledge graph system, which is set to k=21 in this system, namely the 21 power grid evaluation indicators such as node voltage qualification rate, power grid expansion margin, and power supply capacity margin mentioned above; n is the number of power grid comprehensive evaluation performance categories, which is set to n=5 in this system, indicating that there are 5 categories of power grid comprehensive evaluation performance, namely the five aspects of regional power grid sufficiency, coordination, reliability, security, and economy; a is the influence factor of each evaluation index in a certain power grid performance evaluation item; b is a constant term, which serves as the threshold for judging missing values ​​of power grid evaluation indicators.

[0142] In the process of training the linear regression model, the complete data in the power grid comprehensive evaluation knowledge graph system is used as the training set, and the power grid evaluation indicators with missing values ​​are used as the test set. The missing values ​​are the dependent variable x to be predicted.

[0143] In this embodiment, the data verification of the power grid comprehensive evaluation index refers to the data verification of the regional power grid comprehensive evaluation index involved in the knowledge graph construction based on the created entities, entity attributes, and corresponding relationship attributes.

[0144] The data verification methods for the aforementioned comprehensive power grid evaluation indicators include the following five items:

[0145] First, verify whether the data volume of each power grid evaluation indicator is correct, that is, the completeness of the data;

[0146] Check that the data format and type are correct;

[0147] The correlation between the five comprehensive power grid performance indicators is used to determine whether the relationship attributes of each power grid evaluation indicator correspond.

[0148] Check for duplicate or redundant indicators under a specific power grid performance evaluation item;

[0149] Verify the standardization of data to prevent situations where the names of power grid evaluation indicators are not standardized, resulting in the inability to summarize relevant power grid evaluation indicators under a certain power grid performance assessment item in the knowledge graph.

[0150] In this embodiment, storing the knowledge graph in the graph database includes: submitting the power grid comprehensive evaluation system knowledge graph and its corresponding comprehensive evaluation indicators after completing the above series of processing operations to the Neo4j graph database through the py2neo library of Python, which can realize the long-term preservation and dynamic updating of the power grid comprehensive evaluation system knowledge graph.

[0151] This embodiment also provides a power grid comprehensive evaluation index processing system based on knowledge graphs, which includes:

[0152] The knowledge graph construction module is used to classify the power grid evaluation indicators based on the influence factors of each power grid evaluation indicator, realize the differentiated processing of the power grid evaluation indicators, obtain the initialized power grid evaluation indicator data, and construct a knowledge graph from it.

[0153] The normalization processing module is used to normalize the knowledge graph and convert the initialized power grid evaluation index data into normalized power grid evaluation index data.

[0154] The fault-tolerant processing module is used to perform fault-tolerant processing on the standardized power grid evaluation index data, and then use the imputation function to perform regression imputation on the missing values;

[0155] The secure transcoding processing module is used to perform secure transcoding processing on the waveform data in the regression-filled power grid evaluation index data;

[0156] The data cleaning module is used to clean the power grid evaluation index data after the security transcoding process, and then verify the data cleaning results until the training model and triplet classification model of the knowledge graph are fully converged. Then, the data is filled in by the linear regression model.

[0157] The data verification module is used to verify the data of the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph;

[0158] The data storage module is used to store the knowledge graph of the power grid comprehensive evaluation indicators after the data verification is completed into a graph database.

[0159] In summary, this invention provides a comprehensive power grid performance evaluation index system that considers the sufficiency, coordination, reliability, security, and economy of regional power grids. This system is applied to actual power grid operations and uses a knowledge graph to visually represent the entire evaluation system.

[0160] This invention provides a method and system for processing comprehensive power grid evaluation indicators based on knowledge graphs. The method and system can perform targeted standardization, fault tolerance, and secure transcoding preprocessing operations on the comprehensive power grid evaluation indicators themselves, thereby improving the accuracy and reliability of various comprehensive power grid evaluation indicators.

[0161] This invention has strong practicality and feasibility, and can provide important technical support and research platform for the scheduling, planning and related decision-making of the regional power grid, and has high value for promotion and application.

[0162] The above-described invention merely illustrates implementation methods of the present invention and should not be construed as limiting the scope of the invention patent, nor as imposing any form of limitation on the structure of the embodiments of the present invention. It should be noted that those skilled in the art can make various changes and improvements without departing from the concept of the embodiments of the present invention, and these all fall within the protection scope of the embodiments of the present invention.

Claims

1. A method for processing comprehensive power grid evaluation indicators based on knowledge graphs, characterized in that, Includes the following steps: Using the influencing factors of various power grid evaluation indicators as a benchmark, the power grid evaluation indicators are classified to achieve differentiated processing of the power grid evaluation indicators, obtain initialized power grid evaluation indicator data, and construct a knowledge graph from it. The knowledge graph is normalized to convert the initialized power grid evaluation index data into normalized power grid evaluation index data. The standardized power grid evaluation index data is subjected to fault tolerance processing, and then missing values ​​are filled by regression using an imputation function; The waveform data in the power grid evaluation index data after regression imputation is subjected to secure transcoding processing; Data cleaning is performed on the power grid evaluation index data after the security transcoding process, and then the data cleaning results are verified until the training model and triplet classification model of the knowledge graph are fully converged. Then, data is filled in using a linear regression model. Data verification was performed on the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph; The knowledge graph of the power grid comprehensive evaluation indicators after the data verification is completed is stored in a graph database; The data cleaning and verification results include: Obtain an uncleaned knowledge graph to be cleaned, the knowledge graph to be cleaned includes entities composed of power grid comprehensive evaluation index data, and their corresponding entity attributes and relation attributes, wherein each group of entities, entity attributes and their relationships form a triple. The power grid comprehensive evaluation index data to be processed is cleaned for the first time, and invalid and outlier values ​​are removed using the Schauville criterion; The knowledge graph-based power grid comprehensive evaluation index processing system is equipped with a pre-set knowledge graph training model and a triplet classification model. The system uses a pre-set knowledge graph training model and a triplet classification model for training and analysis, and then performs a second cleaning process to filter out and remove erroneous triplets. Using a linear regression model, missing values ​​were imputed in the removed data. The cleaned power grid comprehensive evaluation index is tested by double cumulative analysis. If the test result meets the predetermined value, the data cleaning can be ended; otherwise, the data is iterated and the first cleaning, second cleaning and missing value filling processes are repeated until the test result meets the predetermined value. The first cleaning includes: Using the Chauville criterion, all power grid evaluation indicators are used as samples to construct a dataset, and a probability band centered on the mean of the normal distribution is determined. Any sample data value that is not within the probability band will be judged as an outlier and removed from the dataset. The formula for calculating the Schauville criterion is as follows: in, Here, x represents the maximum set deviation value, μ represents the suspected outlier, and μ is the sample mean. The standard deviation is the sample standard deviation. The second cleaning includes: Input all triples into the Trans-E training model and use random negative sampling to train the noise-aware knowledge graph model. The Trans-E model has a pre-set triplet scoring formula. Based on the triplet scores after training, all triplets are input into the triplet classification model for training. After training, the confidence of each triplet is refreshed, and the confidence of different triplets are different. A preset confidence threshold is set: if the confidence of a triple is greater than the preset threshold, the triple is considered correct and retained; if the confidence of a triple is less than the preset threshold, the triple is considered incorrect and discarded. The triplet classification model uses the Sigmoid function to constrain the classifier's output to 0-1. The Trans-E model and the triplet classification model are combined and trained iteratively until both the knowledge graph training model and the triplet classification model converge completely.

2. The method for processing power grid comprehensive evaluation indicators based on knowledge graphs according to claim 1, characterized in that: The power grid evaluation indicators include at least one of the following: sufficiency, coordination, reliability, security, and economy.

3. The method for processing power grid comprehensive evaluation indicators based on knowledge graphs according to claim 1, characterized in that: After the fault-tolerant processing is performed, the missing data is automatically filled in using a filling function, thereby obtaining relatively complete data in the power grid comprehensive evaluation system.

4. The method for processing comprehensive power grid evaluation indicators based on knowledge graphs according to claim 1, characterized in that, The secure transcoding process includes: The waveform data of the comprehensive evaluation index of the power grid are smoothed and filtered using Hilbert-Huang transform, and the correlation between waveform data is found using configuration files. The waveforms are then drawn in parallel to form a graphical waveform file.

5. The method for processing power grid comprehensive evaluation indicators based on knowledge graphs according to claim 1, characterized in that, The filling through a linear regression model includes: Convert the mathematical expression of the linear regression model into a vector expression: The mathematical expression for the linear regression model is: The mathematical expression is transformed into the following vector expression: Where x represents a power grid evaluation index with missing values; y represents a power grid evaluation index without missing values; k represents the total number of power grid evaluation indices; n represents the number of power grid comprehensive evaluation performance categories; a k is the influence factor of each evaluation index on a certain power grid performance evaluation item; b is a constant term, which serves as the threshold for judging missing values ​​of power grid evaluation indicators.

6. The method for processing comprehensive power grid evaluation indicators based on knowledge graphs according to claim 1, characterized in that, The data verification of the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph includes the following steps: Verify the accuracy of the data for each power grid evaluation indicator; Check that the data format and type are correct; The correlation between the different power grid evaluation indicators is used to determine whether the relationship attributes of each power grid evaluation indicator correspond. Check for duplicate or redundant indicators under a specific power grid performance evaluation item; Verify the standardization of the data.

7. A system for processing power grid comprehensive evaluation indexes based on knowledge graphs as described in any one of claims 1 to 6, characterized in that, It includes: The knowledge graph construction module is used to classify the power grid evaluation indicators based on the influence factors of each power grid evaluation indicator, realize the differentiated processing of the power grid evaluation indicators, obtain the initialized power grid evaluation indicator data, and construct a knowledge graph from it. The normalization processing module is used to normalize the knowledge graph and convert the initialized power grid evaluation index data into normalized power grid evaluation index data. The fault-tolerant processing module is used to perform fault-tolerant processing on the standardized power grid evaluation index data, and then use the imputation function to perform regression imputation on the missing values; The secure transcoding processing module is used to perform secure transcoding processing on the waveform data in the regression-filled power grid evaluation index data; The data cleaning module performs data cleaning on the various power grid evaluation index data that have undergone the security transcoding process, and then verifies the data cleaning results until the training model and triplet classification model of the knowledge graph are fully converged. Finally, the data is filled in using a linear regression model. The data verification module is used to verify the data of the regional power grid comprehensive evaluation indicators involved in the construction of the knowledge graph; The data storage module is used to store the knowledge graph of the power grid comprehensive evaluation indicators after the data verification is completed into a graph database.

Images