CN117078035A - LSTM-based power system transient stability evaluation method, system and storage medium - Google Patents
LSTM-based power system transient stability evaluation method, system and storage medium Download PDFInfo
- Publication number
- CN117078035A CN117078035A CN202310916659.0A CN202310916659A CN117078035A CN 117078035 A CN117078035 A CN 117078035A CN 202310916659 A CN202310916659 A CN 202310916659A CN 117078035 A CN117078035 A CN 117078035A
- Authority
- CN
- China
- Prior art keywords
- power system
- transient stability
- training
- vector
- time sequence
- 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.)
- Pending
Links
- 230000001052 transient effect Effects 0.000 title claims abstract description 95
- 238000011156 evaluation Methods 0.000 title claims abstract description 64
- 238000003860 storage Methods 0.000 title claims abstract description 17
- 239000013598 vector Substances 0.000 claims abstract description 98
- 238000012549 training Methods 0.000 claims abstract description 67
- 238000000034 method Methods 0.000 claims abstract description 41
- 238000013210 evaluation model Methods 0.000 claims abstract description 27
- 238000004458 analytical method Methods 0.000 claims abstract description 24
- 238000005065 mining Methods 0.000 claims abstract description 16
- 230000006870 function Effects 0.000 claims description 22
- 238000013097 stability assessment Methods 0.000 claims description 13
- 230000004913 activation Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 7
- 230000002068 genetic effect Effects 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 6
- 238000004088 simulation Methods 0.000 claims description 5
- 238000005520 cutting process Methods 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 238000004590 computer program Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 8
- 230000007246 mechanism Effects 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000009826 distribution Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000003064 k means clustering Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Educational Administration (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Marketing (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Tourism & Hospitality (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Quality & Reliability (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Operations Research (AREA)
- Water Supply & Treatment (AREA)
- Public Health (AREA)
- Power Engineering (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
Abstract
The invention discloses an LSTM-based power system transient stability evaluation method, an LSTM-based power system transient stability evaluation system and a storage medium, belonging to the field of power system transient stability evaluation, and comprising the following steps: a training stage and an application stage; the training stage comprises the steps of training a transient stability evaluation model of the power system by adopting a training sample set; the evaluation model includes: the time sequence feature analysis module is used for extracting time sequence information vectors at corresponding moments in the training samples; the depth feature mining module extracts association features among different feature dimensions from the time sequence information vector at each moment by adopting a self-attention network to obtain a depth information vector corresponding to each moment; and the transient stability evaluation module is used for predicting whether the running state of the power system is in transient stability after the depth information vectors are spliced according to the time sequence. According to the method and the system, the stability of the power system can be evaluated more accurately according to the time sequence data in a short time after the fault occurs, and the evaluation precision and efficiency of the model are improved.
Description
Technical Field
The invention belongs to the field of power system transient stability evaluation, and particularly relates to an LSTM-based power system transient stability evaluation method, an LSTM-based power system transient stability evaluation system and a storage medium.
Background
The stability of the power system is critical to the safe operation of the power system. Transient stability is widely used, among other things, to evaluate the performance of a power system to maintain and reestablish after a disturbance. However, transient stability assessment remains challenging in terms of power system operation and control, particularly in increasingly popular large-scale power grids.
Conventional transient stability assessment mainly includes two types: one is to generalize the li-apunov theory, which is evaluated by energy equation modeling, which can be applied to simple grid models and power units. However, this approach is hindered by modeling difficulties and low accuracy of the simplified model. Therefore, it is difficult to use in large-scale power grids. The other is time domain simulation, which simulates a large number of samples with interference and obtains accurate evaluation results through detection system response. However, due to its high computational complexity, the analysis is time consuming and this method is not easily adaptable to large power systems. That is, none of the conventional transient stability assessment methods is applicable to large power systems.
With the popularization and application of Phasor Measurement Units (PMUs), it is possible to collect and store a large amount of grid operation data in real time. Therefore, data-driven power system transient stability assessment has become popular. Traditional machine learning can learn patterns and relationships between samples efficiently and automatically. However, decision Trees (DTs), support Vector Machines (SVMs), artificial neural networks, random Forests (RF), K-means clustering algorithms, etc., which cannot easily model nonlinear relationships and exhibit poor generalization ability in terms of high-dimensional data.
The model based on deep learning is the most potential method for evaluating the transient stability of the power system due to the excellent nonlinear fitting capability. For data-driven power system transient stability assessment, very short time response after failure is an important basis for neural network identification and stability. Long short term memory networks (LSTM) are the most widely used time series feature extraction networks, which perform well in small scale time feature extraction. However, the time information extracted by the LSTM is directly used for judging that the effect of a classification layer of the transient stability of the power system is poor, so that the evaluation accuracy of the model is lower. In addition, the number difference between stable samples and unstable samples in the power system is extremely large, so that the stable samples with large numbers are stressed in actual evaluation, and the more essential characteristics of model learning and the robustness of the model are influenced.
Disclosure of Invention
Aiming at the defects and improvement demands of the prior art, the invention provides an LSTM-based power system transient stability evaluation method, an LSTM-based power system transient stability evaluation system and a storage medium, and aims to improve the evaluation accuracy of a model.
To achieve the above object, according to a first aspect of the present invention, there is provided an LSTM-based power system transient stability evaluation method, including:
training phase: training the transient stability evaluation model of the power system by adopting a training sample set; the training samples are power angle cluster sequences of the historical running state of the power system after fault removal and real transient stability labels of the power system; the evaluation model includes:
the time sequence feature analysis module adopts T LSTM units which are connected in series and have shared parameters, and each LSTM unit is used for extracting time sequence information vectors at corresponding moments in the training samples; t is the number of sampling time points in the training sample;
the depth feature mining module is used for extracting association features among different feature dimensions from the time sequence information vector at each moment by adopting a self-attention network to obtain a depth information vector corresponding to each moment;
the transient stability evaluation module is used for predicting whether the running state of the power system is in transient stability after the depth information vectors are spliced according to the time sequence;
the application stage comprises the following steps: and inputting the power angle cluster sequence of the operation state of the power system after the current collected faults are removed into a trained evaluation model to obtain a transient stability judgment result of the operation of the power system.
Further, the loss function of the training phase is:
wherein, alpha represents an adjusting factor, gamma represents a concern factor, p represents a prediction result of the transient stability evaluation module, and when the prediction result is transient stability, p=1, otherwise, p=0; m is the number of training samples.
Further, the value of the adjustment factor alpha is the proportion of stable and unstable samples in the training samples.
Further, the transient stability evaluation module comprises a splicing unit and two fully-connected layers which are connected in series;
the splicing unit is used for splicing the depth information vectors according to time sequences to obtain spliced depth information vectors; and the spliced depth information vector sequentially passes through the two serially connected full-connection layers to obtain a predicted transient stability judgment result.
Further, the self-attention network comprises: a generation layer, a matching layer, an output layer and a normalization layer;
the generation layer is used for obtaining an index vector, a keyword vector and a value vector of the corresponding moment from the time sequence information vector of each moment;
the matching layer is used for carrying out softmax normalization on the product of the index vector and the keyword vector to obtain a feature matrix;
the output layer is used for multiplying the characteristic matrix and the value vector and obtaining a depth information vector at a corresponding moment after passing through the normalization layer.
Further, the activation functions of the output layer of the self-attention network and the two fully connected layers in series are ReLU activation functions.
Further, the training sample acquisition method comprises the following steps:
s1, sampling a power angle value from a power system through a phasor measurement device, obtaining a power angle value data sequence and marking;
or adding faults to the power system, cutting off the faults after a certain time, performing time domain simulation, obtaining a power angle value data sequence and marking;
s2, carrying out Max-Min standardization on the marked data to be in [0,1 ];
and S3, extracting the standardized data based on a power angle cluster extraction technology to obtain an F-dimension power system running state power angle cluster sequence serving as the training sample.
Further, before the training samples are input into the evaluation model, feature screening is performed on the training samples by adopting a wrapper method based on genetic search.
According to a second aspect of the present invention, there is provided an LSTM based power system transient stability assessment system comprising a computer readable storage medium and a processor;
the computer-readable storage medium is for storing executable instructions;
the processor is configured to read executable instructions stored in the computer readable storage medium to perform the method of any one of the first aspects.
According to a third aspect of the present invention there is provided a computer readable storage medium having stored thereon a computer program which when executed by a processor implements a method according to any of the first aspects.
In general, through the above technical solutions conceived by the present invention, the following beneficial effects can be obtained:
(1) According to the evaluation method, in addition to the time sequence characteristics, the data of the power system are considered, the high characteristic correlation among the data of different characteristic dimensions is considered, the time sequence characteristic correlation information among the different time points of the data of the power system is extracted by using the LSTM shared by parameters, the correlation information among the different latitude characteristics in the time sequence information at each time point is extracted by using a self-attention mechanism, the deep analysis of the characteristic information is realized, the output depth information vector simultaneously contains the time sequence correlation information among the different time points and the characteristic correlation information among the different dimension at the same time point, the transient stability of the running state of the power system is evaluated based on the depth information vector, the analysis capability of the time sequence data is improved, the misjudgment and the misjudgment probability are reduced, and the evaluation precision of the model is improved.
(2) Furthermore, the loss function designed according to the distribution characteristics of the training samples has the function of balancing the sample size, the model weight distribution can be changed during the training of the model through the adjustment factor alpha and the attention factor gamma, so that the transient stability evaluation model of the power system is ensured to pay more attention to the unstable samples with less sample size but important function in the training set, the unstable samples are more accurately identified by the stability evaluation model, the risk of missed judgment is further reduced, and the method has practical significance.
(3) Preferably, the proportion of stable and unstable samples in the data set is used as the value of the regulating factor alpha, and experiments prove that the attention of the transient stability evaluation model of the power system to the unstable samples in the training set can be further improved.
(4) Preferably, the stability evaluation module adopts two fully connected layers connected in series, integrates the information of the spliced depth information vector, and evaluates the integrated information, so that the utilization of the information can be increased, and the evaluation accuracy is further improved.
(5) Preferably, the ReLU activation function is selected as the activation function of the depth feature mining module output layer and the two fully-connected layers which are connected in series, so that gradient disappearance and gradient explosion can be reduced, and stable operation of the power transient stability assessment model in the training process is realized.
(6) According to the training sample, the original power angle time sequence data are converted into the power angle cluster sequence with fixed characteristic dimensions for analysis, so that the problem of difficult transient stability judgment caused by different input dimensions of power systems with different topological structures is solved, the generalization performance of the model is improved, and meanwhile, the calculation speed of the model is also improved.
(7) Preferably, feature screening is performed on the training samples by adopting a wrapper method based on genetic search, and more representative features are selected as inputs of the model, so that the robustness of the model for the power system with different complex topological structures and the efficiency of transient stability evaluation are enhanced.
In summary, the evaluation method solves the technical problems of lower evaluation precision, lower evaluation efficiency and higher unstable sample missing judgment rate of the existing power transient stability evaluation method, and improves the evaluation precision of the model; the invention can more accurately evaluate the stability of the power system according to the time sequence data in a short time after the fault occurs.
Drawings
FIG. 1 is a schematic diagram of an LSTM-SAF model in an embodiment of the invention.
Fig. 2 is a flowchart of an LSTM-based power system transient stability evaluation method according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of a timing characteristic analysis module according to an embodiment of the invention.
Fig. 4 is a schematic structural diagram of a depth feature mining module according to an embodiment of the present invention.
FIG. 5 is a schematic diagram of a process for a wrapper selection feature based on a genetic algorithm in an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention. In addition, the technical features of the embodiments of the present invention described below may be combined with each other as long as they do not collide with each other.
As shown in fig. 1 and 2, the LSTM-based power system transient stability assessment method of the present invention includes: a training stage and an application stage;
the training phase comprises the following steps: training the transient stability evaluation model of the power system by adopting a training set; the training samples in the training set are power angle cluster sequences of the historical running state of the power system after fault removal and real transient stability labels of the power system. Specifically, the power angle cluster sequence of the historical operating state of the power system after fault removal comprises motor power angle data X epsilon R of T sampling time points after fault removal T*F ，X＝{X 1 ，X 2 ，...，X T F represents the characteristic dimension of the sample data, F in fig. 1 i Represents the i-th data in any data sample, i= {1, 2..the, F }; in the embodiment of the present invention, f=27, that is, the data sample is a 27-dimensional power angle cluster sequence X.
Wherein, the power system transient stability evaluation model includes: the system comprises a time sequence feature analysis module, a depth feature mining module and a transient stability evaluation module;
the time sequence feature analysis module is used for extracting time sequence information vectors h, h= { h at each moment in the training sample 1 ，h 2 ，...，h T -a }; specifically, the time sequence feature analysis module comprises a plurality of LSTM units which are connected in series and have shared parameters, and each LSTM unit is used for extracting a time sequence information vector at a corresponding moment in a training sample; currently, the method is thatTime sequence information vector h of moment t For analysis of the timing information at the next time, T is 1,2, …, or T.
The depth feature mining module is used for performing depth mining on the time sequence information vectors from the feature dimension at all moments output by the time sequence feature analysis module, extracting deep information from the features of the time sequence information vectors, constructing a set of new depth features, and outputting depth information vectors H, H= { H corresponding to each moment 1 ，H 2 ，...，H T }. Wherein the new data feature dimension is F'. Specifically, the depth feature mining module extracts association features among different feature dimensions of the time sequence feature vector at each moment output by the time sequence information analysis module from the feature dimensions by adopting a self-attention network, and obtains a depth information vector corresponding to each moment.
The stability evaluation module is used for evaluating the running state of the power system after splicing the depth information vectors which are output by the depth feature mining module and reflect the feature association relations between different feature dimensions at each moment according to the time dimension, and the evaluation result is whether the power system is in transient stability.
Specifically, in the timing characteristic analysis module, as shown in fig. 3, in the embodiment of the present invention, T LSTM units connected in series and having shared parameters are adopted; in other embodiments, the number of LSTM cells is determined based on the length of the input power system operation state vector, and each sampling time corresponds to one LSTM cell, and is used to extract the timing information vector and the timing state vector at the corresponding time. Meanwhile, to establish long-term dependencies, LSTM maintains a sequential state vector c throughout its lifecycle.
Each LSTM cell includes three inputs and two outputs, the three inputs being: training a power system running state vector at a corresponding moment in a sample, a time sequence information vector and a time sequence state vector which are output by an LSTM unit corresponding to the last moment; that is, the input of the LSTM unit corresponding to the current moment is the running state vector x of the power system at the current moment in the training sample t The time sequence information vector h of the last moment output by the LSTM unit corresponding to the last moment t-1 Time sequence state vector c of last moment t-1 The method comprises the steps of carrying out a first treatment on the surface of the The output of the LSTM unit corresponding to the current moment is the time sequence information vector h of the current moment t Time sequence state vector c at current moment t The method comprises the steps of carrying out a first treatment on the surface of the Wherein, the input of the first moment is only the running state vector x of the electric power system at the first moment 1 Time sequence information vector h of last moment 0 Time sequence state vector c of=0 and last moment 0 ＝0。
Specifically, each LSTM cell includes three gating structures: an input gate, a forget gate, and an output gate. Aiming at the LSTM unit corresponding to the current moment, the running state vector x of the power system at the current moment in the sample is trained t Time sequence information vector h of last moment output by LSTM unit corresponding to last moment t-1 A new vector i is calculated in the input gate, and the new information which needs to be reserved is represented; at the same time, state vector x t And the timing information vector h at the previous time t-1 A new vector f is calculated in the forgetting door, and the new vector f represents old information to be reserved; calculating the vector i and the vector f according to the following calculation formula to obtain a time sequence state vector c at the current moment t The method comprises the steps of carrying out a first treatment on the surface of the State vector x t And the timing information vector h at the previous time t-1 A new vector o is calculated in the input and output gate, and the new vector o represents information to be output; the time sequence state vector c of the current moment is obtained t The following formula calculation is carried out with the new vector o to obtain a time sequence information vector h at the current moment t ；
c t ＝f⊙c t +i⊙tanh(Wcx t +Uch t-1 +bc)
h t ＝o⊙tanh(c t )
Where Wc, uc and bc are learnable parameters, and by which it is indicated that element-wise multiplication, tan h is a hyperbolic tangent function.
The various gating mechanisms of the LSTM cells may enable efficient timing analysis of the power system operational data sequences. Therefore, the invention evaluates the future transient stability condition of the system based on short-time operation data.
Specifically, as shown in fig. 4, the depth feature mining module in the embodiment of the present invention includes a self-attention network, which is configured to perform depth feature analysis and extraction on the time sequence information vectors output by the time sequence feature analysis module at all times, and includes: a generation layer, a matching layer, an output layer and a normalization layer.
The generation layer is used for obtaining an index vector, a keyword vector and a value vector of the corresponding moment according to the time sequence information vector of each moment; the matching layer obtains a feature matrix by softmax normalization of the product of the keyword vector and the index vector; the output layer is used for multiplying the feature matrix and the value vector to obtain a constructed new feature dimension feature vector, and outputting the features of the time sequence feature vector at each moment through the normalization layer. In actual modeling, the feature dimensions of the output are adjusted as hyper-parameters.
Specifically, the stability evaluation module comprises a splicing unit and two fully-connected layers connected in series. The depth feature mining module outputs a depth feature vector of each time step, and the depth feature vector is spliced according to time sequence through the splicing unit to obtain a spliced depth information vector H, and the spliced depth information vector H is input to a first full-connection layer for information synthesis and then is output as a vector with a set length; and inputting the vector with the set length into a second full-connection layer to obtain a transient stability evaluation result. Through two full-connection layers, the information of the depth information vector H is integrated, and then the integrated information is evaluated, so that the utilization of the information can be increased, and the evaluation accuracy is further improved. In the embodiment of the invention, the set length is 10, and the transient stability evaluation result is represented by a two-dimensional vector.
Preferably, the ReLU activation function is selected as an output layer activation function of each neural network in the power transient stability evaluation model. Specifically, an output layer of the self-attention network of the depth feature mining module is connected with the stability evaluation module by adopting a ReLU activation function; the two fully-connected layers of the stability evaluation module are connected through a ReLU activation function, so that gradient disappearance and gradient explosion can be reduced, and stable operation of the power transient stability evaluation module in the training process is realized.
In the training process, the loss function of the power system transient stability evaluation model is the loss between the power system transient stability evaluation result predicted by the stability evaluation module and the true power system stability label. The loss between the predicted result and the label is minimized, so that the transient stability evaluation misjudgment rate of the offline power system is minimized.
Specifically, a loss value is calculated by a loss function L; in the embodiment of the invention, the designed loss function can change the model weight distribution when the model is trained by adjusting the factor alpha and the attention factor gamma, so that the transient stability evaluation model of the power system is ensured to pay more attention to unstable samples in the training set. The loss function L designed in the embodiment of the invention is as follows:
wherein y is a tag in a real stable or unstable state in the running state of the power system, y=0, and represents that the running state of the power system is transient unstable, and y=1, and represents that the running state of the power system is transient stable; p represents the power system transient stability evaluation result predicted by the stability evaluation module, wherein when the predicted evaluation result is transient stability, p=1, otherwise, p=0; m is the number of training samples in the training set.
The adjustment factor alpha and the attention factor gamma determine specific values from the data set. Preferably, the proportion of stable and unstable samples in the data set is used as the value of the adjusting factor alpha, the value of the attention factor gamma is selected according to experience, and at the moment, the attention of the transient stability evaluation model of the power system to the unstable samples in the training set can be improved.
It should be noted that, in the training process, the training set may be divided into different batches to train the power system transient stability evaluation model, and the loss value of the power system transient stability evaluation model is minimized under each batch. Parameters of the transient stability evaluation model of the power system are learned through a back propagation mechanism. In an alternative embodiment, the parameters of the model are optimized during training using a random gradient descent method.
Preferably, the evaluation method of the present invention further comprises pre-screening the characteristics of the samples in the dataset. In an embodiment of the present invention, a wrapper method based on genetic search is used for sample feature screening, as shown in fig. 5. Specifically, the training sample is subjected to feature screening on a pre-training model by using a wrapper feature selection method, and the screened features are subjected to optimal feature searching through a genetic algorithm to obtain the screened sample. And rolling segmentation is carried out on the power system operation time sequence characteristic sequence based on the sliding window, and a short-time sequence serving as a model input is obtained in real time.
The application stage comprises the following steps: and inputting the currently acquired power angle cluster sequence of the power system running state after fault removal into the trained power system transient stability evaluation model to obtain a power system running transient stability judgment result.
Specifically, the method for acquiring the power angle cluster sequence of the power system running state after fault removal comprises the following steps:
s1, acquiring key data of a real power grid during operation through a Phasor Measurement Unit (PMU) to calculate a power angle data value, performing discrete sampling to generate available time sequence data, and then marking by combining with a stability criterion;
or adding faults to the power system, cutting off the faults after a certain time, performing time domain simulation to obtain a power angle value data sequence, and marking by combining with a stability criterion.
S2, carrying out Max-Min standardization on the marked data into [0,1], and dividing the marked data into training set data and test set data;
s3, extracting standardized power running state data based on a power angle cluster extraction technology to obtain an F-dimension power system running state power angle cluster sequence as a training sample, wherein F=27 in the embodiment of the invention. The original power angle time sequence data are converted into the power angle cluster sequence with fixed characteristic dimensions for analysis, so that the problem of difficult transient stability judgment caused by different input dimensions of power systems with different topological structures is solved, the generalization performance of the model is improved, and meanwhile, the calculation speed of the model is also improved.
The stability label of the sample data is obtained by cutting off all power angle data of the power system in a longer time through faults and combining with a stability criterion. The adopted power system stability criteria are as follows:
wherein delta max And when the TSI is larger than 0, the system transient stability is indicated. TSI is less than 0, indicating system transient instability.
Training the standardized training data by using the constructed power transient stability evaluation model, inputting the standardized test set data into the trained power transient stability evaluation model for stability evaluation, and finally obtaining a final evaluation result.
In the embodiment of the invention, the power transient stability evaluation model is marked as an LSTM-SAF model. The training set data are obtained from the public IEEE39 node system, the NPCC140 node system and the IEEE145 node system through batch simulation of PST3.0 toolkits in MATLAB, and are compared with a plurality of transient stability evaluation reference models aiming at power systems with three different topological structures, so that the performance of the model constructed by the invention is illustrated.
And constructing a model by using a PyTorch library, wherein the hyper-parameters in the model are determined by using a grid search strategy. The batch_size of the training set is set to 100 and the batch_size of the test set is set to 1. The learning rate was set to 0.001, and the effect of model evaluation was measured using the accuracy rate (ACC), the misjudgment rate (MIS), the false judgment rate (FAL), the G-mean, and the F1 score (F1-score).
Table 1 shows the comparison values of the evaluation indexes obtained by the power transient stability evaluation model (LSTM-SAF) and the reference model on the power systems with three different topologies.
TABLE 1 comparison of LSTM-SAF with evaluation index of reference model
It can be seen from table 1 that the LSTM-SAF model of the present invention achieves optimal performance compared to the reference model for power systems of different topologies. For example, in an IEEE39 node system, the proposed model achieves a significant improvement in accuracy over the reference model, although the proposed model does not differ much from the simple 1D-CNN result in MIS, the proposed model still achieves good results considering overall performance, and the proposed model achieves very low results on the false positive rate of unstable samples of interest, mainly due to the model weight distribution that the loss function of the inventive design changes when training the model. Meanwhile, in the power system results of other topological structures, the model of the invention also obtains excellent performance, and the more complicated power system with a structure is considered to have stronger anti-interference capability against external interference, so that the LSTM-SAF model of the invention also obtains higher results.
In summary, the invention integrates the information depth extraction concept into transient stability judgment, and in the process of gradually extracting the information in the time sequence data, the information which is significant for transient stability evaluation in the time sequence is mined and utilized, so that the accuracy of transient stability evaluation is improved. In the invention, 27-dimensional characteristics extracted from the power angle cluster curve in the operation process of the power system are used as the input of the transient stability evaluation model, and more representative characteristics are further selected by using a wrapper method, so that a characteristic engineering thought combining characteristic extraction and characteristic selection can be realized, more efficient and more robust model input can be realized, the evaluation robustness and accuracy can be improved, and the method is suitable for real-time effective actual operation transient stability evaluation of the power system. In addition, the loss function designed by the invention gradually guides the model in the training process to pay more attention to unstable samples which are more harmful in practical application, and has great significance in reducing economic loss. Therefore, the method and the device are applicable to transient stability evaluation of power systems with different topological structures, and have strong practical applicability.
According to the evaluation method, in addition to the time sequence characteristics, the data of the power system are considered, the high characteristic correlation among the data of different characteristic dimensions is considered, the time sequence characteristic correlation information among the different time moments of the data of the power system is extracted by using the long-short-period memory neural network LSTM shared by parameters, the correlation information among the different latitude characteristics in the time sequence information at each time moment is extracted by using a self-attention mechanism, the deep analysis of the characteristic information is realized, the output depth information vector simultaneously contains the time sequence correlation information among the different time moments and the characteristic correlation information among the different dimensions at the same time moment, the analysis capability of the time sequence data is improved, the misjudgment is reduced, the misjudgment probability is increased, and the more accurate and more robust evaluation result is obtained.
The invention combines the advantages of LSTM extraction time sequence information and the advantages of self-attention mechanism, namely the self-attention mechanism has good characteristic relevance analysis capability, attention calculation is carried out on a characteristic channel, and the transient stability analysis can be carried out by comprehensively utilizing the relation among the characteristics. The depth feature mining module enables information which cannot be learned by the time sequence information analysis module to be submitted to the downstream depth feature mining module for learning, and the depth feature mining module learns the feature association relation between time sequence information vectors at each moment, so that time sequence information in the power system running state sequence can be more fully mined out, and accurate assessment of the transient stability of the power system running state is achieved.
According to a second aspect of the present invention, there is provided an LSTM based power system transient stability assessment system comprising a computer readable storage medium and a processor;
the computer-readable storage medium is for storing executable instructions;
the processor is configured to read executable instructions stored in the computer readable storage medium to perform each step corresponding to the LSTM-based power system transient stability evaluation method in the above embodiment.
According to a third aspect of the present invention, there is provided a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the LSTM-based power system transient stability assessment method in the above-described embodiment.
It will be readily appreciated by those skilled in the art that the foregoing description is merely a preferred embodiment of the invention and is not intended to limit the invention, but any modifications, equivalents, improvements or alternatives falling within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (10)
1. An LSTM-based power system transient stability assessment method, comprising:
training phase: training the transient stability evaluation model of the power system by adopting a training sample set; the training samples are power angle cluster sequences of the historical running state of the power system after fault removal and real transient stability labels of the power system; the evaluation model includes:
the time sequence feature analysis module adopts T LSTM units which are connected in series and have shared parameters, and each LSTM unit is used for extracting time sequence information vectors at corresponding moments in the training samples; t is the number of sampling time points in the training sample;
the depth feature mining module is used for extracting association features among different feature dimensions from the time sequence information vector at each moment by adopting a self-attention network to obtain a depth information vector corresponding to each moment;
the transient stability evaluation module is used for predicting whether the running state of the power system is in transient stability after the depth information vectors are spliced according to the time sequence;
the application stage comprises the following steps: and inputting the power angle cluster sequence of the operation state of the power system after the current collected faults are removed into a trained evaluation model to obtain a transient stability judgment result of the operation of the power system.
2. The method of claim 1, wherein the training phase has a loss function of:
wherein, alpha represents an adjusting factor, gamma represents a concern factor, p represents a prediction result of the transient stability evaluation module, and when the prediction result is transient stability, p=1, otherwise, p=0; m is the number of training samples.
3. The method of claim 2, wherein the adjustment factor α is a ratio of stable and unstable samples in the training sample.
4. The method of claim 1, wherein the transient stability assessment module comprises a splice unit and two fully connected layers in series;
the splicing unit is used for splicing the depth information vectors according to time sequences to obtain spliced depth information vectors; and the spliced depth information vector sequentially passes through the two serially connected full-connection layers to obtain a predicted transient stability judgment result.
5. The method of claim 4, wherein the self-attention network comprises: a generation layer, a matching layer, an output layer and a normalization layer;
the generation layer is used for obtaining an index vector, a keyword vector and a value vector of the corresponding moment from the time sequence information vector of each moment;
the matching layer is used for carrying out softmax normalization on the product of the index vector and the keyword vector to obtain a feature matrix;
the output layer is used for multiplying the characteristic matrix and the value vector and obtaining a depth information vector at a corresponding moment after passing through the normalization layer.
6. The method of claim 5, wherein the activation functions of the output layer of the self-attention network and the two serially connected fully connected layers are ReLU activation functions.
7. The method according to claim 1, wherein the training sample acquisition method comprises:
s1, sampling a power angle value from a power system through a phasor measurement device, obtaining a power angle value data sequence and marking;
or adding faults to the power system, cutting off the faults after a certain time, performing time domain simulation, obtaining a power angle value data sequence and marking;
s2, carrying out Max-Min standardization on the marked data to be in [0,1 ];
and S3, extracting the standardized data based on a power angle cluster extraction technology to obtain an F-dimension power system running state power angle cluster sequence serving as the training sample.
8. The method of claim 7, further comprising feature screening the training samples using a genetic search-based wrapper method prior to inputting the training samples to the assessment model.
9. An LSTM based power system transient stability assessment system comprising a computer readable storage medium and a processor;
the computer-readable storage medium is for storing executable instructions;
the processor is configured to read executable instructions stored in the computer readable storage medium to perform the method of any one of claims 1-8.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310916659.0A CN117078035A (en) | 2023-07-25 | 2023-07-25 | LSTM-based power system transient stability evaluation method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310916659.0A CN117078035A (en) | 2023-07-25 | 2023-07-25 | LSTM-based power system transient stability evaluation method, system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117078035A true CN117078035A (en) | 2023-11-17 |
Family
ID=88718499
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310916659.0A Pending CN117078035A (en) | 2023-07-25 | 2023-07-25 | LSTM-based power system transient stability evaluation method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117078035A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117937521A (en) * | 2024-03-25 | 2024-04-26 | 山东大学 | Power system transient frequency stability prediction method, system, medium and equipment |
| CN118520304A (en) * | 2024-07-19 | 2024-08-20 | 国网山西省电力公司营销服务中心 | Deep learning multilayer active power distribution network situation awareness and assessment method and system |
-
2023
- 2023-07-25 CN CN202310916659.0A patent/CN117078035A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117937521A (en) * | 2024-03-25 | 2024-04-26 | 山东大学 | Power system transient frequency stability prediction method, system, medium and equipment |
| CN118520304A (en) * | 2024-07-19 | 2024-08-20 | 国网山西省电力公司营销服务中心 | Deep learning multilayer active power distribution network situation awareness and assessment method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhang et al. | Deep learning for short-term voltage stability assessment of power systems | |
| Wang et al. | Data-driven mode identification and unsupervised fault detection for nonlinear multimode processes | |
| CN110718910B (en) | Transient stability evaluation method for Bayesian optimization LightGBM | |
| CN117078035A (en) | LSTM-based power system transient stability evaluation method, system and storage medium | |
| CN108022001A (en) | Short term probability density Forecasting Methodology based on PCA and quantile estimate forest | |
| CN110807084A (en) | Attention mechanism-based patent term relationship extraction method for Bi-LSTM and keyword strategy | |
| Li et al. | Deep learning-based transient stability assessment framework for large-scale modern power system | |
| Lin et al. | Electricity theft detection in power consumption data based on adaptive tuning recurrent neural network | |
| Zu et al. | A simple gated recurrent network for detection of power quality disturbances | |
| Zhan et al. | A Two-Stage transient stability prediction method using convolutional residual memory network and gated recurrent unit | |
| Santos et al. | Performance analysis of a DNN classifier for power system events using an interpretability method | |
| Zhou et al. | Real-time anomaly detection in distribution grids using long short term memory network | |
| Cao et al. | Short-term forecasting and uncertainty analysis of photovoltaic power based on the FCM-WOA-BILSTM model | |
| Li et al. | Open set recognition fault diagnosis framework based on convolutional prototype learning network for nuclear power plants | |
| Guo et al. | A stacked ensemble method based on TCN and convolutional bi-directional GRU with multiple time windows for remaining useful life estimation | |
| Yao et al. | Electricity theft detection in incremental scenario: A novel semi-supervised approach based on hybrid replay strategy | |
| CN117726478A (en) | Intelligent decision-making method for dispatching of power system unit, terminal equipment and storage medium | |
| Xu et al. | Short-term wind power prediction based on anomalous data cleaning and optimized lstm network | |
| CN115510961A (en) | Community comprehensive energy system operation safety assessment method based on active learning | |
| Senesoulin et al. | Estimation of dominant power oscillation modes based on ConvLSTM approach using synchrophasor data and cross-validation technique | |
| Qin | Software reliability prediction model based on PSO and SVM | |
| Ma et al. | A prediction-based cycle life test optimization method for cross-formula batteries using instance transfer and variable-length-input deep learning model | |
| Chen et al. | Transient voltage stability assessment based on an improved TCN-BiLSTM framework | |
| Liu et al. | Line loss prediction method of distribution network based on long short-term memory | |
| CN117973639B (en) | Photovoltaic power prediction method based on multiscale similarity days and improved integration algorithm |
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 |