Method and system for extracting security and stability control rules based on gmm and decision tree

Abstract

The application discloses a kind of based on GMM and decision tree's safe and stable control rule extraction method and system, the method includes: obtaining short-circuit current historical calculation result and its corresponding time mark and time mark corresponding mode data, constructs bus short-circuit current control rule extraction sample;Decision tree is constructed, if all operating modes in sample are same, then generate leaf node;Gaussian mixture model is used to classify sample;If the difference of the mean of short-circuit current of different class sample is less than a constant A, then generate leaf node;If the number of a certain class sample is less than a constant B, then generate leaf node;Based on attribute division index, select optimal division attribute;Select each value of optimal division attribute, constitute sample subset and return to re-sample classification, update sample and mode data;Based on the generated decision tree, generate bus short-circuit current control rule.The application can make full use of short-circuit current information to realize the fast, reliable extraction of short-circuit current control rule.

Description

Technical Field

[0001] This invention relates to power system automation technology, and more particularly to a method and system for extracting safety and stability control rules based on GMM and decision tree. Background Technology

[0002] Large-scale centralized renewable energy integration has a direct impact on short-circuit current levels, while massive distributed renewable energy integration has altered the power grid's source-load distribution, posing a severe challenge to traditional short-circuit current control strategies. Methods based on physical models to mine short-circuit current control rules face challenges such as the large volume and dispersed distribution of renewable energy unit data, and the small impact of individual renewable energy units, making it difficult to uncover short-circuit current control rules for novel power systems.

[0003] Chinese patent application number 202110428472.7 discloses a short-circuit current prevention and control decision-making method considering safety and stability limit constraints. It calculates the control effect of candidate short-circuit current control measures on the short-circuit current of key circuit breakers, and selects the optimal short-circuit current prevention and control measure based on this. However, the candidate measures include a large number of renewable energy measures, leading to long calculation times. Chinese patent application number 202110441474.X discloses a transmission network structure optimization method for adjusting outgoing line schemes in 500 kV substations. For the initially screened outgoing line adjustment schemes, it needs to calculate the short-circuit current level under different outgoing line adjustment schemes. However, it fails to specify whether the evaluated short-circuit current level is for all circuit breakers in the near area or relies on expert experience for screening, posing a risk of calculation redundancy or omission. Chinese patent application number 201911022581.8 discloses an intelligent auxiliary decision-making method and system for limiting short-circuit current operation. It uses machine learning to obtain predictive models of auxiliary decision-making control measures corresponding to different current-limiting components and short-circuit control measures. During the control process, the corresponding predictive model is selected based on the monitored current-limiting components and the short-circuit control measures already implemented, thereby quickly obtaining auxiliary decision-making control measures. The machine learning prediction results are simulated and verified; if the prediction results are inapplicable, conventional auxiliary decision-making methods are used. However, situations where machine learning predictions are inapplicable can be known in advance through short-circuit current rule mining. If there are no relevant rules in the historical data, the adaptability of the machine learning prediction results is random, and the reliability is reduced.

[0004] The aforementioned short-circuit current control strategy based on "online computation-assisted decision-making + human experience judgment" is no longer sufficient to meet the short-circuit current level requirements of new power systems due to its long computation time, the risk of computational redundancy or omissions, or low reliability of prediction results. However, the massive historical calculation results of short-circuit current contain a large amount of short-circuit current control rule information, and how to extract short-circuit current control rules from this information is an urgent problem to be solved. Summary of the Invention

[0005] Purpose of the invention: The present invention aims to provide a method and system for extracting safe and stable control rules based on GMM and decision tree, which can make full use of information such as short-circuit current adjustment and prevention control to improve decision-making efficiency and reliability, and realize the rapid and reliable extraction of short-circuit current control rules.

[0006] Technical solution: The present invention provides a method for extracting safe and stable control rules based on GMM and decision tree, comprising the following steps:

[0007] (1) Obtain the historical calculation results of short-circuit current and its corresponding time scale, as well as the mode data S corresponding to the time scale of the short-circuit current calculation results, and perform data preprocessing to construct the bus short-circuit current control rule extraction sample set D;

[0008] (2) Construct a decision tree. If all samples in the sample set operate in the same way, then generate leaf nodes.

[0009] (3) Based on the distribution of short-circuit current in the busbar, the Gaussian mixture model is used to classify the samples;

[0010] (4) If the difference between the mean short-circuit currents of different types of samples is less than a fixed value A, then a leaf node is generated.

[0011] (5) If the number of samples of a certain type is less than a fixed value B, then a leaf node is generated;

[0012] (6) Calculate the attribute partitioning index for all operating modes S, and select the optimal partitioning attribute s. * ;

[0013] (7) Select the optimal partitioning attribute s * For each value of , its corresponding sample constitutes a sample subset; return to step (3) and update the sample and mode data;

[0014] (8) Generate bus short-circuit current control rules based on the generated decision tree.

[0015] Preferably, the mode data in step (1) includes the start-up and shutdown status of generator units and the commissioning status of lines in the entire grid, which are obtained from the historical data of grid operation mode and correspond to the time scale of all historical calculation results of short-circuit current.

[0016] Preferably, the data preprocessing in step (1) includes merging the bus short-circuit current calculation results with the start-up and shutdown status and line commissioning status data of the whole network generator units with the same time scale. If a certain time scale only has the bus short-circuit current calculation results or the start-up and shutdown status and line commissioning status data of the whole network generator units, then the data of that time scale is removed.

[0017] Preferably, the Gaussian mixture model used in step (3) to classify samples includes:

[0018] (3.1) Searching for the optimal Gaussian Mixture Model (GMM) based on the Bayesian information criterion k ;

[0019] (3.2) Based on the optimal Gaussian mixture model (GMM) k The historical minimum value of the short-circuit current at bus k is Initially, with a step size of λ, calculate the cumulative distribution function (CDF) of the short-circuit current at bus k. k ;

[0020] (3.3) If there exists a CDF with consecutive n steps k If all values ​​remain unchanged, then the short-circuit current control rule is extracted, and the average short-circuit current corresponding to n consecutive step sizes is denoted as . Otherwise, a leaf node is generated;

[0021] (3.4) The short-circuit current in the sample is greater than The sample with the higher short-circuit current is denoted as sample D. high The short-circuit current in the sample is less than The sample with the lower short-circuit current is denoted as sample D. high .

[0022] Preferably, the calculation method for the attribute division index of the operating mode in step (6) includes:

[0023] The product of information gain ratio and short-circuit current sensitivity is used as the attribute classification index, and the calculation formula is as follows:

[0024]

[0025] In the formula, F a The dividing criterion for attribute a is... The sensitivity of attribute a to the short-circuit current at bus k. Information gain rate.

[0026] Preferably, in step (7), a subset of samples is used as the sample, S\{s *} is the mode data, return to step (3), update the sample and mode data.

[0027] Preferably, the method for generating the bus short-circuit current control rule in step (8) is as follows:

[0028] (8.1) Each leaf node corresponds to one control rule;

[0029] (8.2) The optimal attribute of the node corresponding to the leaf node is s. m ;

[0030] (8.3) If s mWhen the system is shut down, the short-circuit current decreases by the following amount:

[0031] The present invention discloses a safety and stability control rule extraction system based on GMM and decision tree, comprising:

[0032] The data acquisition and preprocessing module is used to acquire historical calculation results of short-circuit current and their corresponding time scales, as well as the mode data corresponding to the time scales of short-circuit current calculation results, and to perform data preprocessing to construct a sample set for extracting bus short-circuit current control rules.

[0033] The decision tree construction module is used to build decision trees. If all samples in the sample have the same operation mode, a leaf node is generated. According to the distribution of bus short-circuit current, a Gaussian mixture model is used to classify the samples. If the difference between the mean short-circuit current of different classes of samples is less than a fixed value A, a leaf node is generated. If the number of samples of a certain class is less than a fixed value B, a leaf node is generated.

[0034] The decision tree update module is used to calculate the attribute partitioning index for all operating modes, select the optimal partitioning attribute, and construct a subsample set for each value of the optimal partitioning attribute to update the sample and mode data.

[0035] The short-circuit current control rule generation module is used to generate bus short-circuit current control rules based on the generated decision tree.

[0036] The present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the aforementioned method for extracting security and stability control rules based on GMM and decision tree.

[0037] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the above-described method for extracting security and stability control rules based on GMM and decision tree.

[0038] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: It integrates big data technology and physical mechanisms, makes full use of information such as the adjustment and prevention control of short-circuit current to improve decision-making efficiency and reliability, judges whether there are short-circuit current control rules in massive historical operation mode data, screens out effective short-circuit current control measures, realizes the rapid and reliable extraction of short-circuit current control rules, supports the operation mode adjustment and prevention control decision of power grid regulation, and ensures the safe and stable operation of the new power system. Attached Figure Description

[0039] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0040] like Figure 1 As shown, the present invention provides a method for extracting safe and stable control rules based on GMM and decision trees, which includes the following steps:

[0041] (1) Obtain the historical results of the online short-circuit analysis function, and extract the time scale and the corresponding short-circuit current of each bus in the historical results;

[0042] (2) Obtain the start-up and shutdown status of all generating units and the line commissioning status of the entire network corresponding to the time scale of the historical calculation results of all bus short-circuit currents from the historical data of power grid operation modes, and denot it as operation mode S=[s1,s2,…,s…]. n In the formula, s1~s n This represents the operating status of devices 1 to n, i.e., the mode data;

[0043] (3) After merging the bus short-circuit current calculation results with the start-up and shutdown status and line commissioning status data of the whole network generator units, the data is removed. The data removal means that if a certain time point only has the bus short-circuit current calculation results or the start-up and shutdown status and line commissioning status data of the whole network generator units, the data of that time point is removed.

[0044] (4) Based on the busbar list order, extract sample D to form busbar short-circuit current control rules. Sample D refers to the data after merging and data removal, arranged according to the busbar list order. Begin constructing the decision tree, denoted as bus k, where the historical maximum short-circuit current value of bus k is... The historical minimum value is

[0045] (5) Generate a node;

[0046] (6) If all samples operate in the same way, then mark the node as a leaf node and the attribute is the mean of all bus short-circuit currents in the sample.

[0047] (7) Based on the short-circuit current distribution of bus k, the Gaussian mixture model is used to classify the samples. The type with higher short-circuit current is denoted as 1, and the type with lower short-circuit current is denoted as 0; the sample with higher short-circuit current is denoted as D. high The sample with the lower short-circuit current is D. low ;

[0048] The Gaussian mixture model is used to classify samples in the following ways:

[0049] (7.1) Searching for the optimal Gaussian Mixture Model (GMM) based on the Bayesian information criterion k ;

[0050] (7.2) Based on the optimal Gaussian mixture model (GMM)k ,from Initially, with a step size of λ, calculate the cumulative distribution function (CDF) of the short-circuit current at bus k. k ;

[0051] (7.3) If there exists a CDF with consecutive n steps k If all values ​​remain unchanged, it indicates that an accurate and effective short-circuit current control rule can be extracted. Let the average short-circuit current corresponding to n consecutive step sizes be denoted as . Otherwise, mark the node as a leaf node, with the attribute being the mean short-circuit current of the corresponding sample;

[0052] (7.4) Update the short-circuit current to be less than The sample type is 0, and the short-circuit current is greater than 0. The sample type is 1, update the sample D with higher short-circuit current. high And sample D with lower short-circuit current low .

[0053] (8) Calculate the mean short-circuit current for the two types of samples respectively. If |μ high -μ low If |≤1, then the node is marked as a leaf node, and its attribute is the mean short-circuit current of the corresponding sample; where μ high and μ low These are the sample mean values ​​for high short-circuit current and low short-circuit current, respectively.

[0054] (9) Count the number of samples in the two classes. If N high ≤N min or N low ≤N min Then, the node is marked as a leaf node, and its attribute is the mean short-circuit current of the corresponding sample; where N high and N low N represents the number of samples with higher short-circuit current and the number of samples with lower short-circuit current, respectively. min This is a specified value; that is, if the number of samples in the sample subset is less than a certain value, the search rules will no longer be applied.

[0055] (10) Calculate the attribute partitioning index for all operating modes S, and select the optimal partitioning attribute s. * ;

[0056] The calculation method for the attribute segmentation index includes:

[0057] The product of information gain ratio and short-circuit current sensitivity is used as the attribute classification index, and the calculation formula is as follows:

[0058]

[0059] In the formula, F aThe dividing criterion for attribute a is... The sensitivity of attribute a to the short-circuit current at bus k. This refers to the information gain ratio.

[0060] (11) Select the optimal partitioning attribute s * Each value The corresponding samples constitute a sample subset D v ;

[0061] (12) D v Given the sample and S\{s*} as the mode data, proceed to step (7) to update the sample and mode data;

[0062] (13) Based on the generated decision tree T, generate the short-circuit current control rule R for bus k;

[0063] The method for generating the short-circuit current control rule R is as follows:

[0064] (13.1) Each leaf node corresponds to one control rule;

[0065] (13.2) The optimal attribute of the node corresponding to the leaf node is s. m ;

[0066] (13.3) If s m When the system is shut down, the short-circuit current decreases by the following amount:

[0067] The present invention discloses a safety and stability control rule extraction system based on GMM and decision tree, comprising:

[0068] The data acquisition and preprocessing module is used to acquire historical calculation results of short-circuit current and their corresponding time scales, as well as the mode data corresponding to the time scales of short-circuit current calculation results, and to perform data preprocessing to construct a sample set for extracting bus short-circuit current control rules.

[0069] The decision tree construction module is used to build decision trees. If all samples in the sample have the same operation mode, a leaf node is generated. According to the distribution of bus short-circuit current, a Gaussian mixture model is used to classify the samples. If the difference between the mean short-circuit current of different classes of samples is less than a fixed value A, a leaf node is generated. If the number of samples of a certain class is less than a fixed value B, a leaf node is generated.

[0070] The decision tree update module is used to calculate the attribute partitioning index for all operating modes, select the optimal partitioning attribute, and construct a subsample set for each value of the optimal partitioning attribute to update the sample and mode data.

[0071] The short-circuit current control rule generation module is used to generate bus short-circuit current control rules based on the generated decision tree.

[0072] The present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the aforementioned method for extracting security and stability control rules based on GMM and decision tree.

[0073] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the above-described method for extracting security and stability control rules based on GMM and decision tree.

Claims

1. A method for extracting a security and stability control rule based on a GMM and a decision tree, characterized by, The method comprises the following steps: (1) obtaining short-circuit current historical calculation results, corresponding time tags, and mode data S corresponding to short-circuit current calculation result time tags, and performing data preprocessing to construct a bus short-circuit current control rule extraction sample set D; (2) constructing a decision tree, and if all sample operation modes in the sample set are the same, generating a leaf node; (3) according to bus short-circuit current distribution, using a Gaussian mixture model to classify samples; The Gaussian mixture model classifying the samples comprises: (3.1) Searching for the optimal Gaussian Mixture Model (GMM) based on the Bayesian Information Criterion k ; (3.2) According to the optimal Gaussian mixture model GMM k The short-circuit current history minimum value of the bus k is Start, with λ as a step, calculate the short-circuit current cumulative distribution function CDF of bus k k ; (3.3) If there exists CDF of consecutive n steps k If the average of the short-circuit current of the consecutive n steps is unchanged, extract the short-circuit current control rule, and record the average of the short-circuit current corresponding to the consecutive n steps as Otherwise, generate a leaf node; (3.4) samples in which the short-circuit current is greater than are recorded as samples D in which the short-circuit current is high high samples in which the short-circuit current is less than high are recorded as samples D in which the short-circuit current is low​ (4) if the difference between short-circuit current means of different class samples is less than a constant value A, generating a leaf node; (5) if the number of a certain class of samples is less than a constant value B, generating a leaf node; (6) Calculate the attribute partition index of all operation modes S, select the optimal partition attribute s * ; (7) select the optimal partition attribute s * for each value of s, the corresponding samples form a sample subset; return to step (3) to update the samples and mode data; (8) generating a bus short-circuit current control rule based on the generated decision tree. 2.The GMM and decision tree based safe stable control rule extraction method according to claim 1, wherein, The mode data in step (1) comprises obtaining generator set start-stop state and line commissioning state data corresponding to all short-circuit current historical calculation result time tags from power grid operation mode historical data. 3.The GMM and decision tree based safe stable control rule extraction method of claim 2, wherein, The data preprocessing in step (1) comprises merging bus short-circuit current calculation results with the same time tags and generator set start-stop state and line commissioning state data, and if there is only bus short-circuit current calculation results or generator set start-stop state and line commissioning state data at a certain time tag, the data at the time tag is removed. 4.The GMM and decision tree based safe stable control rule extraction method of claim 1, wherein, The attribute division index calculation method of the operation mode in step (6) comprises: The product of information gain rate and short-circuit current sensitivity is used as the attribute division index, and the calculation formula is as follows: In the formula, F a is the partition index of attribute a, is the sensitivity of attribute a to the short-circuit current of bus k, is the information gain rate. 5.The GMM and decision tree based safe stable control rule extraction method of claim 4, wherein, In step (7), the sample subset S\{s *} is taken as the sample, and the mode data is returned to step (3) to update the sample and the mode data. 6.The GMM and decision tree based safe stable control rule extraction method of claim 5, wherein, The generation method of the bus short-circuit current control rule in step (8) is as follows: (8.1) each leaf node corresponds to a control rule; (8.2) The node optimization attribute corresponding to the leaf node is s m ; (8.3) if s m If the system is de-energized, the short circuit current decreases by 7. A GMM and decision tree based security and stability control rule extraction system for implementing the GMM and decision tree based security and stability control rule extraction method according to any one of claims 1-6, characterized by It comprises: A data acquisition and preprocessing module is configured to obtain short-circuit current historical calculation results, corresponding time tags, and mode data corresponding to short-circuit current calculation result time tags, and perform data preprocessing to construct a bus short-circuit current control rule extraction sample set; A decision tree construction module is configured to construct a decision tree, and if all sample operation modes in the sample set are the same, generate a leaf node; according to bus short-circuit current distribution, use a Gaussian mixture model to classify samples; if the difference between short-circuit current means of different class samples is less than a constant value A, generate a leaf node; if the number of a certain class of samples is less than a constant value B, generate a leaf node; A decision tree updating module is configured to calculate attribute division indexes of all operation modes, select an optimal division attribute, and select each value of the optimal division attribute to construct a sub-sample set, and update sample and mode data; A short-circuit current control rule generation module is configured to generate a bus short-circuit current control rule based on the generated decision tree.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is loaded into the processor to implement the GMM and decision tree based safety and stability control rule extraction method according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program is executed by the processor to implement the GMM and decision tree based safety and stability control rule extraction method according to any one of claims 1-6.

Images