A nonparametric independent screening method for key factors of multi-dc sending terminal system planning

Abstract

The application relates to a nonparametric independent screening method for key factors of a multi-direct-current sending terminal system planning, and belongs to the technical field of power transmission system planning and operation. First, a MAP calculation method is provided. On this basis, a multi-layer neural network average influence value method is used to identify the determining factors of accurate planning sending terminals. In addition, a group lasso is used to construct a calculation rule of important planning indexes relative to related operation variables and prospective planning. Embodiments prove the feasibility and accuracy of the method in mining key factors and potential correlations. Compared with traditional empirical sending terminal planning strategies, the method has more accurate, more economical, more scientific and more reliable planning effects.

Description

Technical Field

[0001] This invention belongs to the field of power transmission system planning and operation technology, specifically involving a non-parametric independent screening method for key factors in the planning of multi-DC sending-end systems. Background Technology

[0002] Due to the uneven distribution of energy resources and China's vast territory, long-distance power transmission has become a major characteristic of China's power grid. The West-to-East Power Transmission Project is a key solution for improving energy allocation. High-voltage direct current (HVDC) transmission technology will play a significant role in promoting renewable energy development and reducing carbon emissions. Among several AC / DC coupling modes, asynchronous HVDC transmission systems are the mainstream of today's power grid because they decouple large synchronous grids into multiple asynchronous regional grids, improving dynamic security. However, this interconnection method results in each regional grid bearing a huge DC load, placing higher demands on the AC grid. Strengthening the construction of the AC grid requires adopting reasonable planning schemes and establishing corresponding indicator systems.

[0003] Numerous studies have been conducted to measure the strength of AC systems in high-voltage direct current (HVDC) transmission systems. Among these, multi-DC-point systems have become a research hotspot due to their widespread practical application. To analyze the power stability of multi-DC-point systems, the interactions between the individual DC lines must be considered. Therefore, the International Conference on Large Electric Systems (CIGRE) proposed the Multiple Infeed Effective Short-Circuit Ratio (MIESCR) index to describe the voltage support strength of the AC system relative to the converter station. To a certain extent, MIESCR can roughly assess the power stability of multi-infeed DC systems. For multi-DC-output systems, studies have demonstrated the feasibility of MIESCR. However, some scholars have theoretically questioned the comprehensiveness of MIESCR. Based on this, scholars have derived some non-empirical, rigorous indices from power flow models, such as the node voltage sensitivity factor. Although these indices perform well in theoretical interpretation, they are limited to providing direct guidance for AC grid planning.

[0004] Another key indicator, the Maximum Transmission Power (MAP), reflects the power stability margin of a high-voltage direct current (HVDC) transmission system. Some scholars have analyzed its direct relationship with MIESCRI, making MAP a direct indicator for AC-side planning. However, current research on MAP largely relies on AC system equivalence, which introduces significant errors into MAP calculations, thus reducing the economic efficiency and security of AC grid planning. Furthermore, because it is difficult to analyze the response mechanism of MAP to various operational and structural variables from a global perspective, existing research can only roughly understand the local trends of MAP response through the control variable method.

[0005] Existing literature has greatly simplified multi-output asynchronous AC / DC systems. Furthermore, there is a lack of research on the correlation between system response and short-circuit ratio and maximum DC power under severe DC faults (such as commutation failure). Current research heavily relies on system equivalent techniques, but errors in these techniques can lead to inaccurate calculations of indicators, consequently reducing the economic efficiency and safety of planning. Analyzing the response mechanism of maximum DC power to various operational and structural variables from a global perspective is extremely difficult; existing research primarily uses the controlled variable method to roughly observe trends in DC operating indicators.

[0006] Therefore, at this stage, a non-parametric independent screening method for key factors in the planning of multi-DC transmission end systems needs to be designed to solve the above problems. Summary of the Invention

[0007] The purpose of this invention is to provide a non-parametric independent screening method for key factors in the planning of multiple DC sending-end systems, which solves the technical problems existing in the prior art. In order to accurately capture the correlation between grid structure parameters, DC control parameters and key indicators such as DC power stability and commutation failure transient overvoltage, this invention analyzes a large amount of operational data perturbed by structure and control parameters, and realizes the discovery of key source parameters affecting the stability of DC system and the construction of correlation relationships with the help of data-driven technology, thus providing a reference for DC planning of sending-end systems.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows:

[0009] A nonparametric independent screening method for key factors in the planning of a multi-DC transmission-end system includes the following steps:

[0010] S1. Obtain the basic indicators of the multiple DC transmission system;

[0011] S2. Analyze the power stability of the multi-DC transmission system based on its basic performance indicators.

[0012] S3. The average influence value method based on multi-layer neural networks is used to identify the key factors affecting the key indicators of the multi-DC transmission system.

[0013] Furthermore, the basic indicators in step S1 are as follows:

[0014] 1) Multi-feed interaction factor and multi-feed effective short-circuit ratio.

[0015] 2) Transient overvoltage: Consider temporary overvoltage caused by commutation failure at the receiving end.

[0016] Furthermore, step S2 is detailed as follows:

[0017] The maximum power curve (MPC) is an important concept for analyzing the power stability of high-voltage direct current transmission systems.

[0018] There is a breakpoint in the MPC, which is the maximum value of DC power increasing with DC current, called MAP; in order to develop the MPC of a specific system, the τ of the i-th rectifier station i ,α i The reactive power compensation remains unchanged; then, by increasing each current setting value according to the set path, the MPC of the multi-DC transmission system can be plotted.

[0019] Two strategies for determining the DC current growth path are proposed: one is to keep the DC current other than the selected DC current constant, and the other is to keep the setpoint of all DC currents in the power grid increasing synchronously. This invention makes the following assumptions regarding the obtained MPC:

[0020]

[0021] In the formula, I dni This represents the rated current of the i-th DC feeder, and M is the number of DC feeder lines.

[0022] By combining the DC current gain, we finally obtain MPC and MAP.

[0023] Furthermore, step S3 is as follows:

[0024] S301: Non-parameter independent screening;

[0025] Before identifying critical impacts, operational data from the multi-DC transmission system needs to be prepared in advance. This is achieved through sampling using a set of optimal points in the parameter space. The parameter space is established using operational constraints and planning methods, and GPS is deployed in this space. This determines the input characteristics as {impedance of each AC line, reactive power limit of the generator, reactive power compensation of the rectifier station, active power generation, reactive power generation, DC current setpoint, rated DC current, rectifier firing angle, load, MOIF, MOESCR}, and the outputs as {TOV, MAP, and corresponding DC current}. All inputs are directly obtainable. To indicate input, use Let represent the output, where n represents the sample size, and p and t represent the dimensions of the input and output, respectively; furthermore, all data-driven tasks will be separated into t subtasks, the output of which is y. i Let i = 1, ..., t, and y represent y in the following description. i ;

[0026] The basic principle of nonparametric independent screening is to find the numerical relationship between a single input feature and the target / output feature through univariate regression, and to quantify the strength of the relationship through the sum of squared residuals.

[0027] For different DC system indicators, the directly related factors vary. Each related factor can be treated independently to observe the correlation strength between a single factor and the DC system indicator. During the non-parametric independent screening process, the correlation between a certain dimension of the input features and the output indicator is considered independently. Based on the above analysis, the correlation analysis problem is simplified as shown in the following formula:

[0028]

[0029] In the formula, i is the index of any randomly selected branch in the target feature set. Consistent with the meaning of x; let The non-parameter independent filtering process is as follows:

[0030] i) Assume there exists a numerical relationship model between the j-th input feature and the target. Therefore:

[0031]

[0032] In the formula, ε is the correlation error that follows a zero-mean normal distribution;

[0033] ii) Use univariate nonparametric regression to establish an approximate correlation model between each input feature and the output. And calculate the target value estimated by the approximate correlation model:

[0034]

[0035] In the formula: This represents the target value predicted using the j-th input feature and an approximate correlation model; noting the strong nonlinearity of TTC, a cubic B-spline function is used to construct...

[0036] iii) Calculate the estimation error of the target for each input feature and the approximate correlation model:

[0037]

[0038] iv) For sets

[0039] RSS = {RSS} j j = 1, 2, ..., n F}

[0040] Input features are sorted in ascending order, with those appearing earlier contributing more to the target. Input features strongly correlated with the target are then selected using certain criteria, as shown below:

[0041]

[0042] S302: Lasso set;

[0043] The group lasso is developed based on the minimum absolute contraction and lasso algorithm; it can extract explicit calculation rules for relevant variables; the numerical correlation between inputs and relevant indicators can be explicitly represented as an additivity model with a constant intercept ρ; then GL is constructed as a structural risk minimizer with the l1 norm:

[0044]

[0045]

[0046]

[0047] The extrapolation residual method is used to eliminate the lasso constraint and solve problems (6)-(8).

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] One of the beneficial effects of this solution is that it proposes a data-driven key factor identification method for the planning of multiple DC transmission systems. First, a MAP calculation method is proposed. Based on this, a multi-layer neural network average influence value method is used to identify the determinants of accurate transmission planning. Furthermore, a set of rules is used to construct calculation rules for important planning indicators relative to relevant operational variables and long-term planning. Examples demonstrate the feasibility and accuracy of this method in uncovering key factors and potential correlations. Compared with traditional empirical transmission planning strategies, this method achieves more accurate, economical, scientific, and reliable planning results. Attached Figure Description

[0050] Figure 1 This is a single-line schematic diagram of the two DC transmission systems of the present invention.

[0051] Figure 2 This is a schematic diagram illustrating the importance scores of the candidate factors in this invention.

[0052] Figure 3 This is a schematic diagram illustrating the evaluation performance of the rules extracted by GL in this invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0054] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0055] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0056] This invention proposes a method for identifying key factors in the planning of multi-DC transmission-end systems based on non-parametric independent screening and grouping; the details are as follows.

[0057] 1. Basic Indicators of Multiple DC Transmission Systems

[0058] To date, numerous indices have been proposed for measuring the strength of AC systems with DC feeders. Inspired by some generalized indices, we propose to similarly define several sending-end indices, as follows:

[0059] 1) Multi-outlet interaction factor (MOIF) and multi-outlet effective short-circuit ratio (MOESCR): These metrics are the same as the corresponding metrics at the receiving end, namely MIIF and MIESCR.

[0060] 2) Transient Overvoltage (TOV): Consider temporary overvoltage caused by receiving-end commutation failure. After the fault occurs, a large amount of reactive power compensation will be injected back into the AC system, leading to the TOV phenomenon. PSCAD will be used to simulate the maximum TOV value.

[0061] 2. Power stability of multiple DC transmission systems

[0062] The maximum power curve (MPC) is an important concept for analyzing the power stability of high-voltage direct current (HVDC) transmission systems. The MPC has an extreme point, which is the maximum value of DC power as DC current increases, called MAP (Maximum Available Power). To allow the MPC of a specific system to develop, the τ of the i-th rectifier station... i ,α i The reactive power compensation usually remains unchanged. Then, by increasing each current setpoint according to the set path, the MPC of the multi-DC transmission system can be plotted.

[0063] Two strategies for determining the DC current growth path have been proposed. One is to keep the DC current other than the selected DC current constant, and the other is to keep the setpoints of all DC currents in the power grid increasing synchronously. The latter is better because it can reflect the interaction between multiple DC currents to a certain extent. This invention makes the following assumptions for obtaining the MPC:

[0064]

[0065] In the formula, I dni This represents the rated current of the i-th DC feed, and M is the number of DC feed lines.

[0066] By combining the DC current gain, we finally obtain MPC and MAP.

[0067] 3. Identification Methods for Key Factors in the Sending-End Planning of Multi-DC Transmission Systems

[0068] Introducing actual dynamic models into computational metrics helps to plan the sending-end power grid more accurately and realistically, but it undoubtedly increases the nonlinearity of multi-DC transmission system models and the complexity of theoretical analysis. Control variable combination simulation methods are the mainstream approach to address these challenges, but the persuasiveness and accuracy of illustrating the global response using fewer local experiments are insufficient. This invention employs a data-driven approach to identify key factors affecting AC / DC intensity. This novel approach can describe the relationships between complex metrics or physical and grid-planning-related parameters (such as impedance, shunt arrays, etc.) from a broader perspective. Furthermore, this method avoids reducing the accuracy of equivalent system modeling.

[0069] 3.1 Non-parametric independent screening

[0070] Before identifying critical impacts, operational data for the multi-DC transmission system must be prepared in advance. This can be achieved using optimal point set sampling (GPS) in the parameter space. Specifically, the parameter space is established using existing operational constraints and planning methods, and GPS is deployed in this space. The input characteristics are defined as {impedance of each AC line, reactive power limit of the generator, reactive power compensation of the rectifier station, active power generation, reactive power generation, DC current setpoint, rated DC current, rectifier firing angle, load, MOIF, MOESCR}, and the outputs include {TOV, MAP, and corresponding DC current}. All inputs are directly available. For simplicity, use... To indicate input, use Let represent the output, where n represents the sample size, and p and t represent the dimensions of the input and output, respectively. Furthermore, all data-driven tasks will be divided into t subtasks, whose output is y. i Let i = 1, ..., t, and y represent y in the following description. i .

[0071] Nonparametric independence screening (NIS) was proposed to address this problem. The basic principle of NIS is to find the numerical relationship between a single input feature and the target / output feature through univariate regression, and to quantify the strength of the relationship through the residual sum of squares (RSS). The lower the RSS, the stronger the correlation between the input and output (or, the stronger the contribution and influence of the input feature on the target feature).

[0072] The directly related factors vary for different DC system indicators. However, generally speaking, each related factor can be treated independently to observe the correlation strength between a single factor and the DC system indicator. Therefore, during NIS execution, the correlation between a certain dimension of the input features and the output indicator can be considered independently. Based on the above analysis, the correlation analysis problem can be simplified as follows:

[0073]

[0074] In the formula, i is the index of any randomly selected branch in the target feature set. The meaning is consistent with x. To simplify the description, let... The NIS process is as follows:

[0075] i) Assume there exists a numerical relationship model between the j-th input feature and the target. Therefore:

[0076]

[0077] In the formula, ε is the correlation error that follows a zero-mean normal distribution, which is usually 0;

[0078] ii) Use univariate nonparametric regression to establish an approximate correlation model between each input feature and the output. And calculate the target value estimated by the approximate correlation model:

[0079]

[0080] In the formula: This represents the target value predicted using the j-th input feature and an approximate correlation model. Note the strong nonlinearity of TTC; simple linear regression cannot effectively reflect the correlation between the input and the target. Therefore, this invention uses a cubic B-spline function to construct...

[0081] iii) Calculate the estimation error of the target for each input feature and the approximate correlation model:

[0082]

[0083] iv) For sets

[0084] RSS = {RSS} j j = 1, 2, ..., n F}

[0085] The input features are sorted in ascending order, with those appearing earlier contributing more to the target. Strongly correlated features (SCFs) are then selected based on certain criteria, one of which is shown below:

[0086]

[0087] 3.2 Set of lassos

[0088] Group Lasso is developed based on the traditional Lasso algorithm (Minimum Absolute Contraction). It is a data-driven regression algorithm that extracts explicit computational rules for relevant variables. The numerical correlation between the input and relevant indicators can be explicitly represented as an additivity model with a constant intercept ρ. GL is then constructed as a structural risk minimizer with the l1 norm.

[0089]

[0090]

[0091]

[0092] This invention uses the extrapolation residual method (Celer) to eliminate the lasso constraint and solve problems (6)-(8).

[0093] The proposed method was tested on a two-DC transmission system, the topology of which is as follows: Figure 1 As shown.

[0094] The power flow model in the AC system is solved using Matpower, and then the algebraic equations (9) to (13) are solved using the voltage at the converter. The DC power can then be calculated and fed back to the AC system as a load.

[0095] Through the data preparation process in 5.3.1, 1500 data samples were prepared for the later analysis. 80% of the entire dataset was used for training, and the remainder was used for testing.

[0096] Key factors in determining the delivery end planning

[0097] Key factors were identified for all outputs. Scores were sorted in descending order of importance as follows: Figure 2 As shown.

[0098] Table 1 lists the key factors.

[0099]

[0100]

[0101] Note: Busbar 3 and Busbar 4 are the conversion busbars connected to Busbar 1 and Busbar 2, respectively.

[0102] according to Figure 2 The importance scores identified the top nine most critical factors for the relevant indicators, as shown in Table 1. Additionally, the main influencing factors on the TOV of Bus 3 are also given, although their importance scores are not displayed. To demonstrate the feasibility and correctness of this method, we can combine the results in Table 1 with qualitative theoretical analysis. Taking the MAP of Bus 3 as an example (for which there are many mature theoretical analyses), we can see a strong correlation between the active power generation of Bus 1 and MAP, as the output DC power mainly comes from MAP. Some indicators of Bus 4 are also important because the output DC power of Bus 4 has a significant impact on the active power generation of Bus 1. For the TOV of Bus 3, local reactive power compensation is the main influencing factor, leading to most of the reactive power generation, which is the root cause of TOV.

[0103] The results of the data analysis are in high agreement with the physical data. These arguments support the effectiveness of the method.

[0104] Discover the calculation rules for key indicators

[0105] Based on the identification of key factors, GL (Geometry and LiGenius) is applied to learn explicit calculation rules for relevant indicators regarding the key factors of the sending-end planning. The performance of the rules extracted by GL is as follows: Figure 3 As shown.

[0106] from Figure 3 As can be seen, the method proposed in this invention can achieve accurate rule discovery. The mean square error between the estimated value and the true value is 0.043 pu. Although there is significant distortion at the edges of the test set distribution, this is expected to be improved by expanding the sampling space and using piecewise or nonlinear regression methods.

[0107] In summary, this invention proposes a data-driven key factor identification method for the planning of multiple DC transmission systems. First, a MAP calculation method is proposed. Based on this, a multi-layer neural network average influence value method is used to identify the determinants of accurate transmission planning. Furthermore, a set of rules is used to construct calculation rules for important planning indicators relative to relevant operational variables and long-term planning. Examples demonstrate the feasibility and accuracy of this method in uncovering key factors and potential correlations. Compared with traditional empirical transmission planning strategies, this method achieves more accurate, economical, scientific, and reliable planning results.

[0108] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A non-parametric independent screening method for key factors in the planning of a multi-DC transmission-end system, characterized in that, include The following steps: S1. Obtain the basic indicators of the multiple DC transmission system; S2. Analyze the power stability of the multi-DC transmission system based on its basic performance indicators. S3. Use the average influence value method based on multi-layer neural network to identify the key factors affecting the key indicators of the multi-DC transmission system. Step S3 is as follows: S301: Non-parameter independent screening; Before identifying key impacts, operational data from the multiple DC transmission systems must be prepared in advance; this can be achieved through parameter... The method utilizes optimal point set sampling in the numerical space; it establishes a parameter space using operational constraints and planning methods, with GPS deployed within this space; thus determining the input characteristics as {impedance of each AC line, reactive power limit of the generator, reactive power compensation of the rectifier station, active power generation, reactive power generation, DC current setpoint, rated DC current, rectifier firing angle, load, MOIF, MOESCR}, and the output includes {TOV, MAP, and corresponding DC current}, all of which are directly obtainable; using To indicate input, use Indicates the output, where n Indicates sample size. p and t These represent the dimensions of the input and output, respectively. Furthermore, all data-driven tasks will be broken down into t subtasks, the output of which is: In the following description, y represent y i ; The basic principle of nonparametric independent screening is to find the numerical relationship between a single input feature and the target / output feature through univariate regression, and to quantify the strength of the relationship through the sum of squared residuals. For different DC system indicators, the directly related factors vary. Each related factor can be treated independently to observe the correlation strength between a single factor and the DC system indicator. During the non-parametric independent screening process, the correlation between a certain dimension of the input features and the output indicator is considered independently. Based on the above analysis, the correlation analysis problem is simplified as shown in the following formula: ； In the formula, is the index of an randomly selected branch in the target feature set. Consistent with meaning; let The non-parameter independent screening process is as follows: i) Assume there exists a first j Numerical relationship model between input features and target Then we have: In the formula, It is the correlation error that follows a normal distribution with a mean of 0. ii) Use univariate nonparametric regression to establish an approximate correlation model between each input feature and the output. And calculate the target value estimated by the approximate correlation model: In the formula, This represents the target value predicted using the j-th input feature and an approximate correlation model; noting the strong nonlinearity of TTC, a cubic B-spline function is used to construct... ; iii) Calculate the estimation error of the target for each input feature and the approximate correlation model: iv) For sets The input features are sorted in ascending order, with those appearing earlier contributing more to the target. Input features strongly correlated with the target are then selected using certain criteria, as shown below: S302: Lasso set; The group lasso algorithm is developed based on the minimum absolute contraction and lasso algorithm; it can extract explicit calculation rules for relevant variables; the numerical correlation between inputs and relevant indicators can be explicitly represented as having a constant intercept. Additivity model; Then GL is constructed as having Norm-based structural risk minimizer: The extrapolation residual method is used to eliminate the lasso constraint and solve problems (6)-(8).

2. The non-parametric independent screening of key factors for planning a multi-DC transmission-end system as described in claim 1. The method, characterized in that, The basic indicators in step S1 are as follows: 1) Multi-feeder interaction factor and multi-feeder effective short-circuit ratio; 2) Transient overvoltage: Consider temporary overvoltage caused by commutation failure at the receiving end.

3. The non-parametric independent screening method for key factors in planning a multi-DC transmission-end system according to claim 2, characterized in that, Step S2 is as follows: The maximum power curve (MPC) is an important concept in analyzing the power stability of high-voltage direct current (HVDC) transmission systems. The MPC has an extreme point, which is the maximum value of DC power as DC current increases; this is called the maximum transmittable power (MAP). To ensure the MPC of a specific system can be optimized, the MPC of the i-th rectifier station... , Keeping reactive power compensation constant, and then increasing each current setting value according to the set path, the MPC of the multi-DC transmission system can be plotted. Two strategies for determining the DC current growth path are proposed: one is to keep the DC current other than the selected DC current constant, and the other is to keep the setpoint of all DC currents in the power grid increasing synchronously; the following assumptions are made based on the MPC obtained: #（1） In the formula, This represents the rated current of the i-th DC circuit. This refers to the number of DC feed lines. By combining the DC current gain, we finally obtain MPC and MAP.

Images