CN109086913B - Power system transient stability assessment method and system based on deep learning - Google Patents
Power system transient stability assessment method and system based on deep learning Download PDFInfo
- Publication number
- CN109086913B CN109086913B CN201810759783.XA CN201810759783A CN109086913B CN 109086913 B CN109086913 B CN 109086913B CN 201810759783 A CN201810759783 A CN 201810759783A CN 109086913 B CN109086913 B CN 109086913B
- Authority
- CN
- China
- Prior art keywords
- transient stability
- noise reduction
- expected accident
- model
- reduction automatic
- 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
- 230000001052 transient effect Effects 0.000 title claims abstract description 162
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000013135 deep learning Methods 0.000 title claims abstract description 24
- 238000013097 stability assessment Methods 0.000 title claims description 30
- 230000009467 reduction Effects 0.000 claims abstract description 51
- 238000000605 extraction Methods 0.000 claims abstract description 45
- 238000012706 support-vector machine Methods 0.000 claims abstract description 36
- 238000011156 evaluation Methods 0.000 claims abstract description 33
- 238000013210 evaluation model Methods 0.000 claims description 23
- 230000006870 function Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 17
- 238000005259 measurement Methods 0.000 claims description 8
- 230000014509 gene expression Effects 0.000 claims description 7
- 239000013598 vector Substances 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 4
- 230000006978 adaptation Effects 0.000 claims description 3
- 230000006872 improvement Effects 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 230000003466 anti-cipated effect Effects 0.000 claims description 2
- 238000013136 deep learning model Methods 0.000 abstract description 8
- 238000005516 engineering process Methods 0.000 abstract description 5
- 238000007726 management method Methods 0.000 description 7
- 238000012854 evaluation process Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000001960 triggered effect Effects 0.000 description 3
- 238000007418 data mining Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 230000003449 preventive effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000001687 destabilization Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- 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)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Development Economics (AREA)
- Tourism & Hospitality (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Operations Research (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Educational Administration (AREA)
- Life Sciences & Earth Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Primary Health Care (AREA)
- General Health & Medical Sciences (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a transient stability evaluation method and a transient stability evaluation system based on deep learning, wherein the transient stability evaluation method comprises the following steps: constructing a transient stability evaluation original feature set; establishing an offline expected accident set; self-adaptively establishing a stacked noise reduction automatic encoder feature extraction model based on a mutual information theory to realize the multi-layer feature extraction of the original input; establishing an integrated regression model of a support vector machine, and judging transient stability margins of various operation modes under a fixed expected accident; generating a plurality of future operation modes and an online expected accident set; and predicting transient stability margin indexes by using the support vector machine integrated regression model corresponding to each expected accident, and grading the severity of the transient stability degree of the system in different operation modes. The method is combined with the existing distributed parallel computing technology, so that the optimal network structure of the deep learning model with a large number of network nodes and a plurality of hidden layers can be determined in a self-adaptive mode, the transient stability degree of the power system can be evaluated quickly and accurately, and the online application requirement is met.
Description
Technical Field
The invention belongs to the field of dynamic security assessment of power systems, and particularly relates to a method and a system for assessing transient stability of a power system based on deep learning.
Background
In recent years, the scale of an extra-high voltage alternating current-direct current hybrid power grid in China is continuously enlarged, high-permeability intermittent new energy power generation and massive flexible load response aggravate uncertainty of power grid source load and load on both sides, the operation mode and dynamic behavior of a large power grid are gradually complicated, the frequent occurrence of natural disasters makes a possible predicted accident scene more complicated, and higher requirements are provided for the overall planning decision level and the cooperative control capability of scheduling operation. Meanwhile, with the continuous deepening of the construction of the smart power grid, a large amount of scheduling operation data are accumulated in each level of regulation and control center, the dynamic safety assessment of the power system based on the data has the advantages of high online calculation speed, easiness in generation of heuristic rules for decision making and the like, and can form good complementation with the traditional model-based safety and stability analysis method. The online safety and stability assessment based on the data mining technology provides a new idea for intelligent scheduling control of the large power grid.
Most of the existing data mining methods for transient stability assessment are shallow learning models, such as support vector machines, decision trees, extreme learning machines and the like, and have the limitations of limited representation capability on complex functions, high computational complexity and poor generalization capability. Therefore, deep learning models are introduced into transient stability assessment problems, such as deep belief networks, stacked autoencoders, and the like. The deep learning approaches a complex function by utilizing a multilayer nonlinear network structure, learns the distributed feature representation of input data, and has the capability of extracting essential features from a small sample set. However, the existing transient stability evaluation model based on deep learning has the following disadvantages: firstly, how to adaptively determine an optimal network structure of a deep learning model with a large number of network nodes and a plurality of hidden layers; second, the stability of the system was not evaluated.
In view of the foregoing, there is a need for a method and system for adaptively determining an optimal network structure of a deep learning model having a large number of network nodes and a plurality of hidden layers, and evaluating a transient stability degree of a power system.
Disclosure of Invention
In order to solve the deficiencies of the prior art, a first object of the present invention is to provide a transient stability assessment method based on deep learning, which can adaptively determine an optimal network structure of a deep learning model with a large number of network nodes and a plurality of hidden layers, and can assess the transient stability degree of a power system.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention discloses a transient stability evaluation method based on deep learning, which comprises the following steps:
constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and predicting transient stability margin indexes of various operation modes under a fixed expected accident;
generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
and predicting transient stability margin indexes by using a support vector machine integrated regression model corresponding to each fixed expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on a utility theory.
Further, the transient stability assessment raw feature set comprises: pre-failure system characteristics, individual characteristics within the failure point and its vicinity, and the location and duration of the anticipated accident.
Further, the method adaptively establishes a stacked noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory.
The hidden layer output of the stacking noise reduction automatic encoder is abstract expression of original input in different layers, the number of hidden layer nodes determines the dimension of extracted features, the number of hidden layers determines the abstract degree of the extracted features, and the weight vectors between the hidden layers convert the input features into more abstract expression, so that the input features are filtered, the automatic extraction of multiple layers of features can be realized, and the objectivity of the feature extraction process is ensured. The MIFS information measurement standard is used for measuring the correlation between the hidden node weight of the stacked noise reduction automatic encoder and the original input and the redundancy between the hidden node weights, so that the purpose of building a simplified network structure is achieved.
Further, the transient stability margin indicator is a difference between a limit cut time and a fault cut time.
Further, based on the utility theory, the severity of the transient stability of the system under different operation modes is graded, specifically:
wherein, T is a set threshold value, and M is a transient stability margin index;
when the transient stability margin index is larger than T, the system is considered to have no transient instability risk; when the transient stability margin index is less than 0, the system is considered to have transient instability; when the transient stability margin indicator belongs to the interval [0, T ], an exponential function is used as a severity function within the interval.
Further, the severity of the sample was classified into 5 grades, let SrRepresenting the severity rating, the severity rating rules are as follows:
wherein, the sample with the severity level of 3 is positioned near the boundary of the security domain, and the transient stability margin is lower; samples with a severity level of 4 included destabilized samples and critically stable samples.
Further, still include: continuously acquiring online operation data from an energy management system, and updating the structure and parameters of a transient stability evaluation model, including updating the structure and parameters of a stacked noise reduction automatic encoder feature extraction model and updating the structure and parameters of a support vector machine integration model;
and when the on-line forecast accidents occur in a set of untrained forecast accidents, training a transient stability assessment model under the accidents on line.
The second purpose of the invention is to disclose a power system transient stability evaluation system based on deep learning, comprising:
an original feature set construction module configured to: constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
a sample set generation module configured to: establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
a stacked denoising autoencoder feature extraction model adaptation building module configured to: self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
a transient stability assessment model building module configured to: taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and predicting transient stability margins of various operation modes under a fixed expected accident;
a future execution scenario generation module configured to: generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
a transient stability margin prediction and severity ranking module configured to: and predicting the transient stability margin by using the support vector machine integrated regression model corresponding to each expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on the utility theory.
Further, still include: a model online update module configured to: and continuously acquiring online operation data from the energy management system, updating the structure and parameters of the transient stability evaluation model, and training the transient stability evaluation model under the accident on line when the untrained forecast accident occurs in the online forecast accident set.
The third objective of the present invention is to disclose a deep learning based transient stability assessment system for an electric power system, comprising a server, wherein the server comprises a memory, a processor and a computer program stored in the memory and capable of running on the processor, and the processor executes the computer program to implement the following steps:
constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and predicting transient stability margin indexes of various operation modes under a fixed expected accident;
generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
and predicting transient stability margin indexes by using a support vector machine integrated regression model corresponding to each fixed expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on a utility theory.
The invention has the beneficial effects that:
the transient stability evaluation process based on deep learning is divided into 3 links. And an offline training link randomly generates a sample set and a corresponding offline expected fault set, model training is respectively carried out on the sample set under each fault, and a feature extraction model and a transient stability evaluation model under each fault are adaptively established. The starting mode of the real-time application link is triggered periodically (15min), a plurality of future operation modes and an online expected accident set are generated based on the online operation mode, and transient stability assessment is carried out by using a model corresponding to each expected accident. And in the online updating step, online operation data is acquired from the energy management system, the structure and parameters of the transient stability evaluation model are updated, and when untrained forecast accidents occur in the online forecast accident set, the transient stability evaluation model under the accidents is trained online.
According to the method, the self-adaptive establishment process and the transient stability evaluation process of the stacked noise reduction automatic encoder feature extraction model corresponding to each fault are independent, the 3 links can be combined with the existing distributed parallel computing technology, the optimal network structure of the deep learning model with a large number of network nodes and a plurality of hidden layers can be determined in a self-adaptive mode, the transient stability degree of the power system can be evaluated quickly and accurately, and the online application requirements are met.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.
FIG. 1 is a flow chart of a transient stability assessment method based on deep learning according to the present invention;
FIG. 2 is a schematic diagram of a feature extraction model of a stacked noise reduction auto-encoder according to the present invention;
FIG. 3 is a schematic diagram of an integrated regression model of the support vector machine of the present invention;
FIG. 4 is a schematic diagram of the transient stability evaluation system based on deep learning according to the present invention.
Detailed Description
It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
The method comprises three links for predicting the transient stability margin of the power system. The off-line training link realizes off-line training of the transient stability assessment model, comprises steps 1-4, the real-time application link utilizes the trained transient stability assessment model to perform real-time assessment, and comprises steps 5-6, and the on-line updating link realizes on-line updating of the transient stability assessment model to enhance generalization capability, and comprises step 7, as shown in the attached figure 1.
An off-line training link: off-line training for implementing transient stability assessment model
Step 1: and constructing a group of transient stability evaluation primitive feature sets according to the operating characteristics of the power system. The transient stability evaluation primitive feature set is composed of three parts of features.
The dynamic characteristics have high requirements on a dynamic phasor measurement system, and in order to avoid transient stability calculation during stability judgment, the original characteristic set selects static characteristics. Selecting the system characteristics before the fault as a part of the original characteristic set, and representing the overall operation condition of the power system; if all the single-machine characteristics are selected, the problem of dimension disaster can occur, and the more the fault information is contained in the area which is closer to the fault point electrically is inspired by the theory of concentric relaxation, so the single-machine characteristics in the fault point and the adjacent area are selected as a part of the original characteristic set; since only the three-phase short-circuit fault with the most severe fault consequences is considered, the location and duration of the expected accident is selected as part of the original feature set.
Step 2: and (3) considering factors such as a power grid topological structure, generator output and load change, establishing an offline expected fault set, and randomly generating a corresponding sample set for each expected accident.
And step 3: based on mutual information theory, a stacking noise reduction automatic encoder feature extraction model under the security domain concept is built in a self-adaptive mode, and original input multi-layer feature extraction is automatically achieved, as shown in fig. 2.
Mutual information measures the degree of interdependence of two variables and represents the content of shared information between the two variables. Assume that the sample set S has n-dimensional features (f)1,f2,…,fn) Of N samples, P (f)i) Is a characteristic fiThe probability of different possible values. f. ofiThe uncertainty of the value is determined by the information entropyAnd (3) measurement:
on the basis of the above two characteristics fiAnd ftMutual information between can be defined as
I(fi；ft)＝H(fi)+H(ft)-H(fi；ft)
Let D be each feature f in the sample set SiThe mean value of mutual information with the category label C represents the relevance of the feature set and the corresponding category; r is the size of mutual information among the features in the sample set S, and represents the redundancy among the features:
using the MIFS information metric, the evaluation function J (f) is
max J(f)＝D-βR
Wherein β is an adjusting coefficient, and the algorithm performance is better when β epsilon [0.5,1 ].
The hidden layer output of the stacking noise reduction automatic encoder is abstract expression of original input in different layers, the number of hidden layer nodes determines the dimension of extracted features, the number of hidden layers determines the abstract degree of the extracted features, and the weight vectors between the hidden layers convert the input features into more abstract expression, so that the input features are filtered, the automatic extraction of multiple layers of features can be realized, and the objectivity of the feature extraction process is ensured. The MIFS information measurement standard is used for measuring the correlation between the hidden node weight of the stacked noise reduction automatic encoder and the original input and the redundancy between the hidden node weights, so that the purpose of building a simplified network structure is achieved.
The specific process of establishing the stacking noise reduction automatic encoder model based on the mutual information comprises the following steps:
the first step is as follows: defining an initial stacking noise reduction automatic encoder network structure with a smaller scale, wherein the number of nodes of an input layer is equal to the input characteristic dimension; while setting J (f) threshold.
The second step is that: and determining the number of nodes of the ith hidden layer. Adding m nodes, carrying out weight initialization on the whole, and calculating J (f) of the weight vector of the existing node.
The third step: it is determined whether J (f) is greater than a threshold. If the threshold value is smaller than the threshold value, returning to the second step; otherwise, deleting the newly added node and carrying out the fourth step.
The fourth step: it is determined whether m is equal to 1. If m is not equal to 1, increasing the nodes by adopting a step-variable method,
returning to the second step for rounding down; if m is equal to 1, subtracting 1 from the current node number is the number of hidden nodes.
The fifth step: and operating the built i hidden-layer stacking noise reduction automatic encoders, and judging whether the transient stability evaluation effect is improved, namely whether the transient stability margin prediction precision is improved. If the improvement is good, i is i +1, returning to the step 2; if not, the i-1 hidden layers are the final network structure of the stacking noise reduction automatic encoder.
And a sixth step: and (3) using i-1 hidden layer outputs of the stacked noise reduction automatic encoder, namely i-1 feature expressions with different abstract degrees as the input of subsequent transient stability evaluation for subsequent security risk situation perception of the large power grid.
And 4, step 4: an integrated regression model of the support vector machine under the security domain concept is established based on an integrated learning method, as shown in fig. 3.
Given sample set { xi1,2, …, N, using a stacked noise reduction automatic encoder as a trainable feature extraction tool, assuming that the stacked noise reduction automatic encoder has N hidden layers, obtaining a hierarchical feature hj(j＝1,2,…,N)，hjThe feature dimension of the hidden layer is the same as the number of hidden layer nodes; usage feature set hjTraining pairThe vector machine model is supported, namely a different basis regression device is constructed by adopting different input feature sets; the output of the sub-learner is integrated using an "averaging method".
Output y of support vector machine integrated regression modelI：
In the formula: y isiAnd predicting the transient stability margin of the ith sub-regressor.
And (3) real-time application links: and carrying out real-time evaluation by using the trained transient stability evaluation model.
And 5: and generating a plurality of future operation modes and an online forecast accident set by using plan data (such as a section power plan, a maintenance plan and the like) and prediction data (such as ultra-short-term load prediction and the like) based on the online operation mode.
Step 6: and predicting the transient stability margin by using the support vector machine integrated regression model corresponding to each expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on the utility theory.
Adopting a transient stability margin index M based on critical clearing time:
M＝tCCT-tcl
in the formula: t is tclIs the time of fault removal, tCCTIs the critical clearing time.
When M >0, the system stabilizes, otherwise destabilizes. Obtaining the transient margins of the samples in the three partitions by using the method shown in FIG. 1, and constructing a severity function based on an effect theory: setting a threshold value T, and when M is larger than T, determining that the system has no risk of transient instability and the severity function value is 0; when M is less than 0, the system is considered to have transient instability, and the severity function value is 3; when M belongs to the interval [0, T ], an exponential function is used as the severity function within the interval:
Sm＝a1e-M+a2
in the formula: a is1And a2Is coefficient of。
Since the severity function is a continuous function, the functions pass through two points (T,0) and (0,3), and the coordinates of the two points are substituted to obtain an exponential function:
in summary, a severity function of the transient margin is obtained:
the severity of the sample was divided into 5 grades, let SrRepresenting the severity rating, the severity rating rules herein are obtained:
the sample with the severity level of 3 is positioned near the boundary of the security domain, and the transient stability margin is lower; samples with a severity level of 4 included destabilized samples and critically stable samples. And during actual scheduling operation, preferentially presenting the operation modes with the severity grades of 3 and 4 to scheduling personnel, and providing reference information for the scheduling personnel to make preventive control measures aiming at the high-risk operation modes.
And (3) an online updating link: and realizing the online updating of the transient stability evaluation model.
And 7: and acquiring online operation data from the energy management system, and updating the structure and parameters of the transient stability evaluation model. And when the on-line forecast accidents occur in a set of untrained forecast accidents, training a transient stability assessment model under the accidents on line.
Updating the structure of the transient stability evaluation model refers to continuously acquiring new online operation data from the EMS to update the network structure of the transient stability model during online operation; updating the structure and parameters of the transient stability assessment model includes: the method comprises the steps of updating the structure and parameters of a feature extraction model of the stacking noise reduction automatic encoder and updating the structure and parameters of an integrated model of a support vector machine.
The transient stability evaluation process based on deep learning is divided into 3 links. And an offline training link randomly generates a sample set and a corresponding offline expected fault set, and model training is respectively carried out on the sample set under each fault to obtain a feature extraction model and a transient stability evaluation model under each fault. The starting mode of the real-time application link is triggered periodically (15min), a plurality of future operation modes and an online expected accident set are generated based on the online operation mode, and transient stability assessment is carried out by using a model corresponding to each expected accident. And in the online updating step, online operation data is acquired from the energy management system, the structure and parameters of the transient stability evaluation model are updated, and when untrained forecast accidents occur in the online forecast accident set, the transient stability evaluation model under the accidents is trained online. According to the method, the self-adaptive establishment process and the transient stability evaluation process of the stacked noise reduction automatic encoder feature extraction model corresponding to each fault are independent, the 3 links can be combined with the existing distributed parallel computing technology, the optimal network structure of the deep learning model with a large number of network nodes and a plurality of hidden layers can be determined in a self-adaptive mode, the transient stability degree of the power system can be evaluated quickly and accurately, and the online application requirements are met.
The invention relates to a power system transient stability evaluation system based on deep learning, as shown in fig. 4, comprising:
(1) an original feature set construction module configured to: constructing a group of transient stability evaluation primitive feature sets according to the operating characteristics of the power system;
in the original feature set construction module, the transient stability evaluation original feature set is composed of three parts of features.
The dynamic characteristics have high requirements on a dynamic phasor measurement system, and in order to avoid transient stability calculation during stability judgment, the original characteristic set selects static characteristics. Selecting the system characteristics before the fault as a part of the original characteristic set, and representing the overall operation condition of the power system; if all the single-machine characteristics are selected, the problem of dimension disaster can occur, and the more the fault information is contained in the area which is closer to the fault point electrically is inspired by the theory of concentric relaxation, so the single-machine characteristics in the fault point and the adjacent area are selected as a part of the original characteristic set; since only the three-phase short-circuit fault with the most severe fault consequences is considered, the location and duration of the expected accident is selected as part of the original feature set.
(2) A sample set generation module configured to: considering factors such as a power grid topological structure, generator output, load change and the like, establishing an offline expected fault set, and randomly generating a corresponding sample set for each expected accident;
(3) a stacked denoising autoencoder feature extraction model adaptation building module configured to: self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
in the module for establishing the characteristic extraction model of the stacked noise reduction automatic encoder, the correlation between the hidden layer node weight of the stacked noise reduction automatic encoder and the original input and the redundancy between the hidden layer node weights are measured by using an MIFS information measurement standard, so that the aim of establishing a simplified network structure is fulfilled. The feature extraction model under the safety domain concept is to establish a stacked noise reduction automatic encoder feature extraction model for each expected accident by taking tidal current before system failure as an input feature.
(4) A transient stability assessment model building module configured to: establishing a support vector machine integration regression model under a security domain concept based on an integration learning method;
in the transient stability evaluation model establishing module, the support vector machine integrated regression model under the safety domain concept is to use the features of each layer extracted by the stacking noise reduction automatic encoder as the input of the support vector machine sub-learner to judge the transient stability of various operation modes under a fixed expected accident, and if the system has a destabilization risk, preventive control measures can be taken in time.
(5) A future execution scenario generation module configured to: generating a plurality of future operation modes and an online expected accident set by using plan data (such as a section power plan, a maintenance plan and the like) and prediction data (such as ultra-short-term load prediction and the like) based on an online operation mode;
(6) a transient stability margin prediction and severity ranking module configured to: and predicting the transient stability margin by using the support vector machine integrated regression model corresponding to each expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on the utility theory.
In the transient stability margin prediction and severity ranking module, a transient stability margin indicator is a difference between a limit removal time and a fault removal time; and an exponential function is constructed as a severity function based on an effect theory, so that the psychological bearing capacity of system operators on transient stability margin change is embodied, and the actual situation of the power system is met.
(7) A model online update module configured to: and acquiring online operation data from the energy management system, and updating the structure and parameters of the transient stability evaluation model. And when the on-line forecast accidents occur in a set of untrained forecast accidents, training a transient stability assessment model under the accidents on line.
Updating the structure of the transient stability evaluation model refers to continuously acquiring new online operation data from the EMS to update the network structure of the transient stability model during online operation; updating the structure and parameters of the transient stability assessment model includes: the method comprises the steps of updating the structure and parameters of a feature extraction model of the stacking noise reduction automatic encoder and updating the structure and parameters of an integrated model of a support vector machine.
The transient stability evaluation process based on deep learning is divided into 3 links. And an offline training link randomly generates a sample set and a corresponding offline expected fault set, model training is respectively carried out on the sample set under each fault, and a feature extraction model and a transient stability evaluation model under each fault are adaptively established. The starting mode of the real-time application link is triggered periodically (15min), a plurality of future operation modes and an online expected accident set are generated based on the online operation mode, and transient stability assessment is carried out by using a model corresponding to each expected accident. And in the online updating step, online operation data is acquired from the energy management system, the structure and parameters of the transient stability evaluation model are updated, and when untrained forecast accidents occur in the online forecast accident set, the transient stability evaluation model under the accidents is trained online. According to the method, the self-adaptive establishment process and the transient stability evaluation process of the stacked noise reduction automatic encoder feature extraction model corresponding to each fault are independent, the 3 links can be combined with the existing distributed parallel computing technology, the optimal network structure of the deep learning model with a large number of network nodes and a plurality of hidden layers can be determined in a self-adaptive mode, the transient stability degree of the power system can be evaluated quickly and accurately, and the online application requirements are met.
The invention further discloses a power system transient stability evaluation system based on deep learning, which comprises a server, wherein the server comprises a memory, a processor and a computer program which is stored on the memory and can run on the processor, and the processor executes the program to realize the following steps:
constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and judging the transient stability margin of various operation modes under a fixed expected accident;
generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
and predicting transient stability margin indexes by using a support vector machine integrated regression model corresponding to each expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on a utility theory.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (9)
1. A transient stability assessment method based on deep learning is characterized by comprising the following steps:
constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and predicting transient stability margin indexes of various operation modes under a fixed expected accident;
generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
predicting transient stability margin indexes by using a support vector machine integrated regression model corresponding to each fixed expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on a utility theory;
the self-adaptive establishment of the stacking noise reduction automatic encoder feature extraction model for each expected accident based on the mutual information theory specifically comprises the following steps:
the first step is as follows: defining an initial stacking noise reduction automatic encoder network structure with a smaller scale, wherein the number of nodes of an input layer is equal to the input characteristic dimension; simultaneously setting a threshold value of an evaluation function J (f) based on the MIFS information measurement standard;
the second step is that: determining the number of nodes of the ith hidden layer; newly adding m nodes, carrying out weight initialization on the whole, and calculating J (f) of the weight vector of the existing node;
the third step: determining whether J (f) is greater than a threshold; if the threshold value is smaller than the threshold value, returning to the second step; if not, deleting the newly added node and carrying out the fourth step;
the fourth step: judging whether m is equal to 1; if m is not equal to 1, increasing the nodes by adopting a step-variable method,
returning to the second step for rounding down; if m is equal to 1, subtracting 1 from the current node number to obtain the hidden node number;
the fifth step: operating the built i hidden-layer stacking noise reduction automatic encoders, and judging whether the transient stability evaluation effect is improved, namely whether the transient stability margin prediction precision is improved; if the improvement is good, i is i +1, returning to the step 2; if not, the i-1 hidden layers are the final network structure of the stacking noise reduction automatic encoder;
and a sixth step: and (3) using i-1 hidden layer outputs of the stacked noise reduction automatic encoder, namely i-1 feature expressions with different abstract degrees as the input of subsequent transient stability evaluation for subsequent security risk situation perception of the large power grid.
2. The method according to claim 1, wherein the transient stability assessment primitive feature set comprises: pre-failure system characteristics, individual characteristics within the failure point and its vicinity, and the location and duration of the anticipated accident.
3. The transient stability evaluation method based on deep learning of claim 1, wherein the transient stability margin indicator is a difference between a limit cut time and a fault cut time.
4. The transient stability evaluation method based on deep learning of claim 1, wherein the severity of the transient stability of the system in different operation modes is graded based on a utility theory, specifically:
wherein, T is a set threshold value, and M is a transient stability margin index;
when the transient stability margin index is larger than T, the system is considered to have no transient instability risk; when the transient stability margin index is less than 0, the system is considered to have transient instability; when the transient stability margin indicator belongs to the interval [0, T ], an exponential function is used as a severity function within the interval.
5. The method of claim 4, wherein the severity of the sample is divided into 5 levels, let SrRepresenting the severity rating, the severity rating rules are as follows:
wherein, the sample with the severity level of 3 is positioned near the boundary of the security domain, and the transient stability margin is lower; samples with a severity level of 4 included destabilized samples and critically stable samples.
6. The transient stability evaluation method based on deep learning of claim 4, further comprising: continuously acquiring online operation data from an energy management system, and updating the structure and parameters of a transient stability evaluation model, including updating the structure and parameters of a stacked noise reduction automatic encoder feature extraction model and updating the structure and parameters of a support vector machine integration model;
and when the on-line forecast accidents occur in a set of untrained forecast accidents, training a transient stability assessment model under the accidents on line.
7. A power system transient stability assessment system based on deep learning, comprising:
an original feature set construction module configured to: constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
a sample set generation module configured to: establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
a stacked denoising autoencoder feature extraction model adaptation building module configured to: self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
a transient stability assessment model building module configured to: taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and predicting transient stability margins of various operation modes under a fixed expected accident;
a future execution scenario generation module configured to: generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
a transient stability margin prediction and severity ranking module configured to: predicting transient stability margin by using a support vector machine integrated regression model corresponding to each expected accident, and grading the severity of the transient stability of the system in different operation modes based on a utility theory;
the method is characterized in that a stacking noise reduction automatic encoder feature extraction model is adaptively established for each expected accident based on a mutual information theory, and specifically comprises the following steps:
defining an initial stacking noise reduction automatic encoder network structure with a smaller scale, wherein the number of nodes of an input layer is equal to the input characteristic dimension; simultaneously setting a threshold value of an evaluation function J (f) based on the MIFS information measurement standard;
determining the number of nodes of the ith hidden layer; newly adding m nodes, carrying out weight initialization on the whole, and calculating J (f) of the weight vector of the existing node;
determining whether J (f) is greater than a threshold; if the threshold value is smaller than the threshold value, returning to the second step; if not, deleting the newly added node and carrying out the fourth step;
judging whether m is equal to 1; if m is not equal to 1, increasing the nodes by adopting a step-variable method,
returning to the second step for rounding down; if m is equal to 1, subtracting 1 from the current node number to obtain the hidden node number;
operating the built i hidden-layer stacking noise reduction automatic encoders, and judging whether the transient stability evaluation effect is improved, namely whether the transient stability margin prediction precision is improved; if the improvement is good, i is i +1, returning to the step 2; if not, the i-1 hidden layers are the final network structure of the stacking noise reduction automatic encoder;
and (3) using i-1 hidden layer outputs of the stacked noise reduction automatic encoder, namely i-1 feature expressions with different abstract degrees as the input of subsequent transient stability evaluation for subsequent security risk situation perception of the large power grid.
8. The deep learning-based power system transient stability assessment system of claim 7, further comprising: a model online update module configured to: and continuously acquiring online operation data from the energy management system, updating the structure and parameters of the transient stability evaluation model, and training the transient stability evaluation model under the accident on line when the untrained forecast accident occurs in the online forecast accident set.
9. The deep learning based power system transient stability assessment system according to claim 7, comprising a server, the server comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the program:
constructing a transient stability evaluation primitive feature set according to the operating characteristics of the power system;
establishing an offline expected accident set, and randomly generating a corresponding sample set for each expected accident;
self-adaptively establishing a stacking noise reduction automatic encoder feature extraction model for each expected accident based on a mutual information theory to realize the multi-layer feature extraction of the original input;
taking the characteristics of each layer extracted by the stacking noise reduction automatic encoder as the input of a learning machine of the support vector machine, establishing an integrated regression model of the support vector machine, and predicting transient stability margin indexes of various operation modes under a fixed expected accident;
generating a plurality of future operation modes and an online expected accident set by using the plan data and the prediction data based on the online operation mode;
and predicting transient stability margin indexes by using a support vector machine integrated regression model corresponding to each fixed expected accident, and grading the severity of the transient stability degree of the system in different operation modes based on a utility theory.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810759783.XA CN109086913B (en) | 2018-07-11 | 2018-07-11 | Power system transient stability assessment method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810759783.XA CN109086913B (en) | 2018-07-11 | 2018-07-11 | Power system transient stability assessment method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109086913A CN109086913A (en) | 2018-12-25 |
| CN109086913B true CN109086913B (en) | 2020-06-09 |
Family
ID=64837461
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810759783.XA Active CN109086913B (en) | 2018-07-11 | 2018-07-11 | Power system transient stability assessment method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109086913B (en) |
Families Citing this family (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109787222A (en) * | 2019-01-04 | 2019-05-21 | 三峡大学 | Distribution based on support vector machines is transported to electric system transient stability discrimination method |
| 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 |
| CN110298398B (en) * | 2019-06-25 | 2021-08-03 | 大连大学 | Wireless protocol frame characteristic selection method based on improved mutual information |
| CN110728457B (en) * | 2019-10-17 | 2022-11-22 | 广西电网有限责任公司电力科学研究院 | Operation risk situation perception method considering multi-level weak links of power distribution network |
| CN111079805A (en) * | 2019-12-03 | 2020-04-28 | 浙江工业大学 | Abnormal image detection method combining attention mechanism and information entropy minimization |
| CN111049159B (en) * | 2019-12-20 | 2023-09-29 | 四川大学 | Power system transient stability prevention control method embedded in deep belief network |
| CN111311008B (en) * | 2020-02-21 | 2021-04-30 | 山东大学 | AC/DC large power grid dynamic security risk situation rolling prospective early warning method and system |
| CN116562618A (en) * | 2020-05-19 | 2023-08-08 | 三峡大学 | Method for obtaining dynamic security assessment model |
| CN111628501B (en) * | 2020-06-18 | 2023-04-18 | 国网山东省电力公司济南供电公司 | AC/DC large power grid transient voltage stability assessment method and system |
| CN111753963B (en) * | 2020-06-28 | 2024-04-02 | 北京百度网讯科技有限公司 | Method, apparatus, device and storage medium for outputting information |
| CN111950765B (en) * | 2020-07-06 | 2024-04-19 | 四川大川云能科技有限公司 | Probabilistic transient stability prediction method based on stacked noise reduction self-encoder |
| CN112232427B (en) * | 2020-10-13 | 2023-10-03 | 宁波大学 | Wind driven generator fault detection method based on distributed regression model |
| CN112865118B (en) * | 2021-01-19 | 2022-09-02 | 河海大学 | Deep learning model generation method for power grid dynamic reactive power reserve demand calculation |
| CN114330884B (en) * | 2021-12-29 | 2024-04-16 | 山东大学 | Multi-time scale coordinated AC/DC power grid dynamic safety early warning method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107846012A (en) * | 2017-09-27 | 2018-03-27 | 国网湖北省电力公司 | A kind of transient stability evaluation in power system method based on depth learning technology |
-
2018
- 2018-07-11 CN CN201810759783.XA patent/CN109086913B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN109086913A (en) | 2018-12-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086913B (en) | Power system transient stability assessment method and system based on deep learning | |
| WO2021164118A1 (en) | Ac/dc bulk power grid dynamic security risk situation rolling prospective early warning method and system | |
| CN116505665B (en) | Fault monitoring method and system for power grid distribution line | |
| CN110009208B (en) | Method and device for evaluating health state of pole-mounted switch complete equipment based on hybrid intelligent algorithm | |
| CN110264116A (en) | A kind of Electrical Power System Dynamic safety evaluation method explored based on relationship with regression tree | |
| CN108053148A (en) | A kind of efficient diagnostic method of power information system failure | |
| CN113487447A (en) | Power distribution network load prediction and line loss cause analysis method based on big data | |
| CN109670611A (en) | A kind of power information system method for diagnosing faults and device | |
| CN110350510A (en) | A kind of power distribution network service restoration method considering failure disturbance degree | |
| CN110826228B (en) | Regional power grid operation quality limit evaluation method | |
| CN104299070B (en) | Operation of power networks trend real time security evaluation method | |
| CN107295537A (en) | A kind of method and system for wireless sensor network reliability of testing and assessing | |
| CN109378835A (en) | Based on the large-scale electrical power system Transient Stability Evaluation system that mutual information redundancy is optimal | |
| CN115587685A (en) | Tough power distribution network evaluation system containing high-permeability distributed power supply | |
| CN110649627B (en) | Static voltage stability margin evaluation method and system based on GBRT | |
| CN106845711A (en) | Processing method and processing device for power supply reliability data | |
| CN106548284B (en) | Operation regulation-oriented self-adaptive modular power grid safety evaluation method | |
| CN113887809A (en) | Power distribution network supply and demand balance method, system, medium and computing equipment under double-carbon target | |
| CN110071500B (en) | Method and system for decision-making of power distribution line recovery sequence after bus fault power loss | |
| CN110827134A (en) | Power grid enterprise financial health diagnosis method | |
| US20220391504A1 (en) | Leakage Measurement Error Compensation Method and System Based on Cloud-Edge Collaborative Computing | |
| CN111008779A (en) | Transformer health state assessment method based on multi-factor assessment model | |
| Venkatesan et al. | Artificial neural network based contingency ranking | |
| KR102551994B1 (en) | Earth leakage location checking and monitoring method, device and system using artificial intelligence model | |
| CN113055237B (en) | Distribution network main station cooperative self-healing reliability determination method and device and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |