CN111652478B - Umbrella algorithm-based power system voltage stability evaluation misclassification constraint method - Google Patents

Abstract

The utility model relates to an umbrella algorithm-based power system voltage stability evaluation misclassification constraint method, which comprises the following steps: constructing an initial data set, and constructing a voltage stability safety classification label based on a voltage stability evaluation rule of the power system; step 2: selecting key operation variables to construct a high-efficiency sample set; step 3: performing offline training on the voltage stability evaluation model; step 4: sending a new sample set generated under a new operating condition into a voltage stability evaluation model for model update; step 5: and (5) completing online voltage stability evaluation by using the trained voltage stability evaluation model. In order to solve the limitation of the traditional VSA model constructed based on a data driving tool in terms of misclassification constraint and model updating mechanism, the invention provides an umbrella algorithm-based power system voltage stability evaluation misclassification constraint method, so that the VSA model can provide a VSA result for balancing the overall classification precision and one class of classification error constraint.

Description

Umbrella algorithm-based power system voltage stability evaluation misclassification constraint method

The invention belongs to the field of power system voltage stability evaluation, and particularly relates to an umbrella algorithm-based power system voltage stability evaluation misclassification constraint method.

Background

Power systems are one of the most complex industrial systems in the world, and their safe operation has been a concern for system development. In recent years, the trends of wide area interconnection of modern power systems, input of renewable energy sources, application of new equipment, rapid increase of load, limitation of transmission capacity and the like are becoming more and more evident. The operation load of the power system is heavier and the operation state is approaching the limit. Static voltage stabilization is a major concern, and many blackout accidents are related to the static voltage stabilization, which can cause huge economic loss and adverse social influence. Thus, reliable voltage stability assessment (Voltage Stability Assessment, VSA) has been attracting increasing interest and research by researchers, which is of great importance for safe operation of electrical power systems. Traditional analysis methods for real-time VSA have the defects of time consumption, slow calculation speed and the like, and the existing data-driven solution has some limitations:

(1) there are two types of misclassification situations in a power system VSA: judging the unsafe state as a class of classification errors of the safe state; and judging the safe state as the second class classification error of the unsafe state. Most of the current VSA research is mainly focused on how to improve the overall classification accuracy of the VSA, and one class of classification error constraint is easily ignored. In actual power system operation, the first class classification errors and the second class classification errors have different social effects on system operation. The consequences of a class one classification error are more severe than a class two classification error. (2) For the study of VSA performance, conventional VSA models generally focus on the construction of VSA models with a specific data-driven tool as a core. For complex and changeable running environments of a power system, the results provided by the VSA model are easily limited by the excellent performance of the model, and the generalization capability of the model is not strong. (3) The traditional model training mechanism adopts a mode of combining offline and online of single solidification, and when the model is faced with a new operation condition, the model retraining mechanism is started immediately when the current evaluation model does not meet the evaluation requirement. The model updating mechanism can increase the burden of offline training of the model in real-time VSA for the operation condition of invisible system, and has higher requirement on offline training.

In summary, the existing VSA method cannot effectively restrict and evaluate the influence of misclassification on the safe operation of the power system, and has some limitations on the design of the online VSA method for the complex and changeable operation environment of the power system.

The patent document with the authority bulletin number of CN105139289A discloses a power grid transient voltage stability assessment method based on error division cost classification learning, which is based on dynamic measurement data of a synchronous phasor measurement unit, and extracts a key subsequence closely related to a power grid state from a time sequence formed by a large amount of dynamic measurement data; introducing a weight coefficient to a learning sample by setting different misclassification costs of a stable state and a unsteady state of the power grid; and performing classification learning by utilizing a decision tree algorithm integrated with the sample weight coefficient to obtain a decision tree model, and performing evaluation on the transient voltage stability condition of the power grid by using the decision tree model for on-line monitoring. The defects are that:

(1) the overall classification accuracy and the class classification error rate relation cannot be well weighed, class classification errors are restrained according to different severity of different misclassifications of VSA, and the risk of class classification errors on system operation is reduced; (2) depending only on the VSA model of the decision tree construction, there may be missing values for object attributes in the dataset, the performance of the tree may be problematic, and the order of attributes in the tree nodes may have a negative impact on performance. The adaptability of the VSA model constructed by only relying on the decision tree is not guaranteed; (3) for implementation of online VSA, a reliable model update mechanism is lacking, and it is difficult to guarantee the effectiveness of online VSA for invisible operating conditions.

Disclosure of Invention

In order to solve the limitation of the traditional VSA model constructed based on a data driving tool in terms of misclassification constraint and model updating mechanism, the invention provides an umbrella algorithm-based power system voltage stability evaluation misclassification constraint method, so that the VSA model can provide a VSA result for balancing the overall classification precision and one class of classification error constraint.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

the power system voltage stability evaluation misclassification constraint method of the umbrella-based algorithm comprises the following steps:

step 1: constructing an initial data set, and constructing a voltage stability safety classification label based on a voltage stability evaluation rule of the power system;

step 2: utilizing a linear and nonlinear relation exploration tool for feature selection, exploring the association degree between the operation variable of the initial sample set and the voltage stability security classification label, and selecting a key operation variable to construct an efficient sample set;

step 3: simultaneously deploying a plurality of types of Nawman pearson classifiers based on an umbrella algorithm to construct a voltage stability evaluation model, taking a high-efficiency sample set and a voltage stability security classification label corresponding to the high-efficiency sample set as model input, and performing offline training on the voltage stability evaluation model;

step 4: taking the influence of factors such as the change of the operation topological structure of the power system, the change of the power distribution between the generator and the load, the data loss of the phasor measurement unit and the like on the system into consideration, and sending a new sample set generated under the new operation working condition into a voltage stability evaluation model for model update;

step 5: and selecting corresponding characteristics based on the power system operation data collected in real time by the phasor measurement unit of the wide area monitoring system, and completing online voltage stability evaluation by using the trained voltage stability evaluation model.

In step 1, operation data of a historical system working point is obtained through real-time updating and collecting of system operation data of a power system PMUs, an initial data set is constructed, voltage breakdown conditions of various operation working conditions are simulated by using a continuous power flow CPF method, the working point is enabled to gradually move from a basic condition to the breakdown point, when the power system cannot meet the increasing load demand, a jacobian matrix of a power flow equation is singular, power flow cannot converge, a voltage breakdown point occurs, and the relation between the current operation point and the voltage of the breakdown point is quantized by using the load active power difference relation between the working point and the voltage breakdown point.

The voltage stability index VSI is constructed as shown in formula (1):

wherein: p (P) ₀ Initial load active power, P _max Active power is received for the corresponding maximum;

the VSI value represents the safety level of the power system, and varies between 0 and 100%, and in order to establish a safety classification rule, an appropriate acceptable threshold η may be set for the VSI to clearly distinguish the voltage safety state, and then a voltage stability label is constructed as shown in formula (2):

in step 2, complex and variable relation of system operation variables are considered in the characteristic selection process, PCC with linear relation exploration function and MIC with nonlinear relation exploration function are combined to be used as a characteristic selection method, the linear relation and nonlinear relation between operation variables (such as active/reactive power, voltage amplitude and the like of a generator/branch) and voltage stability indexes are effectively explored, screening of effective key operation variables is ensured, a high-efficiency sample set is constructed, and the problem of high data dimension is solved.

Constructing each feature of the ith operating point of the power system and a corresponding VSI thereof into a row vector F of a feature description set F _i ＝{x ₁ ,x ₂ ,...,x _m ,y _i I is more than or equal to 1 and less than or equal to m), the feature description set F= { X of the operation data of the n operation conditions ₁ ,X ₂ ,...,X _m Y }, wherein X _k Column vectors (1.ltoreq.k.ltoreq.n) composed of the same feature quantity under various operating conditions, Y= { Y ₁ ,y ₂ ,...,y _n The VSI set of each operation condition is represented, PCC and MIC are respectively adopted for detecting the correlation between each operation variable and the VSI for the operation data variable set X and the corresponding VSI set Y of each operation condition, and the correlation between each operation variable and the VSI is respectively screenedOutputting a high-ranking linear relation operation variable and a nonlinear operation variable which are highly related to the VSI;

exploring PCC for linear relations: let ρ (X, Y) be the pearson correlation coefficient of the feature quantity X, Y, the calculation formula is shown in formula (3):

wherein: n is the dimension of a single feature quantity;

is->

Average values of elements contained in X and Y respectively;

PCC has a value ranging from-1 to 1 and has several attributes:

1) ρ (X, Y) > 0 indicates that there is a positive correlation between X and Y;

2) ρ (X, Y) =0 illustrates the wireless correlation between X and Y;

3) ρ (X, Y) < 0 indicates that there is a negative correlation between X and Y;

4) The larger the absolute value of rho (X, Y), the stronger the linear correlation exists between X and Y;

exploring MIC for nonlinear relations: given a finite order vector data set d= { (x) _i ,y _i ) If the X and Y axes are divided into X and Y grids, respectively, obtaining an X Y grid G, and the variable values in D fall into the grid of G to obtain a corresponding probability distribution d| _G Wherein x and y are positive integers, and different mutual information values can be obtained by changing grid division positions on the premise of fixing the grid division number, wherein the maximum mutual information value is shown as the following formula (4):

I ^* (D,x,y)＝maxI(D| _G ) (4)

wherein: i (D|) _G ) Represents D| _G Mutual information between the inner data points;

to facilitate comparison between different dimensions, equation (4) is normalized as shown in equation (5) to have its value in interval [0,1]:

given an ordered pair data set D of sample size n, the MIC of two variables X, Y in the set is defined as shown in equation (6):

wherein: xy.ltoreq.B (n) { B (n) =n ^a Is generally set as n ^0.6 }；

The MIC ranges from 0 to 1 and has several attributes:

1) For two variables with a functional relationship that tends to be noiseless, their MIC values tend to be 1;

2) MIC values tend to be 1 for a broader class of noiseless relationships;

3) For two variables that are statistically independent of each other, the MIC value tends to be 0.

The method comprises the steps of dividing a high-efficiency sample set into a training set and a testing set by adopting a ten-time cross validation method, sending the training set and the testing set into an umbrella type NP classifier for model offline training, modifying a plurality of traditional separators based on an umbrella type algorithm to obtain a series of umbrella type NP classifiers corresponding to the NP classifiers, and under the premise of guaranteeing the overall classification precision in a model training stage, setting a class error threshold alpha for class error classification errors based on the umbrella type NP classifiers according to the difference of class error and class error properties, controlling the training mode of each NP classifier by setting training set cyclic splitting training times M, obtaining a series of sub NP classifiers, and obtaining stable and unstable classification results of each NP classifier by adopting a weighted voting mode for the sub NP classifiers of each type NP classifier, thereby optimizing the classification performance of a VSA model.

The feature selection method for the feature selection process in the voltage stability evaluation combines PCC with a linear relation exploration function and MIC with a nonlinear relation exploration function as feature selection methods, explores the linear relation and the nonlinear relation between an operation variable and a voltage stability index, ensures screening out effective key operation variables, and constructs a high-efficiency sample set; the operating variables include the active and/or reactive power of the generator, the active and/or reactive power of the branch, the voltage amplitude.

Constructing each feature of the ith operating point of the power system and a corresponding VSI thereof into a row vector F of a feature description set F _i ＝{x ₁ ,x ₂ ,...,x _m ,y _i I is more than or equal to 1 and less than or equal to m), the feature description set F= { X of the operation data of the n operation conditions ₁ ,X ₂ ,...,X _m Y }, wherein X _k Column vectors (1.ltoreq.k.ltoreq.n) composed of the same feature quantity under various operating conditions, Y= { Y ₁ ,y ₂ ,...,y _n Detecting the correlation between each operation variable and the VSI by PCC and MIC for the operation data variable set X and the corresponding VSI set Y of various operation conditions, and respectively screening out high-ranking linear relation operation variables and nonlinear operation variables highly related to the VSI;

exploring PCC for linear relations: let ρ (X, Y) be the pearson correlation coefficient of the feature quantity X, Y, the calculation formula is shown in formula (3):

wherein: n is the dimension of a single feature quantity;

is->

Average values of elements contained in X and Y respectively;

PCC has a value ranging from-1 to 1 and has several attributes:

1) ρ (X, Y) > 0 indicates that there is a positive correlation between X and Y;

2) ρ (X, Y) =0 illustrates the wireless correlation between X and Y;

3) ρ (X, Y) < 0 indicates that there is a negative correlation between X and Y;

4) The larger the absolute value of ρ (X, Y), the stronger the linear correlation between X and Y,

exploring MIC for nonlinear relations: given a finite order vector data set d= { (x) _i ,y _i ) If the X and Y axes are divided into X and Y grids, respectively, obtaining an X Y grid G, and the variable values in D fall into the grid of G to obtain a corresponding probability distribution d| _G Wherein x and y are positive integers; on the premise of fixing the grid division number, different mutual information values can be obtained by changing the grid division positions, wherein the maximum mutual information value is shown as the following formula (4):

I ^* (D,x,y)＝maxI(D| _G ) (4)

wherein: i (D|) _G ) Represents D| _G Mutual information between the inner data points;

to facilitate comparison between different dimensions, equation (4) is normalized as shown in equation (5) to have its value in interval [0,1]:

given an ordered pair data set D of sample size n, the MIC of two variables X, Y in the set is defined as shown in equation (6):

/>

wherein: xy.ltoreq.B (n) { B (n) =n ^a Is generally set as n ^0.6 }；

The MIC ranges from 0 to 1 and has several attributes:

1) For two variables with a functional relationship that tends to be noiseless, their MIC values tend to be 1;

2) MIC values tend to be 1 for a broader class of noiseless relationships;

3) For two variables that are statistically independent of each other, the MIC value tends to be 0.

A training set circulation splitting training method for voltage stability assessment divides a high-efficiency sample set into a training set and a testing set by adopting a multiple cross validation method, the training set and the testing set are sent into an umbrella type NP classifier to carry out model offline training, a plurality of traditional separators are modified based on an umbrella type algorithm to obtain a series of umbrella type NP classifiers corresponding to the NP classifier, under the premise of guaranteeing overall classification accuracy in a model training stage, the umbrella type NP classifier can be used for setting a class error threshold value alpha for class error classification according to the difference of class error and class error classification error properties, the training mode of each NP classifier can be controlled by setting training set circulation splitting training times M, a series of sub NP classifiers are obtained, and the stable and unstable classification results of each NP classifier are obtained by adopting a weighted voting mode for the sub NP classifier of each type NP classifier, so that the classification performance of a VSA model is optimized.

The umbrella algorithm comprises the following steps:

1) According to the set M, for class 0 sample S in training set ⁰ Equally randomly splitting (unstably judged to be stable misclassified sample) to obtain a sample

And sample->

Binding class 1 sample S ¹ (misclassified samples that are determined to be unstable to be stable) are used for the base classifier;

2)

the upper limit upsilon (k) of the violation rate calculated by the classifier is shown as a formula (7), and the threshold k of each sample level is calculated ^* As in equation (8); will->

Binding S ¹ Training a scoring function f for each classifier (e.g., RF) _i ：

Wherein: alpha is a class of classification error threshold;

k ^* ＝min{k∈(1,...,n):υ(k)≤δ} (8)

wherein: delta is the rate of violation of a class of classification error thresholds,

3) Will f _i Is applied to

Obtaining a group of threshold candidates tau _i As in equation (9), and will be τ _i The elements are sorted according to the increasing sequence of the formula (10) to obtain the sort (tau) _i ) And find the level threshold k ^* Corresponding score threshold +.>

As in formula (11):

τ _i ＝{t _i,1 ,...,t _i,n }＝{f _i (x ₁ ),...,f _i (x _n )} (9)

sort(τ _i )＝{t _i,(1) ,...,t _i,(n )} (10)

4) Constructing an NP classifier based on the scoring function and the threshold as shown in equation (12):

5) Repeating the steps 1) to 4), carrying out cyclic split training on the training set for M times, constructing M multiple NP classifiers, and outputting an integrated method of weighted voting of one class of classification error results according to a formula (13) as a final result:

by adopting the technical scheme, the method has the following technical effects:

(1) According to the technical scheme, the problem of error classification which is easy to ignore in the VSA research of the power system is considered, the class classification error threshold is set according to actual operation requirements aiming at class classification errors with serious consequences, and class classification errors are restrained on the premise of ensuring overall classification accuracy, so that the effect of balancing the overall classification accuracy and class classification errors is achieved, and the operation risk caused by the class classification errors of the power system is reduced.

(2) According to the technical scheme, for the data high-dimensional problem, the linear and nonlinear relations of the data are considered at the same time when feature selection processing is adopted, and feature reliability during data dimension reduction is ensured through MIC (many integrated circuits) of nonlinear relation exploration and PCC (policy and charging control) of linear relation exploration.

(3) According to the technical scheme, based on the umbrella type NP algorithm, a plurality of types of NP classifiers can be deployed at the same time, a new VSA model updating mechanism is designed, the model training burden when the model is applied online can be reduced by calling a plurality of classifiers online at the same time, and seamless VSA realization can be effectively ensured. Meanwhile, a set of training sets are used for training the VSA model in a cyclic split mode, the defect that the traditional VSA model training mode is too dependent on training data is avoided, and the effectiveness aspect of VSA model training is guaranteed.

Drawings

FIG. 1 is a flow chart of the overall scheme of the present invention;

FIG. 2 is a flow chart of a VSA model update of the present invention;

FIG. 3 is a diagram of the topology of a 23-node system of the present invention;

fig. 4 is a schematic representation of ROC bands of different NP classifiers of the present invention.

Detailed Description

The invention will be further described with reference to the drawings and examples.

As shown in fig. 1, the flow chart of the error classification constraint method for voltage stability evaluation of the power system based on the umbrella algorithm comprises the following steps:

step 1: constructing an initial data set, and constructing a voltage stability safety classification label based on a voltage stability evaluation rule of the power system;

step 2: utilizing a linear and nonlinear relation exploration tool for feature selection, exploring the association degree between the operation variable of the initial sample set and the voltage stability security classification label, and selecting a key operation variable to construct an efficient sample set;

step 3: simultaneously deploying various types of NP classifiers based on an umbrella algorithm to construct a voltage stability evaluation VSA model, taking a high-efficiency sample set and a voltage stability safety classification label corresponding to the high-efficiency sample set as model input, and performing offline training on the VSA model;

step 4: taking the influence of factors such as the change of an operation topological structure of the power system, the change of power distribution between a generator and a load, PMU data loss of a phasor measurement unit and the like on the system into consideration, and sending a new sample set generated under a new operation working condition into a VSA model for model update;

step 5: online VSA is performed on PMU real-time measurement data.

In step 1 of fig. 1, by collecting system operation data by the PMUs of the power system in real time, a large amount of operation data of historical system operation points can be obtained from the power company, and an initial data set is constructed. And simulating voltage breakdown conditions of various operation conditions by using a continuous power flow CPF method. As the load demand slowly increases, the operating point will gradually move from the base case to the crash point. When the power system cannot meet the increasing load demands, the jacobian matrix of the power flow equation is singular, the power flow cannot converge, and voltage collapse points appear. The relation between the current operating point and the voltage of the breakdown point is quantified by utilizing the relation of the load active power difference between the operating point and the voltage breakdown point, and a voltage stability index VSI is constructed as shown in formula (1):

wherein: p (P) ₀ Initial load active power, P _max Is the corresponding maximum sustainable active power.

VSI values represent power system safety levels, varying between 0 and 100%. In order to establish the safety classification rule, an appropriate acceptable threshold η may be set for the VSI to clearly distinguish the voltage safety states, and then a voltage stability label is constructed as shown in formula (2):

in step 2 of fig. 1, the online VSA of the power system needs to quickly determine the system state by using the massive data of the real-time operation of the system. In the characteristic selection process, complex and variable relation of system operation variables are considered, PCC with a linear relation exploration function and MIC with a nonlinear relation exploration function are combined to serve as a characteristic selection method, the linear relation and the nonlinear relation between operation variables (such as active/reactive power, voltage amplitude and the like of a generator/branch) and voltage stability indexes are effectively explored, the screening of effective key operation variables is ensured, an efficient sample set is constructed, and the problem of high data dimension is solved.

Constructing each feature of the ith operating point of the power system and a corresponding VSI thereof into a row vector F of a feature description set F _i ＝{x ₁ ,x ₂ ,...,x _m ,y _i I is more than or equal to 1 and less than or equal to m), the feature description set F= { X of the operation data of the n operation conditions ₁ ,X ₂ ,...,X _m Y }, wherein X _k Column vectors (1.ltoreq.k.ltoreq.n) composed of the same feature quantity under various operating conditions, Y= { Y ₁ ,y ₂ ,...,y _n VSI set representing individual operating conditions. And detecting the correlation between each operation variable and the VSI by adopting PCC and MIC for the operation data variable set X and the corresponding VSI set Y of various operation conditions, and respectively screening out high-ranking linear relation operation variables and nonlinear operation variables highly correlated with the VSI.

Exploring PCC for linear relations: let ρ (X, Y) be the pearson correlation coefficient of the feature quantity X, Y, the calculation formula is shown in formula (3):

wherein: n is the dimension of a single feature quantity;

is->

The average values of the elements contained in X and Y are shown.

PCC has a value ranging from-1 to 1 and has several attributes:

1) ρ (X, Y) > 0 indicates that there is a positive correlation between X and Y;

2) ρ (X, Y) =0 illustrates the wireless correlation between X and Y;

3) ρ (X, Y) < 0 indicates that there is a negative correlation between X and Y;

4) The larger the absolute value of ρ (X, Y), the stronger the linear correlation between X and Y.

Exploring MIC for nonlinear relations: given a finite order vector data set d= { (x) _i ,y _i ) If the X and Y axes are divided into X and Y grids, respectively, obtaining an X Y grid G, and the variable values in D fall into the grid of G to obtain a corresponding probability distribution d| _G Wherein x and y are positive integers. On the premise of fixing the grid division number, different mutual information values can be obtained by changing the grid division positions, wherein the maximum mutual information value is shown as the following formula (4):

I ^* (D,x,y)＝maxI(D| _G ) (4)

wherein: i (D|) _G ) Represents D| _G Mutual information between the inner data points.

To facilitate comparison between different dimensions, equation (4) is normalized as shown in equation (5) to have its value in interval [0,1]:

given an ordered pair data set D of sample size n, the MIC of two variables X, Y in the set is defined as shown in equation (6):

wherein: xy.ltoreq.B (n) { B (n) =n ^a Is generally set as n ^0.6 }。

The MIC ranges from 0 to 1 and has several attributes:

1) For two variables with a functional relationship that tends to be noiseless, their MIC values tend to be 1;

2) MIC values tend to be 1 for a broader class of noiseless relationships;

3) For two variables that are statistically independent of each other, the MIC value tends to be 0.

In step 3 of fig. 1, the high-efficiency sample set is divided into a training set and a testing set by adopting a ten-time cross validation method, and the training set and the testing set are sent into an umbrella-type NP classifier to perform model offline training, and a plurality of traditional separators are subjected to the umbrella-type algorithm: RF, adaBoost, SVM, NB, NNB, penlog, etc., to obtain a series of umbrella-type NP classifiers corresponding to the NP classifier: NP-RF, NP-ADA, NP-SVM, NP-NB, NP-NNB, NP-Penlog. In the model training stage, on the premise of ensuring the overall classification precision, the umbrella-type NP classifier can be used for restraining the class-error classification errors according to the difference of the class-error classification errors and the class-error classification errors, the training mode of each NP classifier can be controlled by setting the training set circulation split training times M, a series of sub NP classifiers are obtained, and the stable and unstable classification results of each NP classifier are obtained by adopting a weighted voting mode for the sub NP classifiers of each type of NP classifier, so that the classification performance of a VSA model is optimized.

The specific flow of the umbrella algorithm is as follows:

1) According to the set M, for class 0 sample S in training set ⁰ (instability)Misclassified samples that are determined to be stable) are equally randomly split to obtain samples

And sample->

Binding class 1 sample S ¹ (misclassified samples that are determined to be unstable to be stable) are used for the base classifier;

2)

the upper limit upsilon (k) of the violation rate calculated by the classifier is shown as a formula (7), and the threshold k of each sample level is calculated ^* As in equation (8); will->

Binding S ¹ Training a scoring function f for each classifier (e.g., RF) _i ：

Wherein: alpha is a class of classification error thresholds.

k ^* ＝min{k∈(1,...,n):υ(k)≤δ} (8)

Wherein: delta is the violation rate of a class of classification error thresholds.

3) Will f _i Is applied to

Obtaining a group of threshold candidates tau _i As in equation (9), and will be τ _i The elements are sorted according to the increasing sequence of the formula (10) to obtain the sort (tau) _i ) And find the level threshold k ^* Corresponding score threshold +.>

As in formula (11): />

τ _i ＝{t _i,1 ,...,t _i,n }＝{f _i (x ₁ ),...,f _i (x _n )} (9)

sort(τ _i )＝{t _i,(1) ,...,t _i,(n ）｝ (10)

4) Constructing an NP classifier based on the scoring function and the threshold as shown in equation (12):

5) Repeating the steps 1) to 4), carrying out cyclic split training on the training set for M times, constructing M multiple NP classifiers, and outputting an integrated method of weighted voting of one class of classification error results according to a formula (13) as a final result:

in step 4 of fig. 1, the updating of the model is as shown in fig. 2, and the specific updating mode is as follows:

1) First case: when the changed new operation condition exists in the current offline database list, the current VSA model is immediately replaced by the corresponding new model, and the VSA is performed.

2) Second case: the model continues to be used while currently using the VSA model NP classifier can provide acceptable evaluation results for the changed new operating conditions. A fast VSA is achieved.

3) Third case: when invisible operating conditions occur, none of the NP classifiers in the existing VSA model provide acceptable evaluation results. By using the updated sample set for umbrella NP classifier retraining, a new VSA model is constructed that is suitable for new operating conditions.

In step 4 of fig. 1, for the third-case VSA model retraining update, in order to rapidly evaluate the training effect of each NP classifier on the new operating condition, an ROC band with two ROC curves and its corresponding AUC values evolved from the ROC curves are used to evaluate the classification performance of each NP classifier. Wherein the closer the ROC curve is to the upper left corner, the better the corresponding classifier performance. And the AUC value is between 0 and 1, when the AUC value is larger than 0.5, the higher the AUC value is, the better the classification performance is, and the AUC values corresponding to the upper curve and the lower curve of the ROC band are respectively expressed as AUC.U and AUC.L.

In step 5 of fig. 1, for system operation data collected in real time by the system PMUs, required feature data is quickly selected through the proposed scheme, and is sent into a trained VSA model to perform online VSA, so that deployed S NP classifier results meeting evaluation requirements are quickly given. Meanwhile, in order to ensure the reliability of the VSA result, a trusted VSA result is obtained, and the result of each NP classifier is subjected to the following minority-compliance majority voting mechanism to obtain the final result of the online VSA.

1) When S is odd, greater than

If the NP classifier output security label is 1, then the online VSA result is given as system security; is greater than->

If the NP classifier output security label is 0, then the online VSA result is given as unsafe;

2) And when S is even, eliminating the NP classifier result with the worst classifying performance, and then obtaining a final VSA result according to the step 1).

Examples:

the present invention was tested in a 23-node system and an actual 7917-node system, wherein the 23-node system comprises 23 nodes, 10 transformers and 6 generators as shown in fig. 3. While the 7917 node system contains 7917 nodes, 1325 generators and 5590 loads. To capture more system behavior to enrich the database, a series of simulations were automatically performed in the software PSS/E using the Python program, taking into account the different operating conditions. The generator/load power distribution varies randomly between 80% and 120%, and based on a series of continuous power flow simulations, the 23-node system takes 3896 samples, the 7917 system takes 9876 samples. The test was performed on a computer equipped with an Intel Core i7 processor and 8GB of memory.

Fig. 4 and table 1 show ROC bands and AUC values (auc.u and auc.l represent the areas of the upper ROC band and the lower ROC band, respectively) corresponding to the 6 NP classifiers tested by the 23-node system, it can be seen that the 6 ROC bands are all near the upper left corner and the corresponding AUC values are all greater than 0.5. All 6 NP classifiers deployed by the VSA model were shown to exhibit satisfactory VSA performance, but by comparison it can also be seen that the performance of the NP-RF, NP-ADA, NP-Penlog three classifiers was superior. Therefore, in the following VSA tests of the 23-node system and the 7917-node system, three NP classifiers, namely NP-RF, NP-ADA and NP-SVM, with better performance were selected to construct a VSA model for the VSA model test. Where the 23-node system selects α=0.006, δ=0.05 and m=3 as classifier reference parameters. The 7917 node system selects α=0.02, δ=0.05 and m=3 as classifier reference parameters. And the overall classification Accuracy (AC), the type I classification error rate (FD) and the F value (FM) are adopted as VSA classification performance detection indexes, and the VSA classification performance detection indexes are respectively represented by formulas (14) (15) (16):

wherein: f (F) ₁₁ ,F ₁₀ ,F ₀₀ ,F ₀₁ The number of stable samples is determined as the stable number, the number of unstable samples is determined as the unstable number, and the number of unstable samples is determined as the stable number, respectively.

TABLE 1

AUC values of different NP classifiers

AUC area	NP-RF	NP-ADA	NP-Penlog	NP-SVM	NP-NB	NP-NNB
							AUC.L	0.9899	0.9897	0.9883	0.9700	0.9471	0.9580
AUC.U	0.9956	0.9997	0.9906	0.9866	0.9364	0.9469

Table 2 shows the performance test results of the three NP classifiers and the traditional classifier in two systems under the reference parameter condition, and the classification accuracy, class-I classification error rate and F value of the systems are recorded respectively. As shown by result analysis, compared with the traditional classifier, the NP classifier has higher overall classification precision and lower class classification error rate, and is used for evaluating the F value of the classifier on the classification capability of an unstable sample, so that the three NP classifiers deployed based on the umbrella algorithm provide a foundation for constructing a VSA model with excellent performance.

TABLE 2

Performance comparison result of three NP classifiers and traditional classifier

TABLE 3 Table 3

VSA model voting mechanism test result under reference parameter condition

As shown in Table 3, the evaluation results of the VSA model obtained by voting the results of the three NP classifiers in two systems can be seen that the VSA model subjected to the voting mechanism has better classification performance and better class classification error constraint capability. Therefore, the majority-compliant voting of the minority of the VSA model can improve the quality of the VSA result and reduce the risk of the misclassification of the VSA on the operation of the power system.

Table 4 shows the class-one classification error rates of the three NP classifiers and their VSA model voting mechanisms when the class-one classification error threshold α is adjusted, it can be known that adjusting the class-one error threshold can indeed restrict the class-one classification error within a specified range, and at the same time, the training mode of the training set can indeed obtain a more ideal classification result. Therefore, the threshold parameters of the VSA model can be adjusted to provide an alternative model according to the actual operation requirement of the power system, and the scheme has a certain application prospect in the aspect of ensuring the safe operation of the power system.

TABLE 4 Table 4

Class-one classification error of NP classifier under different threshold alpha conditions

Table 5 shows the evaluation results of the overall classification accuracy and class-one classification error rate of the VSA model and the NP classifier of the VSA model when two systems simulate different topologies. It can be seen that when the topology changes, the two systems can still maintain good overall classification accuracy and class classification error constraints. Therefore, the scheme still has better applicability to the influence of the change of the actual operation topology structure of the power system.

TABLE 5

VSA model under different topological structure conditions and performance of NP classifier thereof

The results of the three NP classifiers tested through various tests and the results of the VSA model voting mechanism show that the three NP classifiers deployed by the umbrella algorithm have good misclassification constraint performance, so that the scheme provided by the method has high VSA quality and misclassification constraint capability, and the applicability and flexibility of the scheme are also proved.

Claims (4)

1. The power system voltage stability evaluation misclassification constraint method based on the umbrella algorithm is characterized by comprising the following steps of:

step 1: constructing an initial data set, and constructing a voltage stability safety classification label based on a voltage stability evaluation rule of the power system;

step 2: utilizing a linear and nonlinear relation exploration tool for feature selection, exploring the association degree between the operation variable of the initial sample set and the voltage stability security classification label, and selecting a key operation variable to construct an efficient sample set;

step 3: simultaneously deploying a plurality of types of Nawman pearson classifiers based on an umbrella algorithm to construct a voltage stability evaluation model, taking a high-efficiency sample set and a voltage stability security classification label corresponding to the high-efficiency sample set as model input, and performing offline training on the voltage stability evaluation model;

step 4: taking the influence of factors such as the change of the operation topological structure of the power system, the change of the power distribution between the generator and the load, the data loss of the phasor measurement unit and the like on the system into consideration, and sending a new sample set generated under the new operation working condition into a voltage stability evaluation model for model update;

step 5: based on the operation data of the power system collected by the phasor measurement unit of the wide area monitoring system in real time, selecting corresponding characteristics, and completing online voltage stability evaluation by using a trained voltage stability evaluation model;

in step 1, a voltage stability index VSI is constructed as shown in formula (1):

wherein: p (P) ₀ Initial load active power, P _max Active power is received for the corresponding maximum;

the VSI value represents the safety level of the power system, and varies from 0 to 100%, and in order to establish a safety classification rule, a suitable acceptable threshold η may be set for the VSI to clearly distinguish the voltage safety state, and then a voltage stable safety classification label is constructed as shown in formula (2):

2. the umbrella algorithm-based power system voltage stability evaluation misclassification constraint method of claim 1, wherein in step 1, operation data of a historical system operating point is acquired through real-time update and collection of power system PMUs on system operation data, an initial data set is constructed, voltage breakdown conditions of various operation conditions are simulated by using a continuous power flow CPF method, the operating point is gradually moved from a basic condition to a breakdown point, when the power system cannot meet the increasing load demand, a jacobian matrix of a power flow equation is singular, power flow cannot converge, the voltage breakdown point occurs, and the relation between the current operating point and the voltage of the breakdown point is quantified by using the load active power difference relation between the operating point and the voltage breakdown point.

3. The umbrella algorithm-based power system voltage stability evaluation misclassification constraint method of claim 1, wherein the method comprises the following steps of: in step 2, complex and variable relation of system operation variables are considered in the characteristic selection process, and the PCC with a linear relation exploration function and the MIC with a nonlinear relation exploration function are combined to serve as a characteristic selection method, so that the linear relation and the nonlinear relation between the operation variables and the voltage stability index are effectively explored, the screening of effective key operation variables is ensured, a high-efficiency sample set is constructed, and the problem of high dimension of data is solved.

4. The umbrella algorithm-based power system voltage stability evaluation misclassification constraint method of claim 3, wherein: constructing each feature of the ith operating point of the power system and a corresponding VSI thereof into a row vector F of a feature description set F _i ＝{x ₁ ,x ₂ ,...,x _m ,y _i I is more than or equal to 1 and less than or equal to m), the feature description set F= { X of the operation data of the n operation conditions ₁ ,X ₂ ,...,X _m Y }, wherein X _k Column vectors (1.ltoreq.k.ltoreq.n) composed of the same feature quantity under various operating conditions, Y= { Y ₁ ,y ₂ ,...,y _n Detecting the correlation between each operation variable and the VSI by PCC and MIC for the operation data variable set X and the corresponding VSI set Y of various operation conditions, and respectively screening out high-ranking linear relation operation variables and nonlinear operation variables highly related to the VSI;

exploring PCC for linear relations: let ρ (X, Y) be the pearson correlation coefficient of the feature quantity X, Y, the calculation formula is shown in formula (3):

wherein: n is the dimension of a single feature quantity;

is->

Average values of elements contained in X and Y respectively;

PCC has a value ranging from-1 to 1 and has several attributes:

1) ρ (X, Y) > 0 indicates that there is a positive correlation between X and Y;

2) ρ (X, Y) =0 illustrates the wireless correlation between X and Y;

3) ρ (X, Y) < 0 indicates that there is a negative correlation between X and Y;

4) The larger the absolute value of rho (X, Y), the stronger the linear correlation exists between X and Y;

exploring MIC for nonlinear relations: given a finite order vector data set d= { (x) _i ,y _i ) If the X and Y axes are divided into X and Y grids, respectively, obtaining an X Y grid G, and the variable values in D fall into the grid of G to obtain a corresponding probability distribution d| _G Wherein x and y are positive integers, and different mutual information values can be obtained by changing grid division positions on the premise of fixing the grid division number, wherein the maximum mutual information value is shown as the following formula (4):

I ^* (D,x,y)＝maxI(D| _G ) (4)

wherein: i (D|) _G ) Represents D| _G Mutual information between the inner data points;

to facilitate comparison between different dimensions, equation (4) is normalized as shown in equation (5) to have its value in interval [0,1]:

given an ordered pair data set D of sample size n, the MIC of two variables X, Y in the set is defined as shown in equation (6):

wherein: xy.ltoreq.B (n) { B (n) =n ^a Is generally set as n ^0.6 }；

The MIC ranges from 0 to 1 and has several attributes:

1) For two variables with a functional relationship that tends to be noiseless, their MIC values tend to be 1;

2) MIC values tend to be 1 for a broader class of noiseless relationships;

3) For two variables that are statistically independent of each other, the MIC value tends to be 0.

Images