CN112051481B - Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM - Google Patents

Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM Download PDF

Info

Publication number
CN112051481B
CN112051481B CN202010810874.9A CN202010810874A CN112051481B CN 112051481 B CN112051481 B CN 112051481B CN 202010810874 A CN202010810874 A CN 202010810874A CN 112051481 B CN112051481 B CN 112051481B
Authority
CN
China
Prior art keywords
fault
judgment
moment
model
judgment model
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
Application number
CN202010810874.9A
Other languages
Chinese (zh)
Other versions
CN112051481A (en
Inventor
姚伟
李舟平
曾令康
艾小猛
马士聪
赵兵
文劲宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
State Grid Corp of China SGCC
China Electric Power Research Institute Co Ltd CEPRI
State Grid Hubei Electric Power Co Ltd
Original Assignee
Huazhong University of Science and Technology
State Grid Corp of China SGCC
China Electric Power Research Institute Co Ltd CEPRI
State Grid Hubei Electric Power Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology, State Grid Corp of China SGCC, China Electric Power Research Institute Co Ltd CEPRI, State Grid Hubei Electric Power Co Ltd filed Critical Huazhong University of Science and Technology
Priority to CN202010810874.9A priority Critical patent/CN112051481B/en
Publication of CN112051481A publication Critical patent/CN112051481A/en
Application granted granted Critical
Publication of CN112051481B publication Critical patent/CN112051481B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
    • G01R31/086—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution networks, i.e. with interconnected conductors
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
    • G01R31/08—Locating faults in cables, transmission lines, or networks
    • G01R31/088—Aspects of digital computing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • 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
    • G06N3/045—Combinations of 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
    • Y04S10/52—Outage or fault management, e.g. fault detection or location

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention belongs to the technical field of power transmission safety of power systems, and discloses an LSTM-based method and system for diagnosing a fault area of an alternating-current and direct-current hybrid power grid. The method comprises the following steps: constructing a power system time domain simulation model to simulate various working conditions to obtain a full-network node voltage amplitude and a corresponding label to obtain a basic sample set; intercepting a basic sample by a sliding window method to obtain a sample set of a training fault moment judgment model and a fault area judgment model for training and properly adjusting relevant parameters of the models; inputting a sample of the current sliding window moment, judging whether a fault occurs by a fault moment judgment model, and inputting the sample into a fault area judgment model if the fault occurs, so as to obtain a fault position; and finally, outputting the current fault diagnosis information. The invention can quickly and effectively determine the fault occurrence time in the power grid and determine the fault area by utilizing the deep neural network, has better performance under different operating conditions and certain noise conditions, and can meet the requirement of fault diagnosis in a complex power system.

Description

Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM
Technical Field
The invention belongs to the technical field of power transmission safety of power systems, and particularly relates to an LSTM-based alternating current-direct current hybrid power grid fault area diagnosis method and system.
Background
The rapid and accurate fault diagnosis of the power system after the fault occurs is of great significance to clearing the fault, recovering the power supply and maintaining the stability of the system. With the large-scale access of a direct current system, the alternating current and direct current coupling of a power grid in China is increasingly tight, and the failure of an alternating current system easily causes the commutation failure and even the locking of the direct current system, so that the stability of the power system is influenced. The traditional diagnosis method based on the relay protection of the alternating current system is difficult to be directly applied to an alternating current-direct current hybrid power grid for effective fault diagnosis. Due to the nonlinearity of a direct current system in the alternating current-direct current hybrid power grid and the complexity of direct current control, the fault transient process is more complex compared with a pure alternating current system, so that the complex fault in the operation of the alternating current-direct current hybrid power grid is more difficult to distinguish compared with the fault of the traditional power grid. In addition, because the frequency of the power electronic switch in the direct current system is very high, the traditional fault diagnosis, isolation and control measures cannot meet the actual requirements in terms of action speed.
At present, many intelligent diagnosis schemes based on expert systems, Bayesian networks, Petri networks, fuzzy logic theory and other methods have been proposed, but the methods still have some problems respectively. For example, expert systems are complex to build, update, and maintain; assignment to nodes in a Bayesian network can be obtained only by a large amount of observation and statistical analysis; the modeling of the Petri network is easy to cause the problem of state combination explosion due to the increase of the number of network nodes; in the fuzzy logic theory, proper membership function needs to be set to obtain better performance. However, with the rapid development of artificial intelligence technology, a new solution is provided for deep learning methods represented by CNN and LSTM (Long Short-Term Memory) in a deep neural network, so that the related problems of diagnosis of fault areas of the alternating current-direct current hybrid power grid are solved.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide an LSTM-based AC/DC hybrid power grid fault area diagnosis method and system, and aims to solve the technical problems that the existing power system fault diagnosis method depends on expert knowledge and experience, needs manual screening of effective characteristics, is difficult in model parameter setting, cannot be effectively applied to an AC/DC hybrid power grid, and accordingly influences the accuracy of fault diagnosis.
In order to achieve the above object, an aspect of the present invention provides an LSTM-based ac/dc hybrid power grid fault area diagnosis method, including:
(1) constructing an alternating current-direct current hybrid power grid time domain simulation model, simulating and obtaining voltage amplitude data of nodes of the whole grid under various operating conditions, and labeling labels corresponding to fault conditions to form a basic sample set;
(2) respectively constructing a fault moment judgment model and a fault area judgment model by using an LSTM neural network, and intercepting the basic sample set by adopting a sliding window mode to form sample sets, namely a training set and a testing set, which are respectively used for training and testing the fault moment judgment model and the fault area judgment model;
(3) training the fault moment judgment model and the fault area judgment model by adopting the training set, then adjusting neural network hyper-parameters and model auxiliary parameters in the two judgment models according to the performances of the fault moment judgment model and the fault area judgment model on the test set, and repeating the training until the two judgment models meet the performance requirements on the test set;
(4) when the online application is put into use, the fault moment judgment model determines whether a fault occurs after accumulative judgment according to the fault moment judgment model, if so, the fault occurrence position is determined according to the fault area judgment model, and if not, the current fault diagnosis information is directly generated.
Further, the fault moment judgment model is an LSTM neural network model; the fault moment judging model receives the sampling voltage amplitude data X intercepted by the sliding window, the sampling voltage amplitude data X is used as the input of a neural network after being standardized by z-score, then a judgment result of whether a fault occurs under the current input is output, and if the fault occurs, the judged fault occurrence moment t is outputf′。
Further, the fault area judgment model is an LSTM neural network model; judging at the time of the fault that the model has confirmed the fault and giving a judged fault occurrence time tf' after, delayed by a number of times Δ tlUntil tf′+ΔtlAnd voltage amplitude data X intercepted by adopting a sliding window at each moment is input into the fault area judgment model after being standardized by z-score.
Further, the step (4) includes:
(401) at each sampling moment, inputting voltage amplitude data at the current moment intercepted in a sliding window mode into a fault moment judgment model, and prejudging whether the current moment has a fault or not; determining whether a fault occurs after the pre-judgment result needs to be subjected to accumulative judgment, if so, resetting the accumulative judgment frequency of the fault moment judgment model, executing a step 402, and if not, executing a step 403;
(402) after the fault is confirmed to occur, inputting voltage amplitude data after delaying a plurality of sampling moments into a fault area judgment model to diagnose the position of a fault area, and determining nodes near the position where the fault occurs;
(403) and generating current fault diagnosis information according to the output results of the fault moment judgment model and the fault area judgment model, acquiring newly sampled data, updating a sliding window, returning to the step 401, and continuing to circularly diagnose the fault process.
Further, the step 401 includes:
intercepting the current ith input matrix X according to the sliding window mode(i)After z-score standardization, inputting the fault time judgment model into a trained fault time judgment model, outputting the fault time judgment vector with a T dimension by the model, wherein each component represents whether a corresponding time is faulted or not, namely a fault pre-judgment value, and the appointed faulted is represented by 1, and the non-faulted is represented by 0;
after a plurality of sliding time windows, aligning a plurality of obtained fault moment judgment vectors according to the actual sampling moment and accumulatively judging the number of times of fault occurrence, and if the accumulated number of times of a certain moment reaches a preset threshold value, comprehensively judging the moment as the fault occurrence moment tf', and accordingly resetting the fault prejudgment accumulated times to zero, and then executing step 402; otherwise, if the accumulated number of times at a certain time has not reached the preset threshold, it is determined that the fault has not occurred, then step 403 is executed.
Further, the step 402 includes:
intercepting the current ith input matrix X according to the sliding window mode(i)And after the z-score standardization, inputting the fault area judgment vector into a trained fault area judgment model, and outputting an N-dimensional fault area judgment vector W ═ W1，w2，…，wN]Wherein N is the number of nodes of the whole network;
if the jth node in the grid is the nearest node to the fault, the jth component w j1, otherwise, wjAnd 0, so that the adjacent node or area of the fault is obtained according to the output of the fault area judgment model.
Further, the step (1) further comprises:
and adding noise in the voltage amplitude data according to a certain signal-to-noise ratio, and simulating data containing noise obtained by real sampling.
The invention also provides an LSTM-based AC/DC hybrid power grid fault area diagnosis system, which comprises:
the system comprises a basic sample set construction unit, a fault detection unit and a fault detection unit, wherein the basic sample set construction unit is used for constructing an alternating current-direct current hybrid power grid time domain simulation model, simulating and acquiring voltage amplitude data of nodes of the whole grid under various operating conditions, and labeling labels corresponding to fault conditions to form a basic sample set;
the fault model building unit is used for respectively building a fault moment judgment model and a fault area judgment model by using an LSTM neural network, and intercepting the basic sample set by adopting a sliding window mode to form sample sets, namely a training set and a testing set, which are respectively used for training and testing the fault moment judgment model and the fault area judgment model;
the training unit is used for training the fault moment judgment model and the fault area judgment model by adopting the training set, then adjusting neural network hyper-parameters and model auxiliary parameters in the two judgment models according to the performances of the fault moment judgment model and the fault area judgment model on the test set, and repeating the training until the two judgment models meet the performance requirements on the test set;
and the diagnosis unit is used for determining whether a fault occurs or not after accumulated judgment according to the fault moment judgment model when the online application is put into use, determining the fault occurrence position according to the fault area judgment model if the fault occurs, and directly generating current fault diagnosis information if the fault occurs.
Further, the fault moment judgment model is an LSTM neural network model;
the fault moment judging model receives the sampling voltage amplitude data X intercepted by the sliding window, the sampling voltage amplitude data X is used as the input of a neural network after being standardized by z-score, then a judgment result of whether a fault occurs under the current input is output, and if the fault occurs, the judged fault occurrence moment t is outputf′。
Further, the fault area judgment model is an LSTM neural network model;
judging at the time of the fault that the model has confirmed the fault and giving a judged fault occurrence time tf' after, delayed by a number of times Δ tlUntil tf′+ΔtlVoltage amplitude data X intercepted by adopting a sliding window at each moment is standardized by z-score and then outputAnd entering the fault area judgment model.
Compared with the common neural network, the LSTM neural network can effectively extract the time sequence characteristic information required by the hidden fault diagnosis in the sampled electrical data sequence, does not depend on expert knowledge and experience to carry out complex mechanism analysis to construct an algorithm when being applied to the fault diagnosis of the alternating current-direct current hybrid power grid, does not need the characteristic construction and screening of the traditional machine learning method, and is easy to train and realize. Compared with the fault diagnosis by methods such as a Bayesian network, a Petri network and fuzzy logic, the method provided by the invention can be suitable for fault diagnosis of a larger power grid, and relevant parameters of the model can be obtained easily through training and testing without complex analysis. Meanwhile, the method has high fault diagnosis speed and high diagnosis capability, and can meet the requirements of diagnosis speed and precision of the alternating current-direct current hybrid power grid.
Drawings
Fig. 1 is a schematic flow chart diagram of a fault area diagnosis method for an LSTM-based ac/dc hybrid power grid according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a fault time determination model according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of the basic structure of an LSTM neural network according to an embodiment of the present invention;
fig. 4 is a wiring diagram of an 8-machine 36-node ac/dc hybrid power grid of the CEPRI according to the embodiment of the present invention;
fig. 5(a) and 5(b) are distributions of fault detection time errors and delays, respectively, for different noise intensities according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The embodiment of the invention provides an LSTM-based AC/DC hybrid power grid fault area diagnosis method, which comprises the following steps:
step 1, constructing an alternating current-direct current hybrid power grid time domain simulation model, simulating and obtaining voltage amplitude data of nodes of a whole grid under various operating conditions, and labeling labels corresponding to fault conditions to form a basic sample set;
step 2, respectively constructing a fault moment judgment model and a fault area judgment model by using an LSTM neural network, and intercepting the basic sample set by adopting a sliding window mode to form sample sets, namely a training set and a testing set, which are respectively used for training and testing the fault moment judgment model and the fault area judgment model;
step 3, training the fault moment judgment model and the fault area judgment model by adopting the training set, adjusting neural network hyper-parameters and model auxiliary parameters in the two judgment models according to the performances of the fault moment judgment model and the fault area judgment model on the test set, and repeating the training until the two judgment models meet the performance requirements on the test set;
and 4, when the fault moment judgment model and the fault area judgment model are put into an online application after training is finished, inputting voltage amplitude data at the current moment intercepted by the sliding window mode at each sampling moment into the fault moment judgment model, and prejudging whether the current moment has a fault or not. Determining whether a fault occurs after the pre-judgment result needs to be subjected to accumulative judgment, if so, resetting the accumulative judgment frequency of the fault moment judgment model, and executing the step 5, otherwise, executing the step 6;
step 5, after the fault is confirmed to occur, inputting voltage amplitude data after delaying a plurality of sampling moments into a fault area judgment model to diagnose the position of a fault area, and determining nodes near the position where the fault occurs;
and 6, generating current fault diagnosis information according to the output results of the fault moment judgment model and the fault area judgment model, acquiring newly sampled data, updating a sliding window, returning to the step 4, and continuing to circularly diagnose the fault process.
After a fault occurs in a complex power system, a large amount of fault information is flushed into a dispatching center, and only if the fault occurring area is found out through rapid analysis, guidance can be provided for the establishment of an emergency control strategy of the power system and time can be strived for, so that the system is ensured to recover to a normal operation state as soon as possible. Therefore, the invention constructs a fault diagnosis method for the alternating current-direct current hybrid power grid. Specifically, an LSTM neural network in a deep learning method is introduced, a fault time judgment model and a fault region diagnosis model are respectively constructed, full-network voltage amplitude data training obtained by an electric power system time domain simulation model is used for judging whether a fault occurs at a certain time and a fault region. The model can detect the fault in a short time after the fault occurs and give the fault time and the node of the area where the fault point is located, thereby providing guidance for making an emergency control strategy. According to the fault diagnosis method, the LSTM neural network is utilized to be capable of mining the hidden associated information in the time sequence, complex model mechanism analysis or artificial feature extraction is not needed, the model construction difficulty is reduced, and therefore the fault diagnosis method has strong adaptability when being applied to different power grids.
It should be noted that the basic sample set is original data generated by using power system simulation software, and a series of inputs with fixed time length are obtained after being intercepted by a sliding window and are used as samples of a fault moment judgment model and a fault area judgment model, wherein the fault moment judgment model and the fault area judgment model are both based on an LSTM neural network, the sample set is divided into a training set and a test set, the two models are trained on the training set, and the model performance and the hyper-parameters are tested on the test set.
Further, the step 1 comprises:
in the simulation model of the AC/DC hybrid power grid, conditions such as different tide operation conditions, fault lines, fault positions, fault types and the like are set, simulation data under different conditions are obtained, sampling is carried out according to a preset sampling frequency, voltage amplitude data of all nodes of the whole grid are obtained, labels corresponding to fault condition information are marked, and a basic sample set is formed.
It should be noted that each slideMatrix X consisting of input data voltage amplitudes obtained after interception of moving window(i)The form is as follows:
Figure BDA0002630060050000081
wherein Vi，jThe voltage amplitude of the ith node and the jth moment is represented, T represents the time length of the sliding window, and N represents the number of nodes in the whole network.
In order to enable the fault moment judgment model and the fault area judgment model to have good fault diagnosis effects under different operating conditions, different tide operating points, fault lines, fault positions and short-circuit fault types can be set, simulation is carried out in the time domain simulation model of the AC/DC hybrid power grid to obtain original data, and data obtained by sampling of the wide area measurement system WAMS of a real power system under different operating conditions are simulated. And then simulating the data sampling frequency of the real wide area measurement system to sample, marking whether a fault occurs, the position of the fault occurrence and other label information to obtain a basic sample set, and then respectively generating the actual sample sets for training a fault moment judgment model and a fault area judgment model.
Preferably, the step 1 further comprises:
noise is added into voltage amplitude data obtained by original simulation according to a certain signal-to-noise ratio, and data containing noise obtained by real sampling is simulated. Noise is added into the original data and used for training the model, so that the noise robustness of the model in application can be improved, the overfitting phenomenon of the model in testing can be inhibited, and the generalization capability of the model can be enhanced.
It should be noted that, when two fault determination models are trained, gaussian white Noise with a certain Signal-to-Noise Ratio (SNR) is usually added to a training sample to test the influence of the Noise intensity on the performance indexes of the two fault determination models. The signal-to-noise ratio is calculated as follows,
Figure BDA0002630060050000082
in the formula Ps、PnRepresenting the power of the signal and noise, respectively, As、AnRepresenting the signal and noise assignments, respectively.
Further, the sliding window method in step 2 specifically includes:
intercepting current sampling time T by using a time window with fixed length of TsAnd the previous data form a time series [ x ]s-T+1，xs-T+2，…，xs]. At the next instant ts+1At time, x is removed at the beginning of the sequences-T+1Addition of x at the tail of the sequences+1I.e. this time input time sequence is [ x ]s-T+2，xs-T+3，…，xs+1]The time length of the sequence remains T.
Further, the fault moment judgment model is specifically an LSTM neural network model, and is composed of an LSTM network layer and a full connection layer. The LSTM neural network is used to extract the temporal correlation features hidden in the input time series, and the classification is implemented by using the fully-connected neural network as a classifier, and the basic structure of the model is shown in fig. 2. The model receives voltage amplitude data X which is input and intercepted by the sliding window and sampled by a sampling device of the WAMS system, the voltage amplitude data X is standardized by z-score and then is used as the input of a neural network, the model outputs the judgment result of whether the fault occurs under the current input, and if the fault occurs, the judged fault occurrence time t is outputf′。
Further, the fault area determination model is specifically an LSTM neural network model, and at the fault time, the determination model determines that the fault has occurred and gives a determined fault occurrence time tf' thereafter, a delay of a number of times Δ tlUntil tf′+ΔtlAnd voltage amplitude data X intercepted by adopting a sliding window at each moment is input into a fault area judgment model after being standardized by z-score. The area where the fault occurs is judged more accurately by analyzing the information before the fault and the information after a small part of the fault.
Further, the method of respectively intercepting the basic sample sets in a sliding window manner to form sample sets respectively used for training and testing the fault moment judgment model and the fault area judgment model specifically comprises the following steps:
intercepting original simulation data by using a sliding window to form a sample set for training and testing a model, and generating a fault at a moment tfCan appear at any time in the sliding window, so that the model can be trained by using a sample at which the fixed fault occurs, so that the overfitting of the model can be avoided. The label of the fault moment model is set according to the fault moment, and the label of the fault area judgment model is set according to the adjacent node of the area where the fault occurs.
In practical application, the fault occurrence time in the real-time data obtained by sampling is uncertain, so that the model can be prevented from being only suitable for a certain specific fault occurrence time t by adopting the sample training model without fixing the fault occurrence timefAnd detection is performed in time, so that the generalization capability and robustness of the model are improved.
Further, the step 2 further comprises:
and processing the sample sets for the training and testing of the two judgment models by adopting a z-score standardization method to obtain a standardized voltage amplitude matrix X' which is used as the actual input of the fault moment judgment model and the fault area judgment model.
Specifically, the z-score normalization method is: for a data set with N number of samples, X ═ X(1)，X(2)，…，X(N)]The data set is z-score normalized as follows, where
Figure BDA0002630060050000101
Represents the ith sample matrix X(i)The m-th row and the n-th column of (1),
Figure BDA0002630060050000102
representing the ith sample matrix X after normalization(i)The element of the m-th row and n-th column of' are included.
Figure BDA0002630060050000103
Figure BDA0002630060050000104
Figure BDA0002630060050000105
The z-score standardization method can normalize different characteristics of the sample, so that the characteristic value range is close to that of the sample, and the model can be trained by using a gradient descent algorithm. The voltage amplitude matrix is processed by adopting a z-score standardization method, so that the characteristics of the samples can be jointly scaled to a close range, the training of a neural network model is facilitated, and the convergence speed is accelerated.
Further, the step 2 comprises:
and (3) respectively constructing two judgment models by using an LSTM neural network, and then randomly disordering the sample set obtained in the sliding window mode and dividing the sample set into a training set and a test set according to a certain proportion (such as 4: 1).
The two judgment models are trained by using a supervised learning method, the accuracy of the models on a test set is used as a training index, after the model training is finished, the structural hyper-parameters and model auxiliary parameters of the two models are properly adjusted, the hyper-parameters comprise the number of layers in a neural network, the number of neurons in each layer, the activation functions of the neurons and the like, the model auxiliary parameters comprise the threshold value of the cumulative judgment of the model at the fault moment and the like, then the training is repeated, the change of the training index is observed, and the model with the optimal performance is selected for the practical online application.
Further, the LSTM neural network is a deep neural network, similar to a fully-connected neural network, and is stacked by multiple neural network layers, but each layer is composed of LSTM neurons with memory capability, and the basic structure of the LSTM neural network is shown in fig. 3. The LSTM neural network can selectively memorize the important information in the input before and forget the unimportant information, can excavate the relation between the dependence on the hidden time sequence in the input and the output, is suitable for processing the time sequence input, and has stronger feature extraction capability compared with the common fully-connected neural network.
The alternating current-direct current hybrid system is used as a space-time high-dimensional nonlinear system, has the characteristic of space-time multi-scale coupling, and needs to consider the electromechanical transient process with large time scale and the electromagnetic transient process with small time scale. The LSTM neural network is adopted to construct the fault diagnosis model, so that the time sequence characteristics of the alternating current-direct current hybrid system in the complex transient process can be fully excavated, and the rapid detection and the area positioning of the fault can be realized more easily.
Further, the step 4 comprises:
when the two judgment models are put into online application after training is finished, intercepting the current ith input matrix X according to the sliding window mode(i)And after the z-score standardization, inputting the fault time judgment model into a trained fault time judgment model, outputting the model as a T-dimensional fault time judgment vector, wherein each component represents whether a corresponding time has a fault or not, namely a fault pre-judgment value, for example, if the fault is expressed by 1, the fault is not expressed by 0. After a plurality of sliding time windows, aligning a plurality of obtained fault moment judgment vectors according to the actual sampling moment and accumulatively judging the number of times of fault occurrence, and if the accumulated number of times of a certain moment reaches a preset threshold value, comprehensively judging the moment as the fault occurrence moment tf', and accordingly, the failure prediction accumulated number is cleared, and then step 5 is performed. Otherwise, if the accumulated times at a certain time have not reached the preset threshold, the fault is considered not to occur, and then step 6 is executed.
Due to the influence of noise or other accidental factors, the situation that the fault moment judgment model is inaccurate in single judgment can occur, so that the accuracy and robustness of the model fault moment judgment can be improved by using the mode of accumulative judgment.
Further, the step 5 comprises:
similar to the fault moment judgment model, intercepting the current ith input matrix X according to the sliding window mode(i)And after the z-score standardization, inputting the fault area judgment into a trained fault area judgmentIn the fault model, the model outputs an N-dimensional fault region judgment vector W ═ W1，w2，…，wN]And N is the number of nodes of the whole network. And defining that if the power grid fault occurs in a certain line, the nodes at two ends of the line are the nearest nodes to the fault. Appointing that if the jth node in the power grid is the nearest node of the fault, the jth component w j1, otherwise, w j0. Therefore, the adjacent node or area of the fault can be obtained from the output of the fault area judgment model.
Further, the step 6 includes:
when the two judgment models are put into online application, at the sampling time of each sliding window, the judgment of whether a fault occurs or not can be obtained through the fault time judgment model and the fault area judgment model, and if the fault occurs, the fault time and the diagnosis information of the area where the fault is represented by the node are obtained after the fault occurs. This information is output and provided to dispatch operators for reference to assist in the formulation of emergency control strategies or other applications. And then, acquiring data of the next sampling moment according to the sliding window mode, and repeating the fault diagnosis process of the step 4-6.
The data are input by using the sliding window mode, and the input matrix X is updated by using new sampling data, so that the on-line continuous monitoring of the power system can be conveniently realized, and the safety of the power system is ensured in real time.
The embodiment of the invention also provides an LSTM-based AC/DC hybrid power grid fault area diagnosis system, which comprises the following steps:
the system comprises a basic sample set construction unit, a fault detection unit and a fault detection unit, wherein the basic sample set construction unit is used for constructing an alternating current-direct current hybrid power grid time domain simulation model, simulating and acquiring voltage amplitude data of nodes of the whole grid under various operating conditions, and labeling labels corresponding to fault conditions to form a basic sample set;
the fault model building unit is used for respectively building a fault moment judgment model and a fault area judgment model by using an LSTM neural network, and intercepting the basic sample set by adopting a sliding window mode to form sample sets, namely a training set and a testing set, which are respectively used for training and testing the fault moment judgment model and the fault area judgment model;
the training unit is used for training the fault moment judgment model and the fault area judgment model by adopting the training set, then adjusting neural network hyper-parameters and model auxiliary parameters in the two judgment models according to the performances of the fault moment judgment model and the fault area judgment model on the test set, and repeating the training until the two judgment models meet the performance requirements on the test set;
and the diagnosis unit is used for determining whether a fault occurs or not after accumulated judgment according to the fault moment judgment model when the online application is put into use, determining the fault occurrence position according to the fault area judgment model if the fault occurs, and directly generating current fault diagnosis information if the fault occurs.
The functions of each unit can be referred to the description of the foregoing method embodiments, and are not described herein again.
For better illustration of the invention, the following examples are given:
the invention is explained by taking an 8-machine 36-node alternating-current and direct-current hybrid power grid of CEPRI as an example. In the off-line training stage of the model, an 8-machine 36-node system simulation model of CEPRI is constructed on a PSASP platform, the system has 27 alternating current lines and 1 direct current line, and a single line diagram is shown in FIG. 4. The simulation time length is set to be 2s in the PSASP, the simulation step length is set to be 0.001s, the time when the fault occurs under the fault working condition or the normal load disturbance occurs under the fault-free working condition is set to be 0.5s, namely the simulation time point length is 2001, and the fault occurrence time point is 501.
For time domain simulation of a sample under a fault operation condition, setting the fault type as a three-phase short circuit grounding fault; the fault lines are set as all alternating current transmission lines in the system, and the number of the fault lines is 27; the positions of short-circuit faults are set to be 9 in 10%, 20%, … and 90%; setting different power flow changes of the whole system, namely setting that all the power of the generators and the loads are multiplied by a coefficient simultaneously to change the running state of the power system, and setting 5 coefficients in total, wherein the coefficients are 1.15, 1.1, 1.0, 0.9 and 0.8 respectively. Thus, a total of 27 × 9 × 5 samples of 1215 base samples of fault conditions can be simulated, and the simulation settings under fault conditions are summarized in table 1.
TABLE 1 Fault Condition settings
Sample set-up Detailed description of the invention Number of species
Type of failure Three-phase short circuit earth fault 1
Fault line All AC lines 27
Location of failure 10％，20％，30％，…，90％ 9
Tidal current change 1.15，1.1，1.0，0.9，0.8 5
For the time domain simulation of the samples under the fault-free operating condition, similar to the fault operating condition, the specific settings are as shown in table 2, and a total of 5 × 10 × 7 to 350 basic samples of the fault-free operating condition can be simulated.
TABLE 2 No-Fault Condition settings
Figure BDA0002630060050000141
Intercepting original simulation data by a sliding window method according to a sample set generation method for the training and testing model, wherein the fault occurrence time tfCan occur at any time within the sliding window, resulting in 1215 and 350 samples for actual training and testing, respectively.
After the samples are obtained, noise of 40dB, 35dB and 30dB is added respectively according to the method for adding the noise to train and test the performance of the model. Recording fault occurrence time t judged by fault time judgment modelf' with the actual moment of failure tfThe difference between them is the fault detection time error Δ tfAnd the moment t when the fault moment judgment model actually detects the faultdAnd the time t of fault occurrencefIs the fault detection time delay deltatdThe cumulative determination threshold of the failure time determination model is set to 3.
Through testing, the fault detection time error delta t under different noise intensities can be obtainedfAnd a delay Δ tdThe distribution of (c) is as shown in fig. 5(a) and 5 (b). As can be seen from fig. 5(a), the failure time determination model can detect a failure within a short time after the failure occurs; as can be seen from fig. 5(b), the detection error of the fault is also small, and the online rapid detection of the power system fault can be realized.
Then, the fault area judgment model is tested, and the model positioning accuracy under different noises is shown in table 3, so that the model can be found to have certain noise robustness, and the online rapid positioning of the power system fault can be realized.
TABLE 3 Fault area determination model location accuracy
Noiseless Containing 40dB of noise Containing 35dB of noise Containing 30dB of noise
99.59％ 98.77％ 97.12％ 95.47％
Different from the traditional artificial intelligence method, the fault diagnosis model constructed by the deep neural network can directly process the original sampling data and obtain effective characteristics from the original sampling data for fault diagnosis analysis without manually selecting the characteristics, does not depend on expert knowledge for complex mechanism analysis and modeling, and can provide a fault diagnosis result more quickly and accurately.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (8)

1. An LSTM-based AC/DC hybrid power grid fault area diagnosis method is characterized by comprising the following steps:
(1) constructing an alternating current-direct current hybrid power grid time domain simulation model, simulating and obtaining voltage amplitude data of nodes of the whole grid under various operating conditions, and labeling labels corresponding to fault conditions to form a basic sample set;
(2) respectively constructing a fault moment judgment model and a fault area judgment model by using an LSTM neural network, and intercepting the basic sample set by adopting a sliding window mode to form sample sets, namely a training set and a testing set, which are respectively used for training and testing the fault moment judgment model and the fault area judgment model;
(3) training the fault moment judgment model and the fault area judgment model by adopting the training set, then adjusting neural network hyper-parameters and model auxiliary parameters in the two judgment models according to the performances of the fault moment judgment model and the fault area judgment model on the test set, and repeating the training until the two judgment models meet the performance requirements on the test set;
(4) when the online application is put into use, determining whether a fault occurs or not after accumulated determination according to the fault moment determination model, if so, determining the fault occurrence position according to the fault area determination model, and otherwise, directly generating current fault diagnosis information;
the step (4) comprises the following steps:
(401) at each sampling moment, inputting voltage amplitude data at the current moment intercepted in a sliding window mode into a fault moment judgment model, and prejudging whether the current moment has a fault or not; determining whether a fault occurs after the pre-judgment result needs to be subjected to accumulative judgment, if so, resetting the accumulative judgment frequency of the fault moment judgment model, and executing the step (402), otherwise, executing the step (403);
(402) after the fault is confirmed to occur, inputting voltage amplitude data after delaying a plurality of sampling moments into a fault area judgment model to diagnose the position of a fault area, and determining nodes near the position where the fault occurs;
(403) generating current fault diagnosis information according to the output results of the fault moment judgment model and the fault area judgment model, acquiring newly sampled data, updating a sliding window, returning to the step (401), and continuing to circularly diagnose the fault process;
the step (401) comprises:
intercepting the current ith input matrix X according to the sliding window mode(i)After z-score standardization, inputting the fault time judgment model into a trained fault time judgment model, outputting the fault time judgment vector with a T dimension by the model, wherein each component represents whether a corresponding time is faulted or not, namely a fault pre-judgment value, and the appointed faulted is represented by 1, and the non-faulted is represented by 0;
after a plurality of sliding time windows, aligning a plurality of obtained fault moment judgment vectors according to the actual sampling moment and accumulatively judging the number of times of fault occurrence, and if the accumulated number of times of a certain moment reaches a preset threshold value, comprehensively judging the moment as the fault occurrence moment tf', and accordingly resetting the fault pre-judgment accumulated times to zero, and then executing the step (402); otherwise, if the accumulated number of times at a certain time has not reached the preset threshold, it is determined that the fault has not occurred, then step (403) is executed.
2. The method for diagnosing the fault area of the AC-DC hybrid power grid according to claim 1, wherein the fault time judgment model receives sampled voltage amplitude data X intercepted by the sliding window, and the sampled voltage amplitude data X is used as an input of a neural network after being standardized by z-score, and then a judgment result of whether a fault occurs under the current input is output, and if the fault occurs, the judged fault occurrence time t is outputf′。
3. The ac-dc hybrid power grid fault area diagnosis method according to claim 2, wherein at the fault time, the judgment model has confirmed the fault and gives a judged fault occurrence time tf' after, delayed by a number of times Δ tlUntil tf′+ΔtlAnd voltage amplitude data X intercepted by adopting a sliding window at each moment is input into the fault area judgment model after being standardized by z-score.
4. The method for diagnosing the fault area of the AC/DC hybrid grid according to claim 1, wherein the step (402) comprises:
intercepting the current ith input matrix X according to the sliding window mode(i)And after the z-score standardization, inputting the fault area judgment vector into a trained fault area judgment model, and outputting an N-dimensional fault area judgment vector W ═ W1,w2,…,wN]Wherein N is the number of nodes of the whole network;
if the jth node in the grid is faultyNearest node, then the jth component wj1, otherwise, wjAnd 0, so that the adjacent node or area of the fault is obtained according to the output of the fault area judgment model.
5. The AC/DC hybrid power grid fault region diagnosis method according to any one of claims 1-4, wherein the step (1) further comprises:
and adding noise in the voltage amplitude data according to a certain signal-to-noise ratio, and simulating data containing noise obtained by real sampling.
6. An alternating current-direct current series-parallel connection power grid fault area diagnostic system based on LSTM is characterized by comprising:
the system comprises a basic sample set construction unit, a fault detection unit and a fault detection unit, wherein the basic sample set construction unit is used for constructing an alternating current-direct current hybrid power grid time domain simulation model, simulating and acquiring voltage amplitude data of nodes of the whole grid under various operating conditions, and labeling labels corresponding to fault conditions to form a basic sample set;
the fault model building unit is used for respectively building a fault moment judgment model and a fault area judgment model by using an LSTM neural network, and intercepting the basic sample set by adopting a sliding window mode to form sample sets, namely a training set and a testing set, which are respectively used for training and testing the fault moment judgment model and the fault area judgment model;
the training unit is used for training the fault moment judgment model and the fault area judgment model by adopting the training set, then adjusting neural network hyper-parameters and model auxiliary parameters in the two judgment models according to the performances of the fault moment judgment model and the fault area judgment model on the test set, and repeating the training until the two judgment models meet the performance requirements on the test set;
the diagnosis unit is used for determining whether a fault occurs or not after accumulated judgment according to the fault moment judgment model when the diagnosis unit is put into online application, if so, determining the fault occurrence position according to the fault area judgment model, and otherwise, directly generating current fault diagnosis information;
the diagnostic unit performs the steps of:
(401) at each sampling moment, inputting voltage amplitude data at the current moment intercepted in a sliding window mode into a fault moment judgment model, and prejudging whether the current moment has a fault or not; determining whether a fault occurs after the pre-judgment result needs to be subjected to accumulative judgment, if so, resetting the accumulative judgment frequency of the fault moment judgment model, and executing the step (402), otherwise, executing the step (403);
(402) after the fault is confirmed to occur, inputting voltage amplitude data after delaying a plurality of sampling moments into a fault area judgment model to diagnose the position of a fault area, and determining nodes near the position where the fault occurs;
(403) generating current fault diagnosis information according to the output results of the fault moment judgment model and the fault area judgment model, acquiring newly sampled data, updating a sliding window, returning to the step (401), and continuing to circularly diagnose the fault process;
the step (401) comprises:
intercepting the current ith input matrix X according to the sliding window mode(i)After z-score standardization, inputting the fault time judgment model into a trained fault time judgment model, outputting the fault time judgment vector with a T dimension by the model, wherein each component represents whether a corresponding time is faulted or not, namely a fault pre-judgment value, and the appointed faulted is represented by 1, and the non-faulted is represented by 0;
after a plurality of sliding time windows, aligning a plurality of obtained fault moment judgment vectors according to the actual sampling moment and accumulatively judging the number of times of fault occurrence, and if the accumulated number of times of a certain moment reaches a preset threshold value, comprehensively judging the moment as the fault occurrence moment tf', and accordingly resetting the fault pre-judgment accumulated times to zero, and then executing the step (402); otherwise, if the accumulated number of times at a certain time has not reached the preset threshold, it is determined that the fault has not occurred, then step (403) is executed.
7. The AC-DC hybrid power grid fault area diagnosis system of claim 6, wherein the fault moment judgment model receives the intercepted data of the sliding windowSampling voltage amplitude data X is subjected to z-score standardization to be used as input of a neural network, then a judgment result of whether a fault occurs under the current input is output, and if the fault occurs, the judged fault occurrence time t is outputf′。
8. The system according to claim 7, wherein the fault time determining model determines the fault and gives a determined fault occurrence time t at the fault timef' after, delayed by a number of times Δ tlUntil tf′+ΔtlAnd voltage amplitude data X intercepted by adopting a sliding window at each moment is input into the fault area judgment model after being standardized by z-score.
CN202010810874.9A 2020-08-12 2020-08-12 Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM Active CN112051481B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010810874.9A CN112051481B (en) 2020-08-12 2020-08-12 Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010810874.9A CN112051481B (en) 2020-08-12 2020-08-12 Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM

Publications (2)

Publication Number Publication Date
CN112051481A CN112051481A (en) 2020-12-08
CN112051481B true CN112051481B (en) 2021-05-18

Family

ID=73602751

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010810874.9A Active CN112051481B (en) 2020-08-12 2020-08-12 Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM

Country Status (1)

Country Link
CN (1) CN112051481B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP4261639A4 (en) * 2020-12-08 2024-01-31 JFE Steel Corporation Time series signal trigger condition determination method, method for diagnosing abnormality in monitored facility, and time series signal trigger condition determination device
CN112926752B (en) * 2021-01-26 2022-07-01 西华大学 Intelligent power transmission network fault line detection and fault recovery method considering energy storage system
CN113011656B (en) * 2021-03-22 2022-08-02 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 Power station auxiliary machine fault early warning method and system
CN113484685B (en) * 2021-07-16 2023-08-18 西华大学 Power grid fault diagnosis method based on time sequence organization type P system
CN113536607B (en) * 2021-09-15 2022-01-25 广东电网有限责任公司东莞供电局 Transformer substation signal transmission system evaluation method and system
CN113884818B (en) * 2021-10-27 2024-04-09 国网江苏省电力有限公司徐州供电分公司 Method for accurately estimating arrival time of fault traveling wave of power distribution network based on LSTM
CN114692674B (en) * 2022-02-21 2024-09-17 武汉船用电力推进装置研究所（中国船舶重工集团公司第七一二研究所） Ship propulsion system fault judging method and system
CN114566964B (en) * 2022-04-29 2022-07-29 国网天津市电力公司电力科学研究院 Power distribution network feeder automation control method, device, equipment and storage medium
CN115184728A (en) * 2022-07-13 2022-10-14 西南交通大学 Fault identification method for AC/DC hybrid power distribution network
CN115905835B (en) * 2022-11-15 2024-02-23 国网四川省电力公司电力科学研究院 Low-voltage alternating current arc fault diagnosis method integrating multidimensional features
CN117454231B (en) * 2023-11-27 2024-07-02 广州得元电力科技有限公司 Subway direct current protection test system applying intelligent simulation technology
CN117725981B (en) * 2024-02-08 2024-04-30 昆明学院 Power distribution network fault prediction method based on optimal time window mechanism
CN117909909B (en) * 2024-03-19 2024-06-14 青岛鼎信通讯股份有限公司 Arc grounding fault identification method for power distribution network

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108120900B (en) * 2017-12-22 2020-02-11 北京映翰通网络技术股份有限公司 Power distribution network fault positioning method and system
US20200011932A1 (en) * 2018-07-05 2020-01-09 Nec Laboratories America, Inc. Battery capacity fading model using deep learning
CN110596506A (en) * 2019-10-30 2019-12-20 福州大学 Converter fault diagnosis method based on time convolution network
CN110829417B (en) * 2019-11-14 2023-03-31 电子科技大学 Electric power system transient stability prediction method based on LSTM double-structure model

Also Published As

Publication number Publication date
CN112051481A (en) 2020-12-08

Similar Documents

Publication Publication Date Title
CN112051481B (en) Alternating current-direct current hybrid power grid fault area diagnosis method and system based on LSTM
Teimourzadeh et al. High impedance single-phase faults diagnosis in transmission lines via deep reinforcement learning of transfer functions
CN110829417B (en) Electric power system transient stability prediction method based on LSTM double-structure model
CN102707256B (en) Fault diagnosis method based on BP-Ada Boost nerve network for electric energy meter
CN111965476B (en) Low-voltage diagnosis method based on graph convolution neural network
CN112200694B (en) Dominant instability mode identification model construction and application method based on graph neural network
CN110687393B (en) Valve short-circuit protection fault positioning method based on VMD-SVD-FCM
CN105606914A (en) IWO-ELM-based Aviation power converter fault diagnosis method
CN112180204A (en) Power grid line fault diagnosis method based on electric quantity information
Li et al. Physics-informed learning for high impedance faults detection
CN111680875A (en) Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model
CN113850320A (en) Transformer fault detection method based on improved support vector machine regression algorithm
CN116205265A (en) Power grid fault diagnosis method and device based on deep neural network
Kezunovic Translational knowledge: From collecting data to making decisions in a smart grid
CN114169395A (en) Construction method and application of dominant instability mode recognition model of power system
CN111062569A (en) Low-current fault discrimination method based on BP neural network
Coban et al. Artificial neural network based fault location on 230 kv transmission line using voltage and current signals
CN114384319A (en) Grid-connected inverter island detection method, system, terminal and medium
Morsalin et al. Induction motor inter-turn fault detection using heuristic noninvasive approach by artificial neural network with Levenberg Marquardt algorithm
Li et al. Artificial intelligence for real-time topology identification in power distribution systems
CN113139601B (en) Switch fault detection method based on LSTM and attention mechanism
CN115456106A (en) High-voltage circuit breaker fault diagnosis model optimization method
Vega-Martinez et al. Hybrid data-driven physics-based model framework implementation: Towards a secure cyber-physical operation of the smart grid
Ali et al. The Transmission line fault detection and identification using Artificial Neural Network
Ji et al. A graph-theoretic method for identification of electric power distribution system topology

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