CN110428005B - Umbrella algorithm-based dynamic security misclassification constraint method for power system - Google Patents
Umbrella algorithm-based dynamic security misclassification constraint method for power system Download PDFInfo
- Publication number
- CN110428005B CN110428005B CN201910705909.XA CN201910705909A CN110428005B CN 110428005 B CN110428005 B CN 110428005B CN 201910705909 A CN201910705909 A CN 201910705909A CN 110428005 B CN110428005 B CN 110428005B
- Authority
- CN
- China
- Prior art keywords
- classifier
- power system
- algorithm
- umbrella
- misclassification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000004422 calculation algorithm Methods 0.000 title claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 47
- 238000004088 simulation Methods 0.000 claims abstract description 7
- 238000013210 evaluation model Methods 0.000 claims abstract description 4
- 238000005259 measurement Methods 0.000 claims abstract description 4
- 230000008859 change Effects 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 238000012706 support-vector machine Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000007637 random forest analysis Methods 0.000 claims description 5
- 125000004122 cyclic group Chemical group 0.000 claims description 4
- 238000007477 logistic regression Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 3
- 230000007547 defect Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims 1
- 230000001360 synchronised effect Effects 0.000 abstract 1
- 238000011156 evaluation Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000001052 transient effect Effects 0.000 description 2
- 206010033799 Paralysis Diseases 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011056 performance test Methods 0.000 description 1
- 238000001558 permutation test Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Marketing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Tourism & Hospitality (AREA)
- Life Sciences & Earth Sciences (AREA)
- Development Economics (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Game Theory and Decision Science (AREA)
- Educational Administration (AREA)
- Probability & Statistics with Applications (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- General Health & Medical Sciences (AREA)
- Primary Health Care (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A power system dynamic security misclassification constraint method based on an umbrella algorithm comprises the following steps: constructing a corresponding dynamic security classification label based on the dynamic security classification rule; step 2: performing feature selection on an initial sample set constructed based on a historical operation database of the power system to obtain a key feature set; and 3, step 3: an on-line dynamic security evaluation model is established by a Neyman Pearson comprehensive classifier based on an umbrella algorithm, and a key feature set and a corresponding dynamic security classification label are used as input and are sent into the classifier for off-line training; and 4, step 4: considering various power system operation influence factors, performing time domain simulation to obtain an update sample set, and sending the update sample set into a comprehensive NP classifier to perform model update; and 5: and carrying out online DSA through synchronous phasor measurement unit data collected in real time. The invention can reduce the influence of DSA misclassification on the safe operation of the power system and reduce the social and economic losses.
Description
Technical Field
The invention belongs to the field of dynamic security assessment of an electric power system, and particularly relates to an umbrella algorithm-based dynamic security mis-classification constraint method for the electric power system.
Background
In recent years, with the continuous development of modern power systems, it has become one of the most complex and largest-scale artificial systems in the world. Meanwhile, the power system also faces a serious challenge in a new operating environment. On the one hand, due to the high degree of interconnection coupling of modern power systems, when disturbances and faults reach a certain level, the power systems may lose stability with serious catastrophic consequences. On the other hand, the operational risk of the power system is increased due to the continuous access of the distributed power sources and various uncertainty factors. If the operator can not quickly and accurately capture the accident safety alarm signal, accidents are easily caused, and the power system is paralyzed. Therefore, it is necessary to perform a reliable DSA.
Among them, there are two types of misclassifications for DSA: (1) a class of classification errors that determine an unstable state as a stable state; (2) the two categories of classification errors, which determine a stable state as an unstable state, are not identical in the degree of damage to the power system. Therefore, from the standpoint of the overall classification accuracy and the influence of misclassification consequences of the power system, it is worth studying to constrain a class of classification errors on the premise of keeping the overall classification accuracy of the system within an acceptable range, but the current DSA research has some limitations:
(1) many classifiers, such as Artificial Neural Networks (ANNs), decision trees ((Decision trees, DTs), support Vector Machine (SVM) classifiers, extreme Learning Machines (ELMs), and so on, are mostly focused on how to improve the overall classification accuracy of DSA, and treat one class of classification errors and two classes of classification errors equally, and cannot perform preferential constraint on the serious class of classification errors, so as to reduce the influence of one class of classification errors on the system (2) a general DSA scheme of an electric power system adopts a classifier to construct a DSA model, and an evaluation result is too monotonous and depends on the performance of the classifier on the system, so that there is a problem that applicability of different systems may not be the same (3) for DSA output, a traditional DSA model outputs a single result, and cannot accurately reflect the safety state of the system.
In summary, the existing DSA classification method cannot effectively restrict and evaluate the influence of misclassification on the safe operation of the power system, and reduce the risk of system operation.
Patent document with publication number CN107171315A discloses a transient stability evaluation method for power system based on RPTSVM, which adopts energy function index and power system index to construct an original feature set, reduces feature set dimension and reduces redundant information; the method comprises the steps of performing feature compression on an index of a power system and an index of a projection energy function through a maximum correlation minimum redundancy feature selection method, further reducing redundancy information and reducing feature dimension; the RPTSVM performs transient stability evaluation on the power system after adding a regular term reconstruction classifier into the optimized objective function, classifies the faults into corresponding classes according to the membership of the characteristic subset of the power system relative to the stable class and the unstable class of the power system aiming at the fault types of the actual power system, and determines that the faults are serious faults if the faults are classified into the unstable class; otherwise, determining the fault as a non-critical fault. It has the defects that:
(1) the fault misclassification severity of the power system cannot be differently constrained, and a misclassification threshold value is set, so that the serious class of misclassification is easily insufficiently constrained, and the safe operation risk of the system is increased;
(2) the comprehensive DSA model can not be constructed by providing a plurality of optimal classifiers with class-one classification error constraint for system operators, and the evaluation performance excellent degree depends on the RPTSVM model;
(3) integrated classification results cannot be provided and the reliability of the results cannot be ensured based only on the evaluated membership.
Disclosure of Invention
The invention provides a dynamic security misclassification constraint method of an electric power system based on a umbrella algorithm, which aims to solve the problem that a plurality of data mining tools and dynamic security evaluation schemes are limited to one class of classification misclassification constraints, so that a DSA mechanism is more reliable and flexible, and the constraint effect on one class of classification errors is ensured on the premise that the overall classification effect is acceptable.
In order to achieve the purpose of the invention, the invention adopts the following technical scheme:
a dynamic safety misclassification constraint method for an electric power system based on an umbrella algorithm comprises the following steps:
step 1: constructing a corresponding dynamic security classification label based on the dynamic security classification rule;
step 2: performing feature selection on an initial sample set constructed based on a historical operation database of the power system to obtain a key feature set;
and step 3: an online Dynamic Security Assessment (DSA) model is constructed by a Neyman-Pearson (NP) comprehensive classifier based on an umbrella algorithm, and a key feature set and a corresponding Dynamic Security classification label are used as input and sent to the classifier for offline training;
and 4, step 4: considering the influence of the operation topological structure change of the power system, the power distribution change among generators/loads and the load characteristic change on the system, performing time domain simulation to obtain an update sample set, and sending the update sample set into a comprehensive NP classifier to perform model update;
and 5: online DSA was performed by Phasor Measurement Units (PMUs) data collected in real time.
In step 1, the steps of constructing the corresponding dynamic security classification label are as follows:
(1) Calculating Critical Clearing Time (CCT) corresponding to various working conditions according to system fault flow simulation;
(2) According to the dynamic security classification rules: when the Actual Cutting Time (ACT) is less than or equal to CCT, the system is safe, and the classification label is 1; ACT is larger than CCT, the system is not safe, the classification label is 0, and a dynamic safety index and a corresponding classification label are constructed as shown in formulas (1) and (2);
in the formula: and m is a user-defined threshold.
In step 2, a DSA is directly carried out on the mass operation data of the power system, the defects of large data size, low calculation speed, time consumption and the like exist, and a nearest Neighbor Prediction Independence Test (BNNPT) method capable of exploring the nonlinear relation between the two variables is adopted as a feature selection tool to carry out data dimension reduction.
In step 3, the new sample set subjected to feature selection is divided into a training set and a testing set according to quintupling cross validation rules, and the training set is used for NP classifier training.
In step 3, aiming at a class of classification errors, a classifier is adopted, and an umbrella algorithm based on NP criterion is applied to a plurality of traditional classifiers: random Forest (RF), adaBoost, support Vector Machine (SVM), linear Discriminant Analysis (LDA), bayesian classifier (Naive Bayes, NB), nonparametric Bayes classifier (NNB), logistic Regression algorithm (Logistic Regression), penalty logic algorithm (Penlog) are modified, and a series of corresponding NP classifiers NP-RF, NP-ADA, NP-SVM, NP-LDA, NP-NB, NP-NNB, NP-Logistic, NP-Penlog can be obtained at the same time.
In step 3, the NP criterion-based umbrella algorithm modifies the traditional classifier to restrain a class of classification errors, and meanwhile, the NP classifier parameters are adjusted according to the actual requirements of the power system, a class-classification error threshold upper limit alpha, an irregularity rate delta and training set cycle splitting training times M are set, DSA is further controlled, and the effectiveness of class-classification error restraint is ensured.
In step 3, the key feature set and the corresponding dynamic security classification label are used as input and sent to the NP classifier training step as follows:
(1) According to the set M, the class 0 samples S in the training set 0 (misclassification sample judged to be stable in unstable condition) performing two-part random splitting to obtain samples
And samples
Binding to class 1 sample S 1 (misclassified samples that are stably judged to be unstable) are used for the basic classifier;
(2)
method for classifier to calculate upper limit upsilon of violation rate k Calculating a per-sample level threshold k * Corresponding score threshold
And then
Binding to S 1 Training each classifier scoring function f i ；
k * ＝min{k∈(1,...,n):υ(k)≤δ} (4)
(3) Will f is i Application to
Deriving a set of fractional threshold candidates τ i And will τ i Sorting the medium elements according to the ascending order to obtain sort (tau) i ) And find the rank threshold k * Corresponding score threshold
τ i ＝{t i,1 ,...,t i,n }＝{f i (x 1 ),...,f i (x n )} (5)
sort(τ i )＝{t i,(1) ,...,t i,(n) } (6)
(4) Constructing an NP classifier as shown in equation (8) based on the score function and the threshold:
(5) And (4) repeating the steps (1) to (4), performing cyclic splitting training on the training set for M times, constructing M NP classifiers, and outputting a class-I classification error result as a final result according to an integration method of weighted voting of a formula (9).
In step 4, the operation data caused by the operation influence factors of the power system is sent to the NP classifier trained in the off-line training stage, and the model is updated according to the NP classifier training process.
In step 4, the classifier performance is evaluated by using the values of ROC zone and lower zone Curve Area (AUC) of the Curve modification of the classical receiver operating characteristic (Price Rate of Change), and an NP classifier with excellent performance is selected to construct a DSA evaluation model.
In step 5, system operation data characteristics collected by the PMUs in real time are sent to a trained excellent NP classifier for online DSA, and the overall classification accuracy and the class-one classification error rate are calculated;
in the formula: f 11 ,F 10 ,F 00 ,F 01 The number of stable samples, the number of unstable samples, and the number of unstable samples are respectively determined as the number of stable samples, the number of unstable samples, and the number of unstable samples.
By adopting the technical scheme, the following technical effects can be brought:
(1) Compared with a general DSA model using a single classifier, the DSA model provided by the technical method can deploy a plurality of NP classifiers at the same time, effectively avoids the limitation of the single model in application to different systems, is convenient for operators to quickly and accurately screen the appropriate classifiers, and ensures the classification performance.
(2) The technical method utilizes NP criterion to modify the traditional classifier, and can set a threshold value of a class of classification errors. And distinguishing different influences of the first class classification error and the second class classification error on safe and stable operation and social economy of the power system. And (4) preferentially processing a class of classification errors with serious consequences on the DSA and controlling the class of classification errors within an allowable range. Meanwhile, in order to ensure the effectiveness of the threshold, the violation rate is set for the part exceeding the threshold to further restrict one class of classification errors, and the influence of the DSA result on the safe and stable operation of the power system is ensured to be minimum.
(3) The technical method is different from the traditional offline training process in order to ensure the reliability of the actual DSA result, and a similar integrated comprehensive learning method is applied to a group of training sets. A series of NP classifiers are constructed by distinguishing stable/unstable samples and utilizing a cyclic splitting training mode, and majority voting is adopted for the result of the NP classifiers. Therefore, the novel training set training mode can effectively keep the reliability of the output result of the class classification error.
Drawings
FIG. 1 is a flow chart of the overall scheme of the present invention;
FIG. 2 is a schematic diagram of the DSA model of the present invention;
FIG. 3 is a flow chart of the umbrella algorithm of the present invention;
FIG. 4 is a schematic diagram of the ROC bands of various NP classifiers of the present invention.
Detailed Description
The invention is further described with reference to the following figures and examples.
As shown in fig. 1, a power system dynamic security misclassification constraint method based on an umbrella algorithm includes the following steps:
step 1: constructing corresponding dynamic security classification labels based on dynamic security classification rules;
and 2, step: performing feature selection on an initial sample set constructed based on a historical operation database of the power system to obtain a key feature set;
and step 3: an online Dynamic Security Assessment (DSA) model is constructed by a Neyman-Pearson (NP) comprehensive classifier based on an umbrella algorithm, and a key feature set and a corresponding Dynamic Security classification label are used as input and sent into the classifier for offline training;
and 4, step 4: considering the influence of the operation topological structure change of the power system, the power distribution change among generators/loads and the load characteristic change on the system, performing time domain simulation to obtain an update sample set, and sending the update sample set into the comprehensive NP classifier to perform model update;
and 5: online DSA was performed by collecting Phasor Measurement Units (PMUs) data in real time.
In step 1 of fig. 1: for an emergency event involving clearing of a fault, in fault localization, a system with a longer CCT is considered more stable, and the relationship between CCT and ACT determines whether the system is stable. Based on the established system dynamic security classification rule, the corresponding dynamic security classification label is established as follows, which respectively correspond to the formulas (1) and (2).
(1) Calculating the limit cut-off time (CCT) corresponding to various working conditions according to the system fault load flow simulation;
(2) According to the dynamic security classification rule: when ACT is less than or equal to CCT, the system is safe, and the classification label is 1; and when the ACT is larger than the CCT, the system is unsafe, the classification label is 0, and a dynamic safety index and a corresponding classification label are constructed.
In the formula: and m is a user-defined threshold.
In step 2 of FIG. 1: based on the massive high-dimensional characteristics of the data of the power system, all dynamic data are directly subjected to DSA, and the problems of low calculation speed, time consumption, low learning efficiency and the like exist, and the requirement on the performance of a DSA model is high. In the case of a power system fault, the importance of the characteristic variables related to the safety state of the power system is different from the analysis of factors such as the electrical distance. In order to facilitate efficient and accurate DSA learning of the system, a BNNPT feature selection link is added in the design scheme, and feature selection is carried out on an initial sample set constructed based on a historical operation database of the power system to obtain a key feature set.
The BNNPT algorithm flow is as follows.
(1) Constructing nearest neighbor X for matrix vector X of length N Neighborhood (W rows and V columns) index matrix, and when X Neighborhood(i,j) Is away from X i When the nearest element is available, for each element Y of Y (W rows and V columns) j Constructing nearest neighbor Structure H i Calculating H i 。
H i ＝sum(Y Neighborhood(i,j) ,j)/bags (3)
In the formula: and the bags is the number of nearest neighbor structures constructed by the matrix.
(2) Calculating the square error SE = | | | H-Y | | non-calculation 2 And adopting SE test to predict the accuracy of Y by using X, carrying out permutation test by using SE as test statistic, and carrying out multiple random repeated tests on elements in Y. Calculating SE random The probability of SE ≦ is recorded as the P value.
(3) And screening out the characteristics of which the P value is ranked at the top 5% as key characteristic quantities to form a new key characteristic set.
In step 3 of fig. 1: in order to facilitate the realization of online DSA, an initial sample set formed based on historical data is used for feature selection, and the obtained key feature set and corresponding dynamic security classification labels are used as input and are sent to an NP comprehensive classifier based on an umbrella algorithm for offline training, as shown by a solid line large arrow flow in FIG. 2. Various traditional classifiers are classified by using an umbrella algorithm based on NP criteria: RF, adaBoost, SVM, LDA, NB, NNB, logistic and Penlog are transformed to obtain a series of corresponding NP classifiers NP-RF, NP-ADA, NP-SVM, NP-LDA, NP-NB, NP-NNB, NP-Logistic and NP-Penlog. The training process is shown in fig. 3, and the specific steps are as follows.
(1) According to the set M, the class 0 samples S in the training set 0 Performing two-part random splitting to obtain a sample
And a sample
(i.ltoreq.M). Binding class 1 sample S 1 For the basic classifier.
(2)
Method for classifier to calculate violation rate upper limit upsilon k Calculating a fractional threshold corresponding to each sample level threshold k
And then
Binding to S 1 Training each classifier score function f i ：
k * ＝min{k∈(1,...,n):υ(k)≤δ} (5)
(3) Will f is i Application to
Deriving a set of fractional threshold candidates τ i And will τ i Sorting the medium elements according to the ascending order to obtain sort (tau) i ) And find the rank threshold k * Corresponding score threshold
τ i ＝{t i,1 ,...,t i,n }＝{f i (x 1 ),...,f i (x n )} (6)
sort(τ i )＝{t i,(1) ,...,t i,(n) } (7)
(4) Constructing NP classifier φ based on score function and threshold i (X)：
(5) Repeating the steps (1) to (4), performing cyclic splitting training on the training set for M times to construct M NP classifiers, and outputting a class of classification error results as final results according to a weighted voting integration method:
in step 4 of FIG. 1: since the operation of the power system is not fixed, it is affected by various factors (system power distribution changes, changes in system topology, and changes in impedance characteristics). Therefore, DSA models based on offline training alone are not suitable for power systems that change in real time, and the DSA models are continually updated by sending new system operating data into the NP classifier. The DSA model can be suitable for more burst conditions through an updating mechanism. The steps are shown as a large arrow with a dotted line in fig. 2, and the DSA model is continuously updated according to the NP classifier training process.
Meanwhile, an ROC curve is adopted on the basis of classical classifier performance evaluation, the classifier performance is better according to the fact that the curve is closer to the upper left corner and the classifier performance is better according to the fact that the AUC of the area under the curve is larger, the specific ROC zone improved on the basis of the ROC curve and the corresponding AUC value aiming at the NP classifier are adopted to evaluate the classifier performance, and the NP classifier with excellent performance is selected to construct a DSA evaluation model.
In step 5 of FIG. 1: and (4) transmitting the system operation data characteristics collected by the PMUs in real time into a trained excellent NP classifier for online DSA, and calculating the overall classification accuracy and class-one classification error rate.
In the formula: f 11 ,F 10 ,F 00 ,F 01 The number of stable samples, the number of unstable samples, and the number of unstable samples are respectively determined as the number of stable samples, the number of unstable samples, and the number of unstable samples.
The embodiment is as follows:
the invention is tested in an IEEE39 node calculation system and a 500 node system, wherein the IEEE39 node system comprises 39 nodes and 10 generators, and simulates a three-phase short-circuit fault in a middle point of a line. And a 500 node system contains 500 nodes, 90 generators, 206 loads and 466 branches. The data generation and feature selection process was the same as for the IEEE39 bus test system, resulting in 5740 records. The test was performed on a computer equipped with an Intel Core i7 processor and 8GB of memory.
As shown in fig. 4 and table 1, the ROC bands of the 6 NP classifiers and the corresponding AUC values thereof are shown, it can be seen that the ROC bands of the 6 NP classifiers are all closer to the upper left corner, and the areas auc.u and auc.l of the upper and lower ROC bands are far greater than 0.5. Therefore, the overall classification performance of the six NP classifiers is better. In the DSA test of the IEEE39 node system and the 500 node system, three NP classifiers of NP-ADA, NP-RF and NP-SVM with better performance are selected. Simultaneously, α =0.02, δ =0.05 and M =1 are selected as classifier reference parameters. The test results based on NP-RF, NP-ADA and NP-SVM are shown in table 2. And respectively recording the classification precision and the class classification error rate of the system. It can be seen that the DSA model constructed by the NP classifier can ensure that one class of classification errors are small enough on the premise of meeting the classification precision, so that the influence on the DSA is minimal, and the provided DSA comprehensive scheme has a good application prospect.
TABLE 1 AUC values for different NP classifiers
| Area of ROC strip | NP-Penlog | NP-SVM | NP-ADA | NP-RF | NP-NB | NP-NNB |
| AUC.L | 0.9743 | 0.9691 | 0.9721 | 0.9737 | 0.8051 | 0.9146 |
| AUC.U | 0.9937 | 0.9967 | 0.9959 | 0.9975 | 0.8628 | 0.9549 |
TABLE 2 Performance test results of three NP classifiers in two systems under reference parameter conditions
As shown in table 3 and table 4, the classification error results of the first class when the threshold α and the violation rate δ of the first class of classification errors are adjusted to be 0.01, 0.03, 0.05, 0.07 and 0.09, respectively. It can be seen that as α increases, the accuracy of the evaluation increases within a certain range, and a class of classification errors can be effectively controlled within a corresponding threshold. And one class of classification errors can be further constrained by setting the violation rate under the reference condition. Therefore, a certain threshold value and a violation rate are set for a class of classification errors, which can be effective
TABLE 3 class of classification errors for NP classifiers under different threshold alpha conditions
TABLE 4 class of classification errors for NP classifiers under different violation Rate δ conditions
The occurrence rate of class classification errors is reduced, and the DSA evaluation performance is enhanced.
Table 5 shows a class of classification error results of training sets with different cycle split times when the training set split times M are adjusted to be 1, 3, 5, 7, and 9. It can be seen that as the number of training set splits increases, class-one classification errors gradually decrease and tend to stabilize. Therefore, the training mode of changing the training set has certain application prospect in the aspect of class classification errors of the constraint system.
TABLE 5 class-one classification errors of NP classifiers under different training set split times M conditions
TABLE 6 class-one misclassification comparison of conventional classifier and NP classifier
As shown in Table 6, in order to prove that the design scheme has the effect of restricting class I classification errors compared with the traditional DSA model, three NP classifiers are compared with classification results of the corresponding traditional classifiers, and the DSA model constructed by the NP classifier based on the umbrella algorithm has the function of restricting class I classification errors of DSA and is stable and reliable in classification results by analyzing the class I classification error rate.
Claims (10)
1. A power system dynamic security misclassification constraint method based on an umbrella algorithm is characterized by comprising the following steps:
step 1: constructing a corresponding dynamic security classification label based on the dynamic security classification rule;
step 2: performing feature selection on an initial sample set constructed based on a historical operation database of the power system to obtain a key feature set;
and 3, step 3: an online Dynamic Security Assessment (DSA) model is constructed by a Neyman-Pearson (NP) comprehensive classifier based on an umbrella algorithm, and a key feature set and a corresponding Dynamic Security classification label are used as input and sent into the classifier for offline training;
and 4, step 4: considering the influence of the operation topological structure change of the power system, the power distribution change among generators/loads and the load characteristic change on the system, performing time domain simulation to obtain an update sample set, and sending the update sample set into the comprehensive NP classifier to perform model update;
and 5: online DSA was performed by collecting Phasor Measurement Units (PMUs) data in real time.
2. The electric power system dynamic security misclassification constraint method based on the umbrella algorithm as claimed in claim 1, wherein: in step 1, the steps of constructing the corresponding dynamic security classification label are as follows:
(1) Calculating the Critical Clearing Time (CCT) corresponding to various working conditions according to system fault flow simulation;
(2) According to the dynamic security classification rules: when the Actual Cutting Time (ACT) is less than or equal to CCT, the system is safe, and the classification label is 1; ACT is larger than CCT, the system is not safe, the classification label is 0, and a dynamic safety index and a corresponding classification label are constructed as shown in formulas (1) and (2);
in the formula: and m is a user-defined threshold.
3. The electric power system dynamic safety misclassification constraint method based on the umbrella algorithm as claimed in claim 1, wherein: in step 2, a DSA is directly carried out on the mass operation data of the power system, the defects of large data size, low calculation speed, time consumption and the like exist, and a Nearest Neighbor Prediction Independence Test (BNNPT) method capable of exploring the nonlinear relation between the two variables is adopted as a feature selection tool to carry out data dimension reduction.
4. The electric power system dynamic safety misclassification constraint method based on the umbrella algorithm as claimed in claim 1, wherein: in step 3, the new sample set subjected to feature selection is divided into a training set and a testing set according to quintupling cross validation rules, and the training set is used for NP classifier training.
5. The electric power system dynamic security misclassification constraint method based on the umbrella algorithm as claimed in claim 1 or 4, wherein: in step 3, the umbrella algorithm based on NP criteria works on multiple traditional classifiers: random Forest (RF), adaBoost, support Vector Machine (SVM), linear Discriminant Analysis (LDA), bayesian classifier (Naive Bayes, NB), nonparametric Bayes classifier (NNB), logistic Regression algorithm (Logistic Regression), penalty logic algorithm (Penlog) are modified, and a series of corresponding NP classifiers NP-RF, NP-ADA, NP-SVM, NP-LDA, NP-NB, NP-NNB, NP-Logistic, NP-Penlog can be obtained at the same time.
6. The electric power system dynamic safety misclassification constraint method based on the umbrella algorithm as claimed in claim 5, wherein: in step 3, the NP criterion-based umbrella algorithm modifies the traditional classifier to restrain a class of classification errors, and meanwhile, the NP classifier parameters are adjusted according to the actual requirements of the power system, a class-classification error threshold upper limit alpha, an irregularity rate delta and training set cycle splitting training times M are set, DSA is further controlled, and the effectiveness of class-classification error restraint is ensured.
7. The electric power system dynamic safety misclassification constraint method based on the umbrella algorithm as claimed in claim 6, wherein: in step 3, the key feature set and the corresponding dynamic security classification label are used as input and sent to the NP classifier training step as follows:
(1) According to the set M, the class 0 samples S in the training set 0 Performing two-part random splitting to obtain a sample
And a sample
(i is less than or equal to M); binding class 1 sample S 1 For a base classifier; class 0 sample S 0 Misclassification samples that are determined to be unstable; class 1 sample S 1 Misclassified samples that are stably determined to be unstable;
(2)
method for classifier to calculate upper limit upsilon of violation rate k Calculating a per-sample level threshold k * Corresponding score threshold
And then
Binding to S 1 Training each classifier scoring function f i ；
k * ＝min{k∈(1,...,n):υ(k)≤δ} (4)
(3) Will f is mixed i Application to
Deriving a set of fractional threshold candidates τ i And will τ i Sorting the medium elements in ascending order to obtain sort (tau) i ) And find the rank threshold k * Corresponding score threshold
τ i ＝{t i,1 ,...,t i,n }＝{f i (x 1 ),...,f i (x n )} (5)
sort(τ i )＝{t i,(1) ,...,t i,(n) } (6)
(4) Constructing an NP classifier as shown in equation (8) based on the score function and the threshold:
(5) Repeating the steps (1) to (4), performing cyclic splitting training on the training set for M times, constructing M NP classifiers, and outputting a class of classification error results as final results according to an integration method of weighted voting of a formula (9);
8. the electric power system dynamic safety misclassification constraint method based on the umbrella algorithm as claimed in claim 7, wherein: in step 4, the operation data caused by the operation influence factors of the power system is sent to the NP classifier trained in the off-line training stage, and the model is updated according to the NP classifier training process.
9. The electric power system dynamic security misclassification constraint method based on the umbrella algorithm as claimed in claim 8, wherein: in step 4, the performance of the classifier is evaluated by using the ROC band and the lower band Curve Area (AUC) values of the classic receiver operating characteristic (Price Rate of Change, ROC) Curve transformation, and an NP classifier with excellent performance is selected to construct a DSA evaluation model.
10. The electric power system dynamic safety misclassification constraint method based on the umbrella algorithm as claimed in claim 1, 2, 3, 4, 6, 7 or 9, wherein: in step 5, system operation data characteristics collected by the PMUs in real time are sent to a trained excellent NP classifier for online DSA, and the overall classification accuracy and the class-one classification error rate are calculated;
in the formula: f 11 ,F 10 ,F 00 ,F 01 The number of stable samples, the number of unstable samples, and the number of unstable samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910705909.XA CN110428005B (en) | 2019-07-31 | 2019-07-31 | Umbrella algorithm-based dynamic security misclassification constraint method for power system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910705909.XA CN110428005B (en) | 2019-07-31 | 2019-07-31 | Umbrella algorithm-based dynamic security misclassification constraint method for power system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110428005A CN110428005A (en) | 2019-11-08 |
| CN110428005B true CN110428005B (en) | 2022-11-08 |
Family
ID=68413625
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910705909.XA Active CN110428005B (en) | 2019-07-31 | 2019-07-31 | Umbrella algorithm-based dynamic security misclassification constraint method for power system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110428005B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11947643B2 (en) * | 2019-12-26 | 2024-04-02 | Rakuten Group, Inc. | Fraud detection system, fraud detection method, and program |
| CN111523785A (en) * | 2020-04-16 | 2020-08-11 | 三峡大学 | Power system dynamic security assessment method based on generation countermeasure network |
| CN111585277B (en) * | 2020-05-19 | 2022-04-08 | 三峡大学 | Power system dynamic security assessment method based on hybrid integration model |
| CN111652478B (en) * | 2020-05-19 | 2023-06-02 | 三峡大学 | Umbrella algorithm-based power system voltage stability evaluation misclassification constraint method |
| CN111651932A (en) * | 2020-05-19 | 2020-09-11 | 三峡大学 | Online dynamic security assessment method for power system based on integrated classification model |
| CN111628531B (en) * | 2020-05-19 | 2022-04-08 | 三峡大学 | Data driving method for static voltage stability evaluation of power system |
| CN111814394B (en) * | 2020-06-30 | 2023-08-25 | 三峡大学 | Power system safety assessment method based on correlation and redundancy detection |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105512799A (en) * | 2015-11-26 | 2016-04-20 | 中国电力科学研究院 | Mass online historical data-based power system transient stability evaluation method |
| CN109726766A (en) * | 2019-01-04 | 2019-05-07 | 三峡大学 | A kind of online dynamic secure estimation method of electric system based on Integrated Decision tree |
| CN109829627A (en) * | 2019-01-04 | 2019-05-31 | 三峡大学 | A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080133434A1 (en) * | 2004-11-12 | 2008-06-05 | Adnan Asar | Method and apparatus for predictive modeling & analysis for knowledge discovery |
-
2019
- 2019-07-31 CN CN201910705909.XA patent/CN110428005B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105512799A (en) * | 2015-11-26 | 2016-04-20 | 中国电力科学研究院 | Mass online historical data-based power system transient stability evaluation method |
| CN109726766A (en) * | 2019-01-04 | 2019-05-07 | 三峡大学 | A kind of online dynamic secure estimation method of electric system based on Integrated Decision tree |
| CN109829627A (en) * | 2019-01-04 | 2019-05-31 | 三峡大学 | A kind of safe confidence appraisal procedure of Electrical Power System Dynamic based on integrated study scheme |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110428005A (en) | 2019-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110428005B (en) | Umbrella algorithm-based dynamic security misclassification constraint method for power system | |
| CN101464964B (en) | Pattern recognition method capable of holding vectorial machine for equipment fault diagnosis | |
| Pinzón et al. | Real-time multi-state classification of short-term voltage stability based on multivariate time series machine learning | |
| Zareipour et al. | Classification of future electricity market prices | |
| CN112508442B (en) | Transient stability assessment method and system based on automatic and interpretable machine learning | |
| CN109740859A (en) | Transformer condition evaluation and system based on Principal Component Analysis and support vector machines | |
| CN111652478B (en) | Umbrella algorithm-based power system voltage stability evaluation misclassification constraint method | |
| Mukherjee et al. | Development of an ensemble decision tree-based power system dynamic security state predictor | |
| CN111400966B (en) | Static voltage stability evaluation method of power system based on improved AdaBoost | |
| CN110705831A (en) | Power angle instability mode pre-judgment model construction method after power system fault and application thereof | |
| CN104268576A (en) | Electric system transient stability classification method based on TNN-SVM | |
| Wang et al. | Transient stability assessment of a power system using multi-layer SVM method | |
| Arefi et al. | Ensemble adaptive neuro fuzzy support vector machine for prediction of transient stability | |
| CN111651932A (en) | Online dynamic security assessment method for power system based on integrated classification model | |
| Ramirez-Gonzalez et al. | Convolutional neural nets with hyperparameter optimization and feature importance for power system static security assessment | |
| CN111585277B (en) | Power system dynamic security assessment method based on hybrid integration model | |
| Ramirez-Gonzalez et al. | Convolutional neural network based approach for static security assessment of power systems | |
| Lin et al. | A data-driven scheme based on sparse projection oblique randomer forests for real-time dynamic security assessment | |
| Javan et al. | On-line voltage and power flow contingencies ranking using enhanced radial basis function neural network and kernel principal component analysis | |
| CN111814394B (en) | Power system safety assessment method based on correlation and redundancy detection | |
| Saeh et al. | Artificial neural network for power system static security assessment: a survey | |
| Massaoudi et al. | Short-term dynamic voltage stability status estimation using multilayer neural networks | |
| Zamzam et al. | A two-stage CNN-LSTM model-based transient stability assessment for power system | |
| Dharmapala et al. | Short-term voltage instability prediction using pre-identified voltage templates and machine learning classifiers | |
| Awadalla et al. | Classification of faults in nuclear power plant |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Liu Songkai Inventor after: Wang Feng Inventor after: Wang Can Inventor after: Zhao Ping Inventor after: Wen Bin Inventor after: Mao Dan Inventor after: Liu Lihuang Inventor after: Zhang Lei Inventor after: Zhong Hao Inventor after: Zhang Tao Inventor after: Li Zhenhua Inventor after: Ye Jing Inventor after: Li Peng Inventor before: Liu Songkai Inventor before: Wang Feng Inventor before: Wang Can Inventor before: Zhao Ping Inventor before: Wen Bin Inventor before: Li Dan Inventor before: Liu Lihuang Inventor before: Zhang Lei Inventor before: Zhong Hao Inventor before: Zhang Tao Inventor before: Li Zhenhua Inventor before: Ye Jing Inventor before: Li Peng |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |