Key branch adaptive screening method based on incremental learning in different operation states

Abstract

The invention discloses a key branch adaptive screening method based on incremental learning in different operation states, and relates to the technical field of power system key branch identification, and the method comprises the steps: constructing a safety boundary preliminary fitting model in different load states through unsupervised clustering historical operation data; fitting corresponding security boundaries according to the state data to generate a security boundary set of the whole network; training an incremental learning model by using historical operation state data and the fitted security boundary, and outputting a plurality of out-of-limit branches; and judging whether the security boundary of the branch is included into the sample pool based on the confidence score. According to the method, the incremental learning and the online key branch identification technology are combined, so that real-time generation and self-adaptive updating of the safety boundary of the power system are realized. The self-adaptive updating mechanism enables the model to be continuously optimized in a complex operation state, accurately identifies key branches which may be out-of-limit, and helps a power grid operator to predict and prevent safety risks in advance, thereby improving the safety, reliability and scheduling efficiency of a power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of identification of key branches in power systems, and specifically to an adaptive screening method for key branches based on incremental learning under different operating states. Background Art

[0002] In modern power systems, with the continuous expansion of the grid scale and the increasing complexity of operating states, how to quickly and accurately identify and evaluate the over-limit risks of key branches has become an important task to ensure the safe operation of the grid. The method of screening key branches or sections through historical data has poor adaptability to the real-time state changes of the grid and is difficult to effectively handle complex and variable operating conditions. Especially in a power system with multiple branches and multiple load states, how to accurately identify potential key branches based on real-time operation data and quickly evaluate their safety remains a technical problem.

[0003] To solve this problem, the incremental learning method has been proposed as an adaptive learning strategy that can gradually update the model when new data arrives, thereby enhancing the real-time adaptation ability of the model in a dynamic environment. The key branch screening method based on incremental learning can not only adjust the boundary model according to the real-time operating state, but also effectively identify key branches that may be over-limit and timely discover potential safety hazards. In addition, the adaptive update ability of the incremental learning model enables it to continuously improve the prediction accuracy under different power system states, reduce the dependence on traditional offline modeling and data preprocessing, and thus realize the real-time and intelligent evaluation of power system security. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing branch screening methods simply rely on historical data, lack real-time performance and adaptability, and are difficult to effectively handle the safety evaluation problem under complex grid operating states.

[0006] To solve the above technical problem, the present invention provides the following technical solution: an adaptive screening method for key branches based on incremental learning under different operating states, including constructing a preliminary fitting model of the safety boundary under different load states by unsupervised clustering of historical operation data; fitting the corresponding safety boundary according to the state data to generate a set of safety boundaries for the whole network; training an incremental learning model using historical operation state data and the fitted safety boundary, and outputting multiple over-limit branches; and judging whether to include the safety boundary of the branch in the sample pool based on a confidence score.

[0007] As a preferred solution of the key branch adaptive screening method based on incremental learning under different operating states of the present invention, wherein: the construction of the preliminary fitting model of the safety boundary under different load states by unsupervised clustering of historical operation data includes obtaining historical data of different operating states of all lines in the power grid through offline research and statistics according to the power grid topology structure, including the active power output distribution P of each node i and the power flow distribution P b,j , and constructing low-dimensional safety boundaries of each boundary of the power grid by fitting through historical data by partial least squares method. The boundary is expressed as:

[0008]

[0009] wherein, B represents the set of all nodes and branches, represents the limit of the active power flow passing through, P i represents the active power injected by node i, β ij represents the node-branch power flow sensitivity, that is, the hyperplane coefficient and the constant term c j is directly obtained by simplifying the DC power flow.

[0010] As a preferred solution of the key branch adaptive screening method based on incremental learning under different operating states of the present invention, wherein: the generation of the safety boundary set of the whole network by fitting the corresponding safety boundary according to the state data includes defining the gap between the actual data and the fitting hyperplane as the fitting plane error according to the safety boundary expression, which is expressed as:

[0011]

[0012] wherein, r represents the fitting plane error, β represents the hyperplane coefficient, c represents the hyperplane constant term, P i represents the active injection of the node, [P i , P b,j , r] represents the input data set, P i represents the active injection of the key node, P b,j represents the active power flow of the branch;

[0013] Conduct DBSCAN clustering analysis on the operation data according to different line loads and biases;

[0014] Define the minimum number of points MinPts and the neighborhood radius Eps, and check the Eps neighborhood of each point;

[0015] If the number of points contained in the neighborhood exceeds MinPts, form a cluster around the core, and directly reach and iteratively aggregate from the core point to merge the clusters;

[0016] If no new points are added to any cluster, the process ends.

[0017] As a preferred solution of the key branch adaptive screening method based on incremental learning under different operating states according to the present invention, wherein: training the incremental learning model by using the historical operating state data and the fitted safety boundary, and outputting multiple over-limit branches includes constructing a safety boundary set, and fitting the whole network safety boundary by least square classification;

[0018] The safety boundary is expressed as:

[0019]

[0020] Wherein, α ij represents the coefficient of non-critical nodes;

[0021] To distinguish different operating states belonging to the same boundary, define the load rate γ j , which reflects the load conditions of the lines corresponding to each boundary, and strengthen the construction of the boundary set, and is expressed as:

[0022]

[0023] As a preferred solution of the key branch adaptive screening method based on incremental learning under different operating states according to the present invention, wherein: training the incremental learning model by using the historical operating state data and the fitted safety boundary, and outputting multiple over-limit branches includes, based on the constructed boundary data set, using the boundary as the training label, and using the active power injection of the nodes and the load rate of the whole network as the training input;

[0024] Extract the first batch of training data to initialize the probability prediction model. The prediction adopts error-correcting output codes, and ECOC constructs a coding matrix M with dimensions of K×L, where K is the number of categories and L is the number of binary classifiers. Each column of the matrix M represents a different binary classifier, and is expressed as:

[0025]

[0026] Wherein, M KL represents the K-category coding of the binary classifier L;

[0027] Each column divides multiple categories into two groups, and trains a binary classifier for each group;

[0028] In the prediction stage, the outputs of all binary classifiers will form a prediction vector Then compare the prediction vector with each row of the coding matrix M to determine the final classification result;

[0029] The prediction result is expressed as:

[0030]

[0031] As a preferred embodiment of the key branch adaptive screening method based on incremental learning under different operating states of the present invention, wherein: training the incremental learning model using historical operating state data and the fitted safety boundary, and outputting multiple over-limit branches includes using the second batch of samples to initialize the relatively ineffective source model, where x represents historical operating state data, y represents the fitted safety boundary, and M represents the number of initialization loops;

[0032] After the next batch of samples is completed, after fitting the historical data, introduce it into the sample pool, where N represents the number of subsequent incremental learning loops;

[0033] Train the source model using incremental samples, and the incremental update is performed through continuous gradient descent. The gradient update is expressed as:

[0034]

[0035] where, |D t | represents the number of samples, represents the gradient of the loss function L with respect to the parameter θ vector in the model, η represents the learning rate, which controls the size of the update step, and h θ (x (i) ) represents the predicted value of the model on the input x (i) ;

[0036] Using the confidence matrix H to comprehensively evaluate and rank the confidence levels of each boundary is expressed as:

[0037]

[0038] where, m kj represents the element (k,j) of the coding design matrix M, the coding corresponding to the k-th class of the j-th binary classifier, and s j represents the score of the j-th binary classifier in the online state, and g represents the binary loss function;

[0039] g = log[1 + exp(-2y j s j )] / [2log(2)].

[0040] As a preferred embodiment of the key branch adaptive screening method based on incremental learning under different operating states of the present invention, wherein: determining whether to include the safety boundary of the branch in the sample pool based on the confidence score includes adaptively updating the model using the label results;

[0041] Introduce the confidence difference Margin and the information entropy evaluation Entropy, which are expressed as:

[0042] Margin = Max H - Second Highest H

[0043] Entropy = -∑H(x)logH(x).

[0044] Another object of the present invention is to provide a key branch adaptive screening system based on incremental learning under different operating states, which can train an incremental learning model by using historical operating state data and the fitted safety boundary, and output multiple over-limit branches, solving the problem that the current branch screening contains x and is difficult to effectively process the safety assessment under complex power grid operating states.

[0045] As a preferred solution of the key branch adaptive screening system based on incremental learning under different operating states described in the present invention, it includes: an offline data analysis and safety boundary modeling module, an offline training incremental learning model construction module, and an online adaptive model update and real-time optimization module; the offline data analysis and safety boundary modeling module is used to describe the safety margin of the power grid under different operating states, and provide data and a safety boundary model for model training; the offline training incremental learning model construction module is used to identify the branches in the power grid; the online adaptive model update and real-time optimization module is used to continuously optimize and update the model during the real-time operation of the power grid.

[0046] A computer device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it realizes the steps of the key branch adaptive screening method based on incremental learning under different operating states.

[0047] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, it realizes the steps of the key branch adaptive screening method based on incremental learning under different operating states.

[0048] Advantages of the present invention: The key branch adaptive screening method based on incremental learning under different operating states provided by the present invention realizes the real-time generation and adaptive update of the power system safety boundary by combining incremental learning and online key branch identification technology. Compared with traditional methods, the present invention can dynamically adjust decisions according to real-time operating states, improving the accuracy and flexibility of power grid safety assessment. Its adaptive update mechanism enables the model to continuously optimize under complex operating states, accurately identify key branches that may exceed the limit, helping power grid operators predict and prevent safety risks in advance, thereby improving the safety, reliability and dispatching efficiency of the power grid. Description of the Drawings

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0050] Figure 1 It is the overall flowchart of a key branch adaptive screening method based on incremental learning under different operating states provided for the first embodiment of the present invention.

[0051] Figure 2 It is the operating state classification diagram of the 16 branches of the IEEE 39-node system for a key branch adaptive screening method based on incremental learning under different operating states provided for the second embodiment of the present invention.

[0052] Figure 3 It is the classification fitting boundary error result diagram of the 16 branches of the IEEE 39-node system for a key branch adaptive screening method based on incremental learning under different operating states provided for the second embodiment of the present invention.

[0053] Figure 4 It is the online operating state incremental learning cumulative error and window error diagram of the IEEE 39-node system for a key branch adaptive screening method based on incremental learning under different operating states provided for the second embodiment of the present invention.

[0054] Figure 5 It is the sample label comparison diagram selected in the adaptive update stage of the IEEE 39-node system for a key branch adaptive screening method based on incremental learning under different operating states provided for the second embodiment of the present invention. Specific Embodiments

[0055] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0056] Embodiment 1, referring to Figure 1 , which is an embodiment of the present invention, provides a key branch adaptive screening method based on incremental learning under different operating states, including:

[0057] S1: Construct a preliminary fitting model of the safety boundary under different load states by unsupervised clustering of historical operation data.

[0058] Furthermore, according to the topology of the power system, historical data of the whole network lines under different operating states are obtained through off-line research and statistics, including the active power output distribution P of each node i and the power flow distribution P b,j , and the low-dimensional safety boundaries of each boundary of the power grid are constructed by fitting through partial least squares method using historical data. The boundary expression formula is as follows:

[0059]

[0060] Among them, B represents the set of all nodes and branches, represents the limit of the active power flow passing through, P i represents the active power injected by node i. β ij Node-branch power flow sensitivity, that is, the hyperplane coefficient and the constant term c j are directly obtained by simplifying the DC power flow.

[0061] S2: Generate the set of safety boundaries of the whole network according to the safety boundaries fitted corresponding to the state data.

[0062] Furthermore, in order to study the correlation between the active power flow between lines and the active power injection of key nodes, according to the safety boundary expression, the gap between the actual data and the fitted hyperplane is defined as r, where β is the hyperplane coefficient and c is the hyperplane constant term, and P i is the active power injection of the node:

[0063]

[0064] Taking the active power injection P of the key node i , the active power flow of the branch P b,j and the fitting plane error r, [P i , P b,j , r] as the input data set. Perform DBSCAN clustering analysis on the operation data according to different line loads and biases. The DBSCAN clustering algorithm has the advantages of strong noise processing ability, no need to preset the number of clusters in advance, strong adaptability, high efficiency, etc., and is suitable for analyzing complex power systems. First of all, DBSCAN needs to define the minimum number of points MinPts and the neighborhood radius Eps. DBSCAN will check the Eps neighborhood of each point; if the number of points contained in the neighborhood exceeds MinPts, a cluster will be formed around the core object. The objects directly reachable from the core points will be iteratively aggregated, and the clusters may be merged. When no new points are added to any cluster, the process ends.

[0065] S3: Train the incremental learning model using historical operating state data and the fitted safety boundaries, and output multiple over-limit branches.

[0066] Furthermore, a safety boundary set is constructed. According to the above results, the least squares classification is used to fit the overall network safety boundary. Since the variables of the entire boundary set must be unified for subsequent analysis, unifying the variables requires expanding the boundary from the low dimension that only focuses on the sensitivity nodes to the high dimension. Given the extensiveness of most power systems, the coefficients of non-critical nodes may be very small or even zero. Therefore, the safety boundary can be expressed as follows:

[0067]

[0068] where α ij is the coefficient of the non-critical node. At the same time, in order to distinguish different operating states belonging to the same boundary, the load rate γ j is also defined to reflect the load conditions of the lines corresponding to each boundary and strengthen the construction of the boundary set.

[0069]

[0070] It should be noted that based on the previously constructed boundary data set, the boundary is used as the training label, and the active power injection of the nodes and the load rate of the entire network are used as the training input. The first batch of training data is extracted to initialize the probability prediction model. In order to improve the flexibility and accuracy of the model in dealing with high-dimensional analogy labels, the error-correcting output code (ECOC) method is adopted for prediction. The core idea of ECOC is to convert the multi-class classification problem into multiple binary classification problems through the coding matrix. ECOC constructs a coding matrix M with the dimension of K×L, where K is the number of classes and L is the number of binary classifiers. Each column of the matrix M represents a different binary classifier.

[0071]

[0072] In this method, M KL represents the K-class coding of the L binary classifiers. Each column divides multiple classes into two groups, and a binary classifier is trained for each group. In the prediction stage, the outputs of all binary classifiers will form a prediction vector Then this prediction vector is compared with each row of the coding matrix M to determine the final classification result. The prediction result can be expressed as:

[0073]

[0074] Furthermore, the second batch of samples (referred to as source samples) are used to initialize the relatively ineffective source model, where the incremental learning model is trained using the historical operating state data x and the fitted safety boundary y, and multiple over-limit branches are output. M is the number of initialization loops. After the next batch of samples is completed, after fitting the historical data, a part of the new samples (incremental samples) Introduce a sample pool. Then, use these incremental samples to train the source model in a way similar to the initialization training process of the ECOC model. The incremental update is carried out through continuous gradient descent to ensure that the parameters of the source model are retained. The gradient update formula is as follows:

[0075]

[0076] where, |D t | represents the number of samples, represents the gradient of the loss function L with respect to the parameter vector θ in the model, η is the learning rate, which controls the size of the update step, and h θ (x (i) ) is the predicted value of the model for the input x (i) .

[0077] It should be noted that after being trained with a large amount of historical data for online status update, the prediction model should be able to directly identify potential over-limit branches under various conditions. However, since there may be multiple branches with a high probability of over-limit under a given status, only selecting the boundary with the highest probability may not provide the most reliable label. To better evaluate potential over-limit situations, a confidence matrix H can be used to comprehensively evaluate and rank the confidence levels of each boundary. Finally, the model is based on

[0078]

[0079] where, m kj represents the element (k,j) of the coding design matrix M, the coding corresponding to the k-th class of the j-th binary classifier, and s j represents the score of the j-th binary classifier in the online status, and g represents the binary loss function.

[0080] g = log[1 + exp(-2y j s j )] / [2log(2)]

[0081] S4: Determine whether to include the safety boundary of the branch in the sample pool based on the confidence score.

[0082] Furthermore, for the adaptive update of the incremental learning model: It is challenging to perform adaptive update of the model using these multi-label results. This step requires effectively selecting samples with single-boundary over-limit, that is, samples with a single label, for the incremental learning model to correct. To solve this problem, two metrics are introduced: confidence difference and information entropy to evaluate the relationship between the highest probability and other values. The confidence difference Margin and the information entropy evaluation Entropy are introduced, expressed as

[0083] Margin = Max H - Second Highest H

[0084] Entropy = -∑H(x)logH(x)

[0085] Finally, the prediction model that realizes potential boundary violation behavior can be adaptively updated to ensure a rapid response to new application environments.

[0086] Example 2, referring to Figure 2 - Figure 5 , which is an embodiment of the present invention, provides a method for adaptively screening key branches based on incremental learning under different operating states. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0087] First of all, taking the IEEE 39 - bus system as an example, this study demonstrates the necessity of considering safety boundaries under different operating conditions. The line No. 6 - 11 is selected as the monitoring line, and the key nodes 12 and 10 are determined through node - branch power flow sensitivity analysis. For the convenience of visualization, the thermal stability safety boundary is reduced to a two - dimensional space and clustering analysis is carried out under different operating states. According to the S2 clustering results as Figure 2 shown. As shown in Figure 2 , this gap shows obvious regularity with the growth trend of power flow: initially, the gap decreases from light load to 0 as the power flow increases, then rises to a local peak and then decreases again, and then rises again under heavy load after another gap. The clustered data is classified and fitted into the safety boundary. As shown in Figure 3 , it clearly shows the influence of different data on the safe area. The boundary fitted by light - load data has a significant deviation from the actual thermal stability limit, while the boundary fitted by heavy - load data is highly consistent with the limit. This is because heavy - load data is closer to the limit point, making it more interpretable. In order to more intuitively evaluate the performance of the model, the present invention adopts two indicators: cumulative error and window error. The cumulative error summarizes all prediction errors since the initial stage of the incremental learning process; while the window error calculates the error of the model within a specified recent time range. The training and test results are as shown in Figure 4 . As shown by the blue line, the cumulative error emphasizes the overall performance of the model, showing that the error gradually decreases and tends to be stable after the initial stage and remains stable with the input of subsequent data in the long term. The window error focuses on the performance of the model within the current window; as shown by the pink line, the model shows small fluctuations in the window error, proving its ability to quickly respond to the dynamic changes of data. In order to verify the screening results of step 6, the present invention mixes the known out - of - limit label data set B with the normal data set A. The results show that all selected samples come from data set B. 10 samples are sampled and their confidence levels are compared with the classification results, as shown in Figure 5 .

[0088] Embodiment 3, an embodiment of the present invention, provides a key branch adaptive screening system based on incremental learning under different operating states, including an offline data analysis and security boundary modeling module, an offline training incremental learning model construction module, and an online adaptive model update and real-time optimization module.

[0089] Among them, the offline data analysis and security boundary modeling module is used to describe the security margin of the power grid under different operating states, and provide data and a security boundary model for model training. The offline training incremental learning model construction module is used to identify the branches in the power grid, and the online adaptive model update and real-time optimization module is used to continuously optimize and update the model during the real-time operation of the power grid.

[0090] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0091] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or in combination with these instruction execution systems, apparatuses, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0092] More specific examples (non-exhaustive list) of computer-readable media include the following: electrical connections (electronic devices) having one or more wirings, portable computer diskettes (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable media can even be paper or other suitable media on which a program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing it as appropriate, and then storing it in a computer memory.

[0093] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An adaptive screening method for key branches based on incremental learning under different operating states, characterized in that, including: Constructing a preliminary fitting model of the safety boundary under different load states by unsupervised clustering of historical operation data; Generating a set of safety boundaries for the entire network by fitting the corresponding safety boundaries according to the state data; Training the incremental learning model using the historical operation state data and the fitted safety boundaries, and outputting multiple over-limit branches; Judging whether to include the safety boundary of the branch in the sample pool based on the confidence score.

2. The key branch adaptive screening method based on incremental learning under different operating states according to claim 1, characterized in that: The construction of the preliminary fitting model of the safety margin under different load conditions by unsupervised clustering of historical operation data includes obtaining historical data of different operation conditions of the entire power grid line through offline survey statistics according to the topology of the power system, including the active output distribution P of each node i and power flow distribution P b,j And the low-dimensional safety boundaries of each boundary of the power grid are constructed by fitting historical data through partial least squares method. The boundaries are expressed as: Among them, B represents the set of all nodes and branches, represents the limit of the active power flow passing through, P i represents the active power injected by node i, β ij represents the node-branch power flow sensitivity, that is, the hyperplane coefficient and the constant term c j is directly obtained by simplifying the DC power flow.

3. The key branch adaptive screening method based on incremental learning under different operating states according to claim 2, characterized in that: The generating a set of safety boundaries for the entire network by fitting the corresponding safety boundaries according to the state data includes defining the gap between the actual data and the fitting hyperplane as the fitting plane error according to the safety boundary expression, which is expressed as: Among them, r represents the fitting plane error, β represents the hyperplane coefficient, c represents the hyperplane constant term, and P i represents the active power injection of the node, [P i ,P b,j ,r] represents the input data set, and P i represents the active power injection of the key node, and P b,j represents the active power flow of the branch; Performing DBSCAN clustering analysis on the operation data according to different line loads and biases; Defining the minimum number of points MinPts and the neighborhood radius Eps, and checking the Eps neighborhood of each point; If the number of points included in the neighborhood exceeds MinPts, forming a cluster around the core, directly reaching and iteratively aggregating from the core point, and merging the clusters; If no new points are added to any cluster, the process ends.

4. The key branch adaptive screening method based on incremental learning under different operating states according to claim 3, characterized in that: The training the incremental learning model using the historical operation state data and the fitted safety boundaries and outputting multiple over-limit branches includes constructing a safety boundary set and fitting the safety boundaries of the entire network through least squares classification; The safety boundary is expressed as: Among them, α ij represents the non-critical node coefficient; Distinguish different operating states belonging to the same boundary and define the load factor γ j , reflecting the load conditions of the lines corresponding to each boundary, and strengthening the construction of the boundary set, which is expressed as:

5. The key branch adaptive screening method based on incremental learning under different operating states according to claim 4, characterized in that: The training the incremental learning model using the historical operation state data and the fitted safety boundaries and outputting multiple over-limit branches includes, based on the constructed boundary data set, using the boundary as the training label and using the active power injection of the node and the load rate of the entire network as the training input; Extracting the first batch of training data to initialize the probability prediction model. The prediction adopts the error-correcting output code, and ECOC constructs the coding matrix M with the dimension of K×L, where K is the number of categories and L is the number of binary classifiers. Each column of the matrix M represents a different binary classifier, which is expressed as: Among them, M KL represents the K-category encoding of the binary classifier L; Each column divides multiple categories into two groups, and trains a binary classifier for each group; During the prediction phase, the outputs of all binary classifiers form a prediction vector Then the prediction vector is compared with each row of the encoding matrix M to determine the final classification result; The prediction result is expressed as:

6. The key branch adaptive screening method based on incremental learning under different operating states according to claim 5, characterized in that: Training the incremental learning model using the historical operation status data and the fitted safety boundary, and outputting multiple over-limit branches including using the second batch of samples to initialize the relatively ineffective source model, where x represents the historical operation status data, y represents the fitted safety boundary, and M represents the number of initialization loops; After the next batch of samples is completed and the historical data is fitted, is introduced into the sample pool, where N represents the number of subsequent incremental learning cycles; Training the source model using the incremental samples, and the incremental update is performed through continuous gradient descent. The gradient update is expressed as: where, |D t | represents the number of samples, represents the gradient of the loss function L with respect to the vector of parameters θ in the model, η represents the learning rate, which controls the size of the update step, and h θ (x (i) ) represents the predicted value of the model for the input x (i) ; Using the confidence matrix H to comprehensively evaluate and rank the confidence of each boundary, which is expressed as: where m kj represents the element (k, j) of the coding design matrix M, the coding corresponding to the k-th class of the binary classifier j, s j represents the score of the binary classifier j in the online state, and g represents the binary loss function; g = log[1 + exp(-2y j s j )] / [2log(2)]。 7. The key branch adaptive screening method based on incremental learning under different operating states according to claim 6, wherein: The judging whether to include the safety boundary of the branch in the sample pool based on the confidence score includes adaptively updating the model using the label result; Introducing the confidence difference Margin and the information entropy evaluation Entropy, which are expressed as: Margin = Max H - Second Highest H Entropy = -∑H(x)logH(x).

8. A system adopting the key branch adaptive screening method based on incremental learning under different operating states as described in any one of claims 1 to 7, characterized in that: including an offline data analysis and safety boundary modeling module, an offline training incremental learning model construction module, and an online adaptive model update and real-time optimization module; The offline data analysis and safety boundary modeling module is used to describe the safety margin of the power grid under different operation states and provide data and a safety boundary model for model training; The offline training incremental learning model construction module is used to identify the branches in the power grid; The online adaptive model update and real-time optimization module is used to continuously optimize and update the model during the real-time operation of the power grid.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for adaptively screening critical branches based on incremental learning in different operating states described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for adaptively screening critical branches based on incremental learning in different operating states described in any one of claims 1 to 7.

Images