A voltage stability domain analysis method based on active sampling of network information

Abstract

The present invention relates to the field of DC network analysis, specifically a voltage stability domain analysis method based on active sampling of network information. The method comprises the following steps: obtaining basic network data of the DC network, constructing a network power flow model, analyzing the network power flow model to obtain a voltage stability domain; analyzing the voltage stability domain based on the network power flow model to obtain a stable set; analyzing the stable set to obtain candidate samples; and performing numerical simulation testing on the DC network based on the candidate samples to obtain test results. The present invention can improve the robustness of the network.

Description

Technical Field

[0001] The present invention relates to the field of DC network analysis, and in particular to a voltage stability domain analysis method based on active sampling of network information. Background Art

[0002] The DC network stability region (SVSR) refers to a concept for stability analysis and control of voltage source rectifiers (VSRs) in DC power systems (such as DC transmission systems). In DC power systems, SVSR is primarily used to ensure that power electronic devices (such as rectifiers and inverters) can maintain stable system operation under various operating conditions. High-voltage direct current (HVDC) transmission systems, in particular, typically rely on voltage source rectifiers (VSRs) to convert alternating current (AC) to direct current (DC) and ensure the stability of the DC grid. SVSR focuses on how to ensure that VSRs in these systems can maintain stable operation under various dynamic conditions and disturbances.

[0003] The security-constrained economic dispatch method for security domains with the publication number CN102983573A discloses a security-constrained economic dispatch method based on security domains, the method comprising: The present invention relates to power systems. In order to provide an effective way to deal with the network security constraints of the system in the traditional optimization operation problem of power system economic dispatch, including branch power flow, static voltage stability and transient stability constraints and the contradiction between economy and safety, the technical solution adopted by the present invention is a security-constrained economic dispatch method based on security domains, comprising the following steps: the first step: respectively calculating the active static safety domain boundary coefficient, the cut set voltage stability domain boundary coefficient and the dynamic safety domain boundary coefficient of the system; the second step: establishing a security-constrained economic dispatch model based on the security domain; the third step: using a cognitive-based social evolution algorithm to solve the unit start-stop state optimization sub-problem; the fourth step: calculating the power generation cost, static voltage stability margin and transient stability margin of each unit within the dispatch period; the fifth step: obtaining a feasible economic dispatch plan; otherwise, returning to the third step. The present invention is mainly used for power load allocation optimization.

[0004] In existing technologies, the intrinsic information of the sample space and the physical characteristics of the network are not effectively utilized, resulting in a significant increase in computing resource consumption when the data volume is huge or the sample distribution is complex; in addition, when dealing with large-scale networks, numerical methods face computational bottlenecks and cannot efficiently deal with high-complexity problems; the characterization flexibility is insufficient, and when faced with complex and changeable actual systems, the adaptability and flexibility of the analysis method are limited, making it difficult to adapt to the dynamic changes brought about by the high penetration rate of renewable energy, resulting in insufficient characterization accuracy; the training data scale is highly dependent, and the geometric characteristics of the sample space and the useful information in the iterative process are not fully utilized, resulting in high data preparation and computing costs, especially when dealing with large-scale systems, where efficiency and accuracy are limited; the geometric structure information of the sample space is not fully utilized, which increases the computing cost and reduces efficiency, and it is impossible to accurately sample and train according to the characteristics of the sample space, which is a problem we need to solve. Summary of the Invention

[0005] The present invention aims to address the problems existing in the background technology and propose a voltage stability domain analysis method based on active sampling of network information.

[0006] The technical solution of the present invention is a voltage stability domain analysis method based on active sampling of network information, comprising the following steps:

[0007] S1. Obtain basic network data of the DC network, build a network power flow model, analyze the network power flow model, and obtain the voltage stability domain;

[0008] S2. Analyze the voltage stability domain according to the network power flow model and obtain the stable set;

[0009] S3. Analyze the stable set to obtain candidate samples;

[0010] S4. Combine the candidate samples and perform numerical simulation tests on the DC network to obtain test results.

[0011] Preferably, the process of obtaining basic network data of the DC network and constructing a network power flow model includes:

[0012] The basic network data includes network structure information and network relationship information; the network structure information includes voltage source information and load information; the network relationship information includes the relationship between the voltage source and the load and the relationship between the loads;

[0013] Through the relationship between the voltage source and the load and the relationship between the loads, the voltage source and the load are connected to construct a DC network model.

[0014] Preferably, the process of analyzing the network power flow model to obtain the voltage stability region includes:

[0015] Obtain the operating points of the DC network model, denoted as node 1, node 2, node 3, ..., node i, ..., node j, ..., node u; when there is a power line between node i and node j, denoted as ; When there is no power line between node i and node j, it is recorded as ; The node and nodes The conductance of the electric line between , is a positive real number, then the Kirchhoff matrix of the DC network model is for:

[0016] ;

[0017] By judging whether the node is a load or a voltage source, the Kirchhoff matrix Divide it into ;in, is the load quantity; is the number of voltage sources; specify the matrix as well as It is positive;

[0018] Obtain the load power, load voltage, external power supply voltage and controllable voltage set point, then the vectors of load power, load voltage, external power supply voltage and controllable voltage set point are , , ,and ,in ;

[0019] The voltage potential vector at the node is denoted as , and stipulates ; The current injected into the network at the node is recorded as ;According to the voltage potential vector at the node and the current injected into the network at the node, the power vector p provided by the node to the grid is obtained;

[0020] ;

[0021] According to Kirchhoff and Ohm's law, Expressed as:

[0022] ;

[0023] According to formula (3) and formula (4), the power vector is obtained;

[0024] ;

[0025] The system steady-state operating point is recorded as , the load bus voltage at the operating point is recorded as , the power vector input from the power node to the system is recorded as According to the system steady-state operating point, the load bus voltage at the operating point, the power vector input from the power node to the system, and equation (5), the parameters of the items on the right side of equation (5) are replaced to obtain the static power flow equation, equation (6). The static power flow equation is input into the DC network model, the network power flow model is constructed, and the static power flow equation is analyzed.

[0026] ;

[0027] in, , as well as are the conductance matrices of the source-source, source-load and load-load subnetworks in the network power flow model respectively; = , and , is a submatrix of the admittance matrix; For state variables is quadratic, for variables and state variables It is bilinear.

[0028] Preferably, the process of analyzing the static power flow equation is:

[0029] Obtain the solution of the static power flow equation, which is recorded as the static operating point. If the static power flow equation has no solution, it means that the static voltage of the system is unstable.

[0030] The voltage stability domain is specified as:

[0031] ;

[0032] If any given controllable voltage set point , all exist , satisfy all The load injection power, Equation (6) has a solution, then the power set , denoted as the voltage stability region.

[0033] Preferably, the process of analyzing the voltage stability region according to the network power flow model and obtaining the stable set includes:

[0034] Specified in the system Under these conditions, there is a high load value , and this load value meets the voltage stability domain conditions of the DC network , then when When , it means that the voltage stability domain condition of the DC network is met;

[0035] when When there is a solution to the static power flow equation, There are solutions for all;

[0036] By using the characteristics of the voltage stability domain, which are monotonicity and convexity, a stable subdomain is constructed. ;

[0037] Assume that there are m elements in the voltage stability domain ) , forming a stable set P;

[0038] .

[0039] Preferably, the process of analyzing the stable set to obtain candidate samples includes:

[0040] Closed-loop training is performed on the stable set of the DC network. After actively sampling the stable set, sample labels and data sets are obtained. The closed-loop training includes classifier training, sampling, screening, labeling and SVSR subset identification. The classifier training refers to training the classifier using an extended training set containing newly selected samples and existing labeled samples. The sampling refers to randomly selecting a batch of samples in the unlabeled sample space. The screening refers to evaluating the posterior probability of the selected samples to determine their uncertainty and selecting the samples with the highest uncertainty as the new candidate labeling set. The labeling refers to labeling the screened candidate samples through network numerical simulation. The SVSR subset identification refers to identifying the SVSR subset that supports the labeling process using the physical properties of the network.

[0041] Set feasibility posterior probability The feasibility posterior probability is used to measure the probability that the tidal current has a solution; the probability that the static tidal current equation has a solution is analyzed through the feasibility posterior probability.

[0042] Preferably, the process of analyzing the probability of having a solution to the static power flow equation by using the feasibility posterior probability includes:

[0043] when When the value of is close to 1 or 0, it means that the sample is solvable (that is, belongs to the target category) or unsolvable (that is, does not belong to the target category); when When the value is close to 0.5, it means that the model is uncertain about the classification result of the sample and the feasibility posterior probability is Perform analysis, if is greater than 0.5, the quantitative uncertainty measure is defined as:

[0044] ;

[0045] like is less than or equal to 0.5, and the quantitative uncertainty measure is defined as: ;

[0046] Set the training cycle, sort the samples according to the quantitative uncertainty, and record the samples with the highest uncertainty as candidate samples through the training cycle; the number of initial samples and the number of samples in each training cycle can be set.

[0047] Preferably, the process of performing a numerical simulation test on the DC network in combination with the candidate samples and obtaining the test results includes:

[0048] Through numerical simulation tests, tests were conducted on IEEE14-node, 39-node and 118-node DC networks. Numerical calculations were performed using IPOPT, and simulations were performed using Python and Julia to obtain candidate samples. The test data was recorded and the test results were obtained.

[0049] Compared with the existing technology, the above technical scheme of the present invention has the following beneficial technical effects: by constructing a network flow model and a voltage stability domain, analyzing the monotonicity of the DC network flow equation and the convex set characteristics of SVSR, an active sampling technology is proposed, which can accurately select the samples with the most information value and reduce the dependence on large-scale data. Through active sampling, the consumption of computing resources is significantly reduced, the computing efficiency is improved, and the accuracy of SVSR characterization is improved; through the stable set, the convex set characteristics of SVSR and the monotonicity of the flow equation are utilized to improve the accuracy and adaptability of SVSR characterization, especially when dealing with dynamic changes in the context of high penetration of renewable energy; through candidate samples, by quantifying the uncertainty and geometric information of the sample space, the sample selection and labeling process is iteratively optimized; through closed-loop screening, the dependence on the scale of training data is reduced, the model robustness and characterization flexibility are improved, and based on the convexity of SVSR and the monotonicity of the flow equation, a mathematical theoretical guarantee is provided, and a complete set of DC network stability domain characterization algorithms is constructed to ensure the high accuracy and wide applicability of the method. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The present invention is a flowchart of an embodiment of the present invention. DETAILED DESCRIPTION

[0051] Example 1, as Figure 1 As shown, the present invention proposes a voltage stability domain analysis method based on active sampling of network information, comprising the following steps:

[0052] S1. Obtain basic network data of the DC network, build a network power flow model, analyze the network power flow model, and obtain the voltage stability domain;

[0053] S2. Analyze the voltage stability domain according to the network power flow model and obtain the stable set;

[0054] S3. Analyze the stable set to obtain candidate samples;

[0055] S4. Combine the candidate samples and perform numerical simulation tests on the DC network to obtain test results.

[0056] It should be further explained that, in the specific implementation process, the basic network data of the DC network is obtained, the network power flow model is constructed, and the network power flow model is analyzed to obtain the stability region. The process is as follows:

[0057] The basic network data includes network structure information and network relationship information; the network structure information includes voltage source information and load information; the network relationship information includes the relationship between the voltage source and the load and the relationship between the loads;

[0058] It should be further explained that, in the specific implementation process, the relationship between the voltage source and the load refers to the transmission relationship between the voltage source and the load in the DC network; the relationship between the loads refers to the transmission relationship between the loads in the DC network;

[0059] By connecting the voltage source and the load and the load and the load to the load, a DC network model is constructed;

[0060] Obtain the operating points of the DC network model, denoted as node 1, node 2, node 3, ..., node i, ..., node j, ..., node u; when there is a power line between node i and node j, denoted as ; When there is no power line between node i and node j, it is recorded as ; The node and nodes The conductance of the electric line between , is a positive real number, then the Kirchhoff matrix of the DC network model is for:

[0061] ;

[0062] By judging whether the node is a load or a voltage source, the Kirchhoff matrix Divide it into ;in, is the load quantity; is the number of voltage sources; specify the matrix as well as It is positive;

[0063] Obtain the load power, load voltage, external power supply voltage and controllable voltage set point, then the vectors of load power, load voltage, external power supply voltage and controllable voltage set point are , , ,and ,in ;

[0064] The voltage potential vector at the node is denoted as , and stipulates ; The current injected into the network at the node is recorded as ;According to the voltage potential vector at the node and the current injected into the network at the node, the power vector p provided by the node to the grid is obtained;

[0065] ;

[0066] According to Kirchhoff and Ohm's law, Expressed as:

[0067] ;

[0068] According to formula (3) and formula (4), the power vector is obtained;

[0069] ;

[0070] The system steady-state operating point is recorded as , the load bus voltage at the operating point is recorded as , the power vector input from the power node to the system is recorded as According to the system steady-state operating point, the load bus voltage at the operating point, the power vector input from the power node to the system, and equation (5), the parameters of the items on the right side of equation (5) are replaced to obtain the static power flow equation, equation (6). The static power flow equation is input into the DC network model to construct the network power flow model.

[0071] ;

[0072] in, , as well as are the conductance matrices of the source-source, source-load and load-load subnetworks in the network power flow model respectively; = , and , is a submatrix of the admittance matrix; For state variables is quadratic, for variables and state variables is bilinear;

[0073] Obtain the solution of the static power flow equation, which is recorded as the static operating point. If the static power flow equation has no solution, it means that the static voltage of the system is unstable.

[0074] The voltage stability domain is specified as:

[0075] ;

[0076] If any given controllable voltage set point , all exist , satisfy all The load injection power, Equation (6) has a solution, then the power set , denoted as the voltage stability domain;

[0077] It should be further explained that, in the specific implementation process, if a specific load power , then the load power There is a solution.

[0078] It should be further explained that, in the specific implementation process, the voltage stability domain is analyzed according to the network power flow model, and the process of obtaining the stable set is as follows:

[0079] Specified in the system Under these conditions, there is a high load value , and this load value meets the voltage stability domain conditions of the DC network , then when When , it means that the voltage stability domain condition of the DC network is met;

[0080] when When there is a solution to the static power flow equation, There are solutions for all;

[0081] By using the characteristics of the voltage stability domain, which are monotonicity and convexity, a stable subdomain is constructed. ;

[0082] Assume that there are m elements in the voltage stability domain ) , forming a stable set P;

[0083] .

[0084] It should be further explained that, in the specific implementation process, the process of analyzing the stable set and obtaining candidate samples is as follows:

[0085] Closed-loop training is performed on the stable set of the DC network. After actively sampling the stable set, sample labels and data sets are obtained. The closed-loop training includes classifier training, sampling, screening, labeling and SVSR subset identification. The classifier training refers to training the classifier using an extended training set containing newly selected samples and existing labeled samples. The sampling refers to randomly selecting a batch of samples in the unlabeled sample space. The screening refers to evaluating the posterior probability of the selected samples to determine their uncertainty and selecting the samples with the highest uncertainty as the new candidate labeling set. The labeling refers to labeling the screened candidate samples through network numerical simulation. The SVSR subset identification refers to identifying the SVSR subset that supports the labeling process using the physical properties of the network.

[0086] Set feasibility posterior probability , the feasibility posterior probability is used to measure the probability that the current has a solution; when When the value of is close to 1 or 0, it means that the sample is solvable (that is, belongs to the target category) or unsolvable (that is, does not belong to the target category); when When the value is close to 0.5, it means that the model is uncertain about the classification result of the sample and the feasibility posterior probability is Perform analysis, if is greater than 0.5, the quantitative uncertainty measure is defined as:

[0087] ;

[0088] like is less than or equal to 0.5, and the quantitative uncertainty measure is defined as: ;

[0089] Set the training cycle, sort the samples according to the quantitative uncertainty, and record the samples with the highest uncertainty as candidate samples through the training cycle; the number of initial samples and the number of samples in each training cycle can be set.

[0090] It should be further explained that, in the specific implementation process, the candidate samples are combined to conduct numerical simulation tests on the DC network, and the process of obtaining the test results is as follows:

[0091] Through numerical simulation tests, tests were conducted on IEEE14-node, 39-node and 118-node DC networks. Numerical calculations were performed using IPOPT, and simulations were performed using Python and Julia to obtain candidate samples. The test data was recorded and the test results were obtained.

[0092] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A voltage stability domain analysis method based on active sampling of network information, characterized in that: The following steps are involved: S1. Obtain basic network data of the DC network, build a network power flow model, analyze the network power flow model, and obtain the voltage stability domain; S2. Analyze the voltage stability domain according to the network power flow model and obtain the stable set; S3. Analyze the stable set to obtain candidate samples; S4. Combine the candidate samples and perform numerical simulation tests on the DC network to obtain test results; The process of obtaining basic DC network data and building a network power flow model includes: The basic network data includes network structure information and network relationship information; the network structure information includes voltage source information and load information; the network relationship information includes the relationship between the voltage source and the load and the relationship between the loads; By connecting the voltage source and the load and the load and the load to the load, a DC network model is constructed; The process of analyzing the network power flow model and obtaining the voltage stability region includes: Obtain the operating points of the DC network model, denoted as node 1, node 2, node 3, ..., node i, ..., node j, ..., node u; when there is a power line between node i and node j, denoted as ; When there is no power line between node i and node j, it is recorded as ; The node and nodes The conductance of the electric line between , is a positive real number, then the Kirchhoff matrix of the DC network model is for: ； By judging whether the node is a load or a voltage source, the Kirchhoff matrix Divide it into ;in, is the load quantity; is the number of voltage sources; specify the matrix as well as It is positive; Obtain the load power, load voltage, external power supply voltage and controllable voltage set point, then the vectors of load power, load voltage, external power supply voltage and controllable voltage set point are , , ,and ,in ; The voltage potential vector at the node is denoted as , and stipulates ; The current injected into the network at the node is recorded as ;According to the voltage potential vector at the node and the current injected into the network at the node, the power vector p provided by the node to the grid is obtained; ； According to Kirchhoff and Ohm's law, Expressed as: ； According to formula (3) and formula (4), the power vector is obtained; ； The system steady-state operating point is recorded as , the load bus voltage at the operating point is recorded as , the power vector input from the power node to the system is recorded as According to the system steady-state operating point, the load bus voltage at the operating point, the power vector input from the power node to the system, and equation (5), the parameters of the items on the right side of equation (5) are replaced to obtain the static power flow equation, equation (6). The static power flow equation is input into the DC network model, the network power flow model is constructed, and the static power flow equation is analyzed. ； in, , as well as are the conductance matrices of the source-source, source-load and load-load subnetworks in the network power flow model respectively; = , and , is a submatrix of the admittance matrix; For state variables is quadratic, for variables and state variables It is bilinear.

2. The voltage stability domain analysis method based on active network information sampling according to claim 1, characterized in that: The process of analyzing the static power flow equation is: Obtain the solution of the static power flow equation, which is recorded as the static operating point. If the static power flow equation has no solution, it means that the static voltage of the system is unstable. The voltage stability domain is specified as: ； If any given controllable voltage set point , all exist , satisfy all The load injection power, Equation (6) has a solution, then the power set , denoted as the voltage stability region.

3. The voltage stability domain analysis method based on active network information sampling according to claim 2, characterized in that: The voltage stability domain is analyzed based on the network power flow model. The process of obtaining the stable set includes: Specified in the system Under these conditions, there is a high load value , and this load value meets the voltage stability domain conditions of the DC network , then when When , it means that the voltage stability domain condition of the DC network is met; when When there is a solution to the static power flow equation, There are solutions for all; By using the characteristics of the voltage stability domain, which are monotonicity and convexity, a stable subdomain is constructed. ; Assume that there are m elements in the voltage stability domain ) , forming a stable set P; 。 4. The voltage stability domain analysis method based on active network information sampling according to claim 3 is characterized in that: The process of analyzing the stable set and obtaining candidate samples includes: Closed-loop training is performed on the stable set of the DC network. After actively sampling the stable set, sample labels and data sets are obtained. The closed-loop training includes classifier training, sampling, screening, labeling and SVSR subset identification. The classifier training refers to training the classifier using an extended training set containing newly selected samples and existing labeled samples. The sampling refers to randomly selecting a batch of samples in the unlabeled sample space. The screening refers to evaluating the posterior probability of the selected samples to determine their uncertainty and selecting the samples with the highest uncertainty as the new candidate labeling set. The labeling refers to labeling the screened candidate samples through network numerical simulation. The SVSR subset identification refers to identifying the SVSR subset that supports the labeling process using the physical properties of the network. Set feasibility posterior probability The feasibility posterior probability is used to measure the probability that the tidal current has a solution; the probability that the static tidal current equation has a solution is analyzed through the feasibility posterior probability.

5. The voltage stability domain analysis method based on active network information sampling according to claim 4, characterized in that: The process of analyzing the probability of a solution to the static power flow equation using the feasibility posterior probability includes: when When the value of is close to 1 or 0, it means that the sample is solvable or unsolvable; when When the value is close to 0.5, it means that the model is uncertain about the classification result of the sample and the feasibility posterior probability is Perform analysis, if is greater than 0.5, the quantitative uncertainty measure is defined as: ； like is less than or equal to 0.5, and the quantitative uncertainty measure is defined as: ; Set the training cycle, sort the samples according to the quantitative uncertainty, and record the samples with the highest uncertainty as candidate samples through the training cycle; the number of initial samples and the number of samples in each training cycle can be set.

6. The voltage stability domain analysis method based on active network information sampling according to claim 5, characterized in that: Combined with the candidate samples, the DC network is numerically simulated and tested. The process of obtaining the test results includes: Through numerical simulation tests, tests were conducted on IEEE14-node, 39-node and 118-node DC networks. Numerical calculations were performed using IPOPT, and simulations were performed using Python and Julia to obtain candidate samples. The test data was recorded and the test results were obtained.

Images