Power system voltage stability assessment method based on improved cascade forest

Abstract

A power system voltage stability evaluation method based on an improved cascade forest comprises the steps that 1, a fast voltage stability index FVSI is constructed to describe the voltage stability state of a power system, and the voltage stability margin of the power system is obtained; step 2, representing a mapping relation between an input variable and an output variable by adopting an improved cascading forest Impressed Cascade Forest model, wherein the mapping relation between the input variable and the output variable is represented by adopting an improved cascading forest Impressed Cascade Forest model; step 3, an improved zebra optimization algorithm (Imported Zebra Optimization Algorithm) is adopted to carry out optimization on the improved cascade forest model; 4, constructing a data set of the IZOA improved ICF model by adopting an offline power flow method, and preprocessing the data to generate an efficient data set; and 5, dividing the efficient data set into a training set and a test set, training the ICF model optimized by using the IZOA by using the training set, and testing the model by using the test set. The method is suitable for a novel power system with new energy access, and has important significance for guaranteeing safe and stable operation of the power system.

Description

Technical Field

[0001] The present invention relates to the field of power system voltage stability and new energy voltage, and in particular to a power system voltage stability assessment method based on improved cascade forest. Technical Background

[0002] In recent years, as energy problems have become increasingly serious, green and low-carbon energy represented by wind and solar energy has been vigorously developed. The access to new energy sources such as wind and solar energy has posed a higher challenge to the voltage stability of the power system. The voltage stability issue is one of the important considerations in the planning, operation and design of the power system. Continuous disturbances in the power system will cause voltage instability and sometimes even cause failure of the entire system. The main reason for the voltage instability of the power system is the insufficient supply of reactive power or the excessive occupation of reactive power by the system itself. In order to ensure the safe operation of the system, it is of great significance to evaluate the voltage stability of the system.

[0003] The document with application publication number CN108988320A discloses a method for analyzing the impact of the response characteristics of dynamic components of a power system on voltage stability. From the perspective of generalized branch potential energy, a new analysis method is proposed to study the impact of generators, high-voltage direct current transmission systems and induction motors on voltage stability. According to the changing law of transient potential energy distribution in the network after the fault and the information of stability margin, the reactive power recovery characteristics of each dynamic component after the fault are used as the basic variable, and an evaluation index is established to analyze the mechanism causing transient voltage instability in the power grid; the document with application publication number CN110233476A discloses a method for evaluating voltage stability during black start and related systems. The specific steps are: analyzing several influencing factors affecting voltage stability during the black start process and sorting them according to the degree of influence, screening out necessary elements as feature inputs of the support vector machine model, and selecting several typical values ​​for each feature input to form a sample set. The support vector machine model is trained with the sample set to obtain the mapping relationship between feature input and output, and a voltage stability evaluation model is constructed. The voltage stability is evaluated according to the actual parameters and voltage stability evaluation model during the black start process; the document with application publication number CN114626757A discloses a method and system for determining the voltage stability of the receiving system. The specific steps are: obtaining the transient voltage stability data of the busbar, and segmenting the voltage curve using the differential equation. In combination with the existing practical criteria for transient voltage, different weights are assigned according to the degree of voltage drop. The cumulative amount of drop is quantified using the weighted integral method. An evaluation index is constructed and a complete determination method is formed.

[0004] The purpose of the present invention is to solve the technical problems of low accuracy, complex calculation and long time consumption in power system voltage stability assessment existing in technical literature, introduce artificial intelligence algorithm into the field of power system voltage stability assessment, and use improved zebra optimization algorithm to optimize the improved cascade forest model, and propose a power system voltage stability assessment method based on improved cascade forest. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problems of low accuracy, complex calculation and long time consumption in power system voltage stability assessment existing in technical literature, introduce artificial intelligence algorithm into the field of power system voltage stability assessment, and use improved zebra optimization algorithm to optimize the improved cascade forest model, and propose a power system voltage stability assessment method based on improved cascade forest.

[0006] The technical solution adopted by the present invention is:

[0007] A voltage stability assessment method based on an improved cascade forest includes the following steps:

[0008] Step 1: Construct a Fast Voltage Stability Index (FVSI) to describe the voltage stability state of the power system and obtain the voltage stability margin of the power system;

[0009] Step 2: Use the improved cascade forest (ICF) model to represent the mapping relationship between input variables and FVSI;

[0010] Step 3: Use the improved zebra optimization algorithm (IZOA) to optimize the ICF model;

[0011] Step 4: Use the offline power flow method to construct the data set of the IZOA improved ICF model, preprocess the data, and generate an efficient data set;

[0012] Step 5: ,The efficient data set is divided into a training set and a test set. The ICF model optimized by IZOA is ,trained using the training set, and the model is tested using the ,testing set to determine the voltage stability of the power system.

[0013] In step 1: construct FVSI to describe the voltage state of the power system and obtain the voltage stability margin of the power system, which specifically includes the following steps:

[0014] Step 1-1: Construct FVSI to describe the voltage status of the power system;

[0015] Voltage stability index can be used to study the voltage stability of the power system. FVSI is used to detect key buses in the power system. The FVSI calculation formula is as follows:

[0016]

[0017] Where: Z sr is the line impedance from the transmitter to the receiver; Q r is the reactive power at the receiving end; V s is the voltage at the transmitting end; X sr is the line reactance from the transmitter to the receiver.

[0018] Step 1-2: Obtain the power system voltage stability margin based on FVSI.

[0019] In order to ensure stable operation of the power system, the FVSI value of the power system needs to be kept in a range far less than 1.

[0020] When the FVSI value is close to 1, it indicates that the line is close to instability, which may cause voltage collapse of the power system in severe cases.

[0021] In step 2: the ICF model is used to represent the mapping relationship between the input variables and the voltage stability margin, which specifically includes the following steps:

[0022] Step 2-1: Improve the traditional Cascade Forest (CF) model to obtain the ICF model;

[0023] Improve the CF model structure and increase the types of base learners. Integrate CatBoost, XGBoost, Random Forest and LightGBM learners to get the ICF model.

[0024] Step 2-2: Represent the mapping relationship between input variables and FVSI based on the ICF model.

[0025] Input variables such as bus voltage amplitude, generator active and reactive power are input into the ICF model. CatBoost, XGBoost, random forest and LightGBM learners are used to train the model in each layer to obtain the prediction probability. These probabilities are added as new features to the input variables of the next layer. Feature conversion and fusion are performed at each layer of the model. Finally, the accurate FVSI can be continuously optimized through the multi-level cascade model.

[0026] In step 3, IZOA is used to optimize the ICF model, which specifically includes the following steps:

[0027] Step 3-1: Set the algorithm parameters, including the number of zebra populations, iterative convergence conditions, the minimum and maximum values ​​of the parameters to be optimized in the ICF model, etc.

[0028] Step 3-2: Introduce Elite Opposition-Based Learning (EOBL) to initialize the population: First, randomly generate a population of zebras, then compare the fitness function values ​​of the existing zebras with the fitness function values ​​of the elite opposition zebras, and select the zebras with lower fitness function values ​​as the next generation;

[0029] Each zebra i is defined as a position vector Z i =(z i,1 ,···,z i,j ,···,z i,m ), m is the hyperparameter value to be optimized in the ICF model in step 2, and the fitness function formula is as follows:

[0030] F(Z i )=min(1-Acc CF ) (9)

[0031] Where: Acc CF Indicates the accuracy of the ICF model.

[0032] The zebra with the lowest fitness function value in the population is defined as the pioneer zebra. During the optimization process, the pioneer zebra guides other zebras to its position.

[0033] EOBL is an optimization strategy that constructs a reverse population by using elite individuals to increase the diversity of the population. i , whose reverse zebra representation is

[0034]

[0035] Where r is a random number between 0 and 1; LB is the lower bound of the elite zebra; UB is the upper bound of the elite zebra.

[0036] Step 3-3: Position update during foraging phase: Use formula (3) to calculate the new position of the zebra, and use formula (10) to update the position;

[0037] In the first stage, the individuals in the population are updated by simulating the behavior of zebras in the foraging process. The best individual in the population is called the pioneer zebra, which guides the rest of the individuals to move closer to its position in the search area. The formula for updating the zebra position is as follows:

[0038]

[0039]

[0040] Where: is the updated position of zebra i in the jth dimension after foraging behavior; z i,j is the position of zebra i in the jth dimension; MZ j is the optimal position of the jth dimension; r is a random number between 0 and 1; l∈{1,2}, the larger the l value, the faster the population moves; The new position of the i-th horse for foraging behavior; Z i is the actual position of zebra i; F i ' P1 is the fitness value at the target position of foraging behavior; F i is the fitness value before moving.

[0041] Step 3-4: Defense phase position update: according to the random number p s The value of is used to calculate the new position of the zebra using equation (3), and the position is updated using equation (11).

[0042] In the second stage, the population position is updated using the zebra's defensive behavior against predators. Facing different predators, the zebra's defensive behavior is different: when attacked by a lion, the zebra will choose an escape strategy as shown in equation (4) S1; when facing other predators, the zebra will choose an attack strategy as shown in equation (4) S2.

[0043]

[0044]

[0045] Where: is the updated position of zebra i in the jth dimension after the defensive behavior; The new position of the i-th horse for defensive behavior; F i ' P2 is the fitness value at the target position after the defense behavior; t and T are the current iteration number and the total number of iterations respectively; k is a constant of 0.01; P s is a random number between [0,1]; NZ j is the optimal position of the jth dimension.

[0046] Step 3-5: Introduce the Firefly Disturbance Strategy (FDS) to update the zebra position again: Use formula (5) to calculate the new position of the zebra, then compare the fitness function value of the existing zebra with the fitness function value of the pioneer zebra, and select the zebra with the lower fitness function value as the next generation zebra;

[0047]

[0048]

[0049] Where β0 and θ are constants; S is the spatial distance between the zebra and the pioneer zebra; δ is a step factor in the interval [0,1].

[0050] Step 3-6: Verify whether the iteration limit is met: If the maximum number of iterations is not reached, return to step 3-2 to continue iterating. If it is reached, the optimization process ends.

[0051] In step 4: an offline power flow method is used to construct a data set for the IZOA improved ICF model, and the data is preprocessed to generate an efficient data set, which specifically includes the following steps:

[0052] Step 4-1: Normalize the data based on the MAXABS method;

[0053] The MAXABS normalization method rescales the data x to the range of [-1,1], and the calculation formula is as follows:

[0054]

[0055] In the formula: x is data; |x max | is the maximum absolute value of the data.

[0056] Step 4-2: Use the Fisher Score algorithm to filter the original data to obtain an efficient data set.

[0057] Step 4-2-1: Mark the state of the data normalized in step 4-1, and divide the data into a training data set and a test data set according to a certain ratio;

[0058] Step 4-2-2: Use Fisher Score algorithm to calculate the importance of electrical features in the original data;

[0059] Assume that the number of stable and unstable samples in all electrical characteristic samples is K s and K m , use the Fisher Score algorithm to calculate the sample a-th dimension feature score F a The formula is as follows:

[0060]

[0061] Where: m n , and are the means of the sample's a-th dimension feature in the entire data set, in the stable sample data set, and in the unstable sample set, respectively; is the eigenvalue of the a-th dimension feature of the i-th stable sample data; is the eigenvalue of the a-th dimension feature of the i-th unstable sample data.

[0062] Step 4-2-3: Arrange all features in descending order according to Fisher Score value;

[0063] Step 4-2-4: Add the features to the training data set in order to train the ICF model, and obtain the feature dimension when the model accuracy is the highest through comparison.

[0064] In step 5: the efficient data set is divided into a training set and a test set, the ICF model optimized by IZOA is trained using the training set, and the model is tested using the test set to determine the voltage stability of the power system.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. In order to solve the impact of new energy on the voltage stability of the power system, the present invention adopts the ICF model and proposes a simple, fast and computationally feasible method for evaluating the voltage stability of the power system;

[0067] 2. The present invention uses the Fisher Score algorithm to screen the original data to obtain an efficient data set. The Fisher Score algorithm is a statistical method for feature selection, which is simple to understand, has high computational efficiency and wide adaptability;

[0068] 3. The present invention improves the zebra optimization algorithm and adopts the EOBL initialization population to improve the global search ability, convergence accuracy and speed of the optimization algorithm while avoiding falling into the local optimum, accelerating the algorithm search ability, and adopting FDS to update the zebra position, thereby improving the algorithm search ability and avoiding premature convergence. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0070] Figure 1 is a flow chart of the method of the present invention;

[0071] Figure 2 It is the improved cascade forest model structure of the present invention;

[0072] Figure 3 It is a flow chart of the improved zebra optimization algorithm of the present invention;

[0073] Figure 4 It is a schematic diagram of the IEEE 14-node system of the present invention. DETAILED DESCRIPTION

[0074] The present invention provides a method for evaluating voltage stability of a power system based on an improved cascade forest, the method comprising the following steps: Figure 1 As shown:

[0075] Step 1: Construct a Fast Voltage Stability Index (FVSI) to describe the voltage stability state of the power system and obtain the voltage stability margin of the power system;

[0076] Step 2: Use the improved cascade forest (ICF) model to represent the mapping relationship between input variables and FVSI;

[0077] Step 3: Use the improved zebra optimization algorithm (IZOA) to optimize the ICF model;

[0078] Step 4: Use the offline power flow method to construct the data set of the IZOA improved ICF model, preprocess the data, and generate an efficient data set;

[0079] Step 5: Divide the efficient data set into a training set and a test set, use the training set to train the ICF model optimized by IZOA, and use the test set to test the model to determine the voltage stability of the power system.

[0080] In step 1: construct FVSI to describe the voltage state of the power system and obtain the voltage stability margin of the power system, which specifically includes the following steps:

[0081] Step 1-1: Construct FVSI to describe the voltage status of the power system;

[0082] Voltage stability index can be used to study the voltage stability of the power system. FVSI is used to detect the key busbars in the power system to obtain the voltage stability margin of the power system. The FVSI calculation formula is as follows:

[0083]

[0084] Where: Z sr is the line impedance from the transmitter to the receiver; Q r is the reactive power at the receiving end; V s is the voltage at the transmitting end; X sr is the line reactance from the transmitter to the receiver.

[0085] Step 1-2: Obtain the voltage stability margin of the power system based on FVSI.

[0086] In order to ensure the stable operation of the power system, the FVSI value of the power system needs to be kept in a range far less than 1. When the FVSI value is close to 1, it indicates that the line is close to instability, which may cause the voltage collapse of the power system in severe cases.

[0087] In step 2: the ICF model is used to represent the mapping relationship between the input variables and the voltage stability margin, which specifically includes the following steps:

[0088] Step 2-1: Improve the traditional Cascade Forest (CF) model to obtain the ICF model;

[0089] Improve the CF model structure and increase the types of base learners. Integrate CatBoost, XGBoost, Random Forest and LightGBM learners to get the ICF model.

[0090] Step 2-2: Represent the mapping relationship between input variables and FVSI based on the ICF model.

[0091] Input variables such as bus voltage amplitude, generator active and reactive power are input into the ICF model. CatBoost, XGBoost, random forest and LightGBM learners are used to train the model in each layer to obtain the prediction probability. These probabilities are added as new features to the input variables of the next layer. Feature conversion and fusion are performed at each layer of the model. Finally, the accurate FVSI can be continuously optimized through the multi-level cascade model.

[0092] In step 3, IZOA is used to optimize the ICF model, complete the training process and find the optimal weights and thresholds, such as Figure 3 As shown, the following steps are included:

[0093] Step 3-1: Set the algorithm parameters, including the number of zebra populations, iterative convergence conditions, the minimum and maximum values ​​of the parameters to be optimized in the ICF model, etc.

[0094] Step 3-2: Introduce Elite Opposition-Based Learning (EOBL) to initialize the population: First, randomly generate a population of zebras, then compare the fitness function values ​​of the existing zebras with the fitness function values ​​of the elite opposition zebras, and select the zebras with lower fitness function values ​​as the next generation;

[0095] Each zebra i is defined as a position vector Z i =(z i,1 ,···,z i,j ,···,z i,m), m is the hyperparameter value to be optimized in the improved CF model in step 2, and the fitness function formula is as follows:

[0096] F(Z i )=min(1-Acc CF ) (9)

[0097] Where: Acc CF Indicates improved CF model accuracy.

[0098] The zebra with the lowest fitness function value in the population is defined as the pioneer zebra. During the optimization process, the pioneer zebra guides other zebras to its position.

[0099] EOBL is an optimization strategy that constructs a reverse population by using elite individuals to increase the diversity of the population. i , whose reverse zebra representation is

[0100]

[0101] Where r is a random number between 0 and 1; LB is the lower bound of the elite zebra; UB is the upper bound of the elite zebra.

[0102] Step 3-3: Position update during foraging phase: Use formula (3) to calculate the new position of the zebra, and use formula (10) to update the position;

[0103] In the first stage, the individuals in the population are updated by simulating the behavior of zebras in the foraging process. The best individual in the population is called the pioneer zebra, which guides the rest of the individuals to move closer to its position in the search area. The formula for updating the zebra position is as follows:

[0104]

[0105]

[0106] Where: is the updated position of zebra i in the jth dimension after foraging behavior; z i,j is the position of zebra i in the jth dimension; MZ j is the optimal position of the jth dimension; r is a random number between 0 and 1; l∈{1,2}, the larger the l value, the faster the population moves; The new position of the i-th horse for foraging behavior; Z i is the actual position of zebra i; F i ′P1 is the fitness value at the target position of foraging behavior; F i is the fitness value before moving.

[0107] Step 3-4: Defense phase position update: according to the random number p s The value of is used to calculate the new position of the zebra using equation (4), and the position is updated using equation (11).

[0108] In the second stage, the population position is updated using the zebra's defensive behavior against predators. Facing different predators, the zebra's defensive behavior is different: when attacked by a lion, the zebra will choose an escape strategy as shown in equation (4) S1; when facing other predators, the zebra will choose an attack strategy as shown in equation (4) S2.

[0109]

[0110]

[0111] Where: is the updated position of zebra i in the jth dimension after the defensive behavior; The new position of the i-th horse for defensive behavior; F i ′P2 is the fitness value at the target position after the defense behavior; t and T are the current iteration number and the total number of iterations respectively; k is a constant of 0.01; P s is a random number between [0,1]; NZ j is the optimal position of the jth dimension.

[0112] Step 3-5: Introduce the Firefly Disturbance Strategy (FDS) to update the zebra position again: Use formula (5) to calculate the new position of the zebra, then compare the fitness function value of the existing zebra with the fitness function value of the pioneer zebra, and select the zebra with the lower fitness function value as the next generation zebra;

[0113]

[0114]

[0115] Where β0 and θ are constants; S is the spatial distance between the zebra and the pioneer zebra; δ is a step factor in the interval [0,1].

[0116] Step 3-6: Verify whether the iteration limit is met: If the maximum number of iterations is not reached, return to step 3-2 to continue iterating. If it is reached, the optimization process ends.

[0117] In step 4: an offline power flow method is used to construct a data set for the IZOA improved ICF model, and the data is preprocessed to generate an efficient data set, which specifically includes the following steps:

[0118] Step 4-1: Normalize the data based on the MAXABS method;

[0119] The MAXABS normalization method rescales the data x to the range of [-1,1], and the calculation formula is as follows:

[0120]

[0121] In the formula: x is data; |x max | is the maximum absolute value of the data.

[0122] Step 4-2: Use the Fisher Score algorithm to filter the original data to obtain an efficient data set.

[0123] Step 4-2-1: Mark the state of the data normalized in step 4-1, and divide the data into a training data set and a test data set according to a certain ratio;

[0124] Step 4-2-2: Use Fisher Score algorithm to calculate the importance of electrical features in the original data;

[0125] Assume that the number of stable and unstable samples in all electrical characteristic samples is K s and K m , use the Fisher Score algorithm to calculate the sample a-th dimension feature score F a The formula is as follows:

[0126]

[0127] Where: m n , and are the means of the sample's a-th dimension feature in the entire data set, in the stable sample data set, and in the unstable sample set, respectively; is the eigenvalue of the a-th dimension feature of the i-th stable sample data; is the eigenvalue of the a-th dimension feature of the i-th unstable sample data.

[0128] Step 4-2-3: Arrange all features in descending order according to Fisher Score value;

[0129] Step 4-2-4: Add the features to the training data set in order to train the WNN model, and obtain the feature dimension when the model accuracy is the highest through comparison.

[0130] In step 5: the efficient data set is divided into a training set and a test set, the ICF model optimized by IZOA is trained using the training set, and the model is tested using the test set to determine the voltage stability of the power system.

[0131] Example:

[0132] To verify the effectiveness of the present invention, a test was conducted on an IEEE 14-node system. The schematic diagram of the IEEE 14-node system is shown in FIG. Figure 4 In the IEEE 14 bus system, the synchronous generators in bus 6 and bus 2 are replaced by wind farms (WF) based on double-fed induction generators, and the accuracy of the improved cascade forest model based on the improved zebra optimization algorithm in voltage stability assessment is verified through simulation.

[0133] After the ICF model is trained with the training set, the test set is used to test the ICF model under various unknown load conditions. The continuous power flow (CPF) experimental results based on the FVSI value are used for verification. The FVSI values ​​of each bus of the IEEE 14-bus system measured by the two methods under different load conditions are obtained, and the FVSI values ​​of each bus are compared to obtain the minimum bus and the maximum bus. Under the baseline load condition, the FVSI minimum bus and the maximum bus are shown in Table 1; under the maximum load condition, the FVSI minimum bus and the maximum bus are shown in Table 2.

[0134] Table 1

[0135]

[0136] Table 2

[0137]

[0138] As can be seen from the above table, by comparing CPF as a benchmark with the method proposed in the present invention, the trained ICF gives an accurate FVSI value, and the FVSI value of bus 14 is closer to 1 than other buses, and it is closer to voltage instability. The FVSI value of bus 4 is lower, indicating that bus 4 has stronger voltage stability. Table 1 and Table 2 give the values ​​of FVSI under the baseline load and maximum load conditions, respectively. In both scenarios, bus 14 is determined to be a voltage-fragile bus, and compensation devices are placed to improve the voltage stability of the entire network. The method proposed in the present invention is compared with the CPF method, and the improved cascade forest model is used to evaluate the voltage stability of the power system, which is not only fast but also approximately accurate.

[0139] The invention proposes a method for evaluating power system voltage stability based on improved cascade forest, which is of great significance for ensuring the safe and stable operation of the power system.

Claims

1. A method for evaluating power system voltage stability based on improved cascade forest, characterized in that: The following steps are involved: Step 1: Construct the fast voltage stability index FVSI to describe the voltage stability state of the power system and obtain the voltage stability margin of the power system; Step 2: Use the improved cascade forest ICF model to represent the mapping relationship between input variables and fast voltage stability index FVSI; Step 3: Use the improved zebra optimization algorithm IZOA to optimize the improved cascade forest ICF model; Step 4: Use the offline power flow method to construct a data set of the algorithm-optimized improved cascade forest ICF model, preprocess the data, and generate an efficient data set; Step 5: Divide the efficient data set into a training set and a test set, use the training set to train the model, and use the test set to test the model.

2. The method according to claim 1, characterized in that: In step 1, a fast voltage stability index FVSI is constructed to describe the voltage state of the power system and obtain the voltage stability margin of the power system, which specifically includes the following steps: Step 1-1: Construct the fast voltage stability index FVSI to describe the voltage status of the power system; The voltage stability index can be used to study the voltage stability of the power system. The fast voltage stability index FVSI is used to detect the key busbars in the power system. The calculation formula is as follows: Where: Z sr is the line impedance from the transmitter to the receiver; Q r is the reactive power at the receiving end; V s is the voltage at the transmitting end; X sr is the line reactance from the sending end to the receiving end; Step 1-2: Obtain the voltage stability margin of the power system based on the fast voltage stability index FVSI; The smaller the fast voltage stability index FVSI value, the more stable the line. When the value is close to 1, it indicates that the line is close to instability, which may cause voltage collapse of the power system in severe cases.

3. The method according to claim 1, characterized in that: In step 2, an improved cascade forest ICF model is used to represent the mapping relationship between input variables and the fast voltage stability index FVSI, which specifically includes the following steps: Step 2-1: Improve the traditional cascade forest CF model to obtain an improved cascade forest ICF model; Improve the structure of the cascade forest CF model and increase the types of base learners; Integrate CatBoost, XGBoost, Random Forest and LightGBM learners to obtain an improved cascade forest ICF model; Step 2-2: Represent the mapping relationship between input variables and fast voltage stability index FVSI based on the improved cascade forest ICF model; The input variables such as bus voltage amplitude, generator active and reactive power are input into the improved cascade forest ICF model. Each layer uses CatBoost, XGBoost, random forest and LightGBM learners to train the model respectively to obtain the prediction probability, and these probabilities are added as new features to the input variables of the next layer. Feature conversion and fusion are performed at each layer of the model. Finally, the multi-level cascade model can be continuously optimized to obtain an accurate fast voltage stability index FVSI.

4. The method according to claim 1, characterized in that: In step 3, the improved zebra optimization algorithm IZOA is used to optimize the improved cascade forest ICF model, which specifically includes the following steps: Step 3-1: Set the algorithm parameters, including the number of zebra populations, iterative convergence conditions, and the minimum and maximum values ​​of the parameters that need to be optimized in the improved cascade forest ICF model; Step 3-2: Introduce elite reverse learning EOBL to initialize the population: First, randomly generate a zebra population, then compare the fitness function value of the existing zebras with the fitness function value of the elite opposition zebras, and select the zebra with the lower fitness function value as the next generation of zebras; Based on elite reverse learning EOBL for any zebra Z i , whose reverse zebra representation is The formula is as follows: Where: r is a random number between 0 and 1; LB is the lower bound of elite zebras; UB is the upper bound of elite zebras; Step 3-3: Position update during foraging phase: Use equation (3) to calculate the new position of the zebra and update the zebra position; The formula for updating the zebra's position after the foraging phase is as follows: Where: is the updated position of zebra i in the jth dimension after foraging behavior; z i,j is the position of zebra i in the jth dimension; MZ j is the optimal position of the jth dimension; r is a random number between 0 and 1; l∈{1,2}, the larger the l value, the faster the population moves; Step 3-4: Defense phase position update: according to the random number p s The value of , calculates the new position of the zebra; When attacked by a lion, the zebra will choose an escape strategy as shown in equation (4) S1; when facing other predators, the zebra will choose an attack strategy as shown in equation (4) S2; Where: is the updated position of zebra i in the jth dimension after the defensive behavior; r is a random number between 0 and 1; t and T are the current iteration number and the total number of iterations, respectively; k is a constant of 0.01; P s is a random number between [0,1]; NZ j is the optimal position of the jth dimension; Step 3-5: Introduce the firefly interference strategy FDS to update the zebra position again: Use formula (5) to calculate the new position of the zebra, then compare the fitness function value of the existing zebra with the fitness function value of the pioneer zebra, and select the zebra with the lower fitness function value as the next generation zebra; Where: β0 and θ are constants; S is the spatial distance between the zebra and the pioneer zebra; δ is a step factor in the interval [0,1]; Step 3-6: Verify whether the iteration limit is met: If the maximum number of iterations is not reached, return to step 3-2 to continue iterating; if it is reached, end the optimization process.

5. The method according to claim 1, characterized in that: In step 4, an offline power flow method is used to construct a data set of an algorithm-optimized improved cascade forest ICF model, and the data is preprocessed to generate an efficient data set, which specifically includes the following steps: Step 4-1: Normalize the data based on the MAXABS method; The MAXABS normalization method rescales the data x to the range of [-1,1], and the calculation formula is as follows: In the formula: x is data; |x max | is the maximum absolute value of the data; Step 4-2: Use Fisher Score algorithm to filter the original data to obtain an efficient data set; Step 4-2-1: Mark the state of the data normalized in step 4-1, and divide the data into a training data set and a test data set according to a certain ratio; Step 4-2-2: Use Fisher Score algorithm to calculate the importance of electrical features in the original data; Assume that the number of stable and unstable samples in all electrical characteristic samples is K s and K m , use the Fisher Score algorithm to calculate the sample a-th dimension feature score F a The formula is as follows: Where: m n , and are the means of the sample's a-th dimension feature in the entire data set, in the stable sample data set, and in the unstable sample set, respectively; is the eigenvalue of the a-th dimension feature of the i-th stable sample data; is the eigenvalue of the a-th dimension feature of the i-th unstable sample data; Step 4-2-3: Arrange all features in descending order according to Fisher Score value; Step 4-2-4: Add the features to the training data set in order to train the improved cascade forest ICF model, and obtain the feature dimension when the model accuracy is the highest through comparison.

Images