CN112234641B - Direct-current commutation failure blocking and preventing control method - Google Patents
Direct-current commutation failure blocking and preventing control method Download PDFInfo
- Publication number
- CN112234641B CN112234641B CN202010997688.0A CN202010997688A CN112234641B CN 112234641 B CN112234641 B CN 112234641B CN 202010997688 A CN202010997688 A CN 202010997688A CN 112234641 B CN112234641 B CN 112234641B
- Authority
- CN
- China
- Prior art keywords
- commutation failure
- fault
- line voltage
- time
- 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
Links
Images
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
- H02J3/36—Arrangements for transfer of electric power between ac networks via a high-tension dc link
-
- 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]
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/60—Arrangements for transfer of electric power between AC networks or generators via a high voltage DC link [HVCD]
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a direct current commutation failure blocking and preventing control method, which comprises the following steps: predicting the tracks of the bus voltage of the power grid inverter station in the fault period and the fault recovery stage by an online disturbed voltage transient track intelligent prediction model, and calculating to obtain the bus voltage effective value of the power grid inverter station; and calculating feasible regions and maximum values of the trigger angle instruction values without commutation failure based on the constraint calculation of the effective values of the commutation line voltages and the trigger angle instruction values so as to block and prevent secondary commutation failure. The method can quickly and accurately predict whether the secondary commutation failure occurs in the fault period within ms-level time, quickly and pertinently pre-judge the impending commutation failure in the system, reserve enough time margin and safety margin for subsequent control, realize quick prediction of short-term commutation phase voltage change, avoid the secondary commutation failure in the fault period, and realize a control strategy which can ensure timeliness, reliability and result constraint.
Description
Technical Field
The invention belongs to the field of power control, and particularly relates to a direct-current commutation failure blocking and preventing control method.
Background
The commutation failure is one of the faults with higher occurrence probability of the direct current transmission system. In the converter, the valve which is out of conduction can not recover the blocking capability within a period of time when the reverse voltage acts, or the phase change process is not completed during the reverse voltage, when the valve voltage changes to the positive direction, the phase of the valve which is changed to the phase of the valve which is originally scheduled to be out of conduction is changed, and the condition is called phase change failure.
Among many factors causing the commutation failure, the reduction of the commutation voltage and the phase change caused by the short-circuit fault of the alternating current system can generate direct adverse effect on the commutation process, and are main factors causing the commutation failure.
The current research generally holds that the first commutation failure (lasting from 2/3 base frequency period to 4/3 base frequency period) after the disturbance occurs can not be avoided and the commutation failure can recover by itself under the normal condition. And the strategies such as preventive control and the like only aim at the secondary commutation failure of the subsequent fault.
The current commutation failure prejudgment criterion can be divided into two categories: based on commutation voltage, based on turn-off angle. The prejudgment based on the commutation voltage comprises a critical commutation voltage judgment method and a waveform transformation judgment method. The method comprises the following steps that 1, pre-judgment based on commutation voltage only considers the influence of external factors such as commutation voltage, and the characteristic change of the commutation voltage is taken as the basis of the pre-judgment, but the control characteristic of a direct current system is not fully considered; the judgment is only carried out aiming at the severity of the fault, but not aiming at the commutation failure risk judgment, so that the currently adopted commutation failure prevention module has a large amount of false start and missed start. 2. In the judgment based on the turn-off angle, the method of actually measuring the arc extinguishing angle signal is adopted, negative zero-crossing voltage is detected through a thyristor voltage detection board, the zero-crossing voltage signal is sent to valve base electronic equipment (VBE) to generate a valve current zero-crossing signal, and the valve current zero-crossing signal is compared with a phase-locked voltage zero-crossing signal to obtain the turn-off angle, but the scheme depends on the zero-crossing signal of current, the sensitivity is easy to be influenced, and the method does not belong to prediction in strict sense; the extinction angle prediction judgment method predicts the commutation failure phenomenon and calculates the commutation failure margin by comparing the relevant solid angle of interruption with the predicted turn-off angle of interruption, but the current calculation of the predicted turn-off angle is based on the prediction of direct current under a quasi-steady-state model, when the system is in a transient state, the calculation may have larger deviation, the prediction accuracy is limited, and the reference is difficult to provide for subsequent prevention and control.
The commutation failure is prevented from the source, the coordination optimization of device-system level multi-layer multi-physical coupling is relied on, and the optimization objects comprise hardware equipment, control strategies, network structures and the like. Among them, optimization measures based on control strategies have been widely studied for economy and convenience. According to a physical mechanism of a phase change process, the currently adopted optimization method is mainly based on an existing controller in direct current, and the phase change margin is increased by limiting direct current and increasing a leading trigger angle to relieve the risk of phase change failure. However, the existing control strategies mainly have the following problems:
1) Only the commutation physical process is controlled qualitatively, so that the control result is sometimes conservative and sometimes aggressive, the commutation feasible domain boundary cannot be clearly defined accurately according to different transient conditions, and the control economy is more difficult to further improve on the basis;
2) The current control optimization is mainly based on the adjustment of parameters and structures by existing controls such as constant current, minimum extinction angle and the like, the original control characteristics of direct current can be changed, and the adaptability is insufficient under different transient conditions and operating conditions;
3) The parameter design mode lacks universality;
4) The control of a single physical factor in the phase conversion process is adopted, so that the optimization feasible area of the controller is limited.
The existing measure for preventing and controlling the commutation failure is difficult to reconcile the contradiction between the accurate and quick suppression of the commutation failure and the economy, and the potential of transient response and regulation and control characteristics of high-voltage direct-current transmission is not exploited.
Disclosure of Invention
In view of the above problems, the present invention provides a method for blocking and preventing dc commutation failure.
The direct current commutation failure blocking and preventing control method comprises the following steps:
uninterrupted acquisition of bus line voltage u of power grid inversion station ab (t)、u bc (t)、u ca (t) when detecting the occurrence of an AC system fault, extracting the time t before said fault occurred 1 To the time t after the fault occurs 3 The bus line voltage u of the grid inverter station ab (t)、u bc (t)、u ca (t), t represents time;
data extraction for the first time: extracting the time t before the fault occurs 1 To the time t after the fault occurs 3 The bus line voltage u of the power grid inversion station ab (t)、u bc (t)、u ca (t)；
Inputting the data obtained by the first data extraction into an online secondary commutation failure intelligent prediction modelJudging whether secondary commutation failure occurs, and if the secondary commutation failure does not occur, ending the execution of the direct current commutation failure blocking and preventing control method; if the secondary commutation failure is judged to occur, performing secondary data extraction: extracting the fault occurrence time t start To the time t after the fault occurs 3 The bus line voltage u of the grid inverter station ab (t)、u bc (t)、u ca (t)；
Inputting the data obtained by the second data extraction into an online disturbed voltage transient state track intelligent prediction model;
predicting the bus line voltage of the power grid inversion station to be [ t ] by the online disturbed voltage transient state track intelligent prediction model 3 ，t 2 ]And calculating to obtain [ t 4 ，t 2 ]The effective value of the bus voltage of the power grid inversion station is obtained;
based on the bus voltage effective value of the power grid inversion station and a trigger angle instruction value constraint formula, calculating to obtain [ t [ [ t ] 4 ，t 2 ]A condition satisfied by a firing angle command value for which commutation failure does not occur, to block and prevent occurrence of the secondary commutation failure,
wherein, let t end If it is the fault clearing time, the time t 1 、t 2 、t 3 And the time t of the fault occurrence start And a fault clearing time t end The relationship between (A) and (B) is:
[t 1 ，t start ]is the fundamental frequency cycle prior to the occurrence of the fault;
[t end ，t 2 ]is 5 fundamental frequency periods after the fault is over;
[t start ，t 3 ]is the first fundamental period after the fault occurs.
Further, in the present invention,
the online secondary commutation failure intelligent prediction model is obtained by the following method, and comprises the following steps:
A. generating a multi-source training data set;
B. and on the basis of a supervised learning method, performing off-line training on the intelligent agent by using the multi-source training data set to construct the online secondary commutation failure intelligent prediction model.
Further, in the present invention, it is preferable that,
the sources of the multi-source training dataset include: the method comprises the steps of simulating samples of an electromagnetic transient state and an electromechanical-electromagnetic transient state hybrid simulation of an alternating-current and direct-current power grid, and real-time measurement data obtained by measuring an alternating-current and direct-current power grid inverter station in real time through a power grid phasor measuring device and a power fault recorder;
the multi-source training data set comprises m samples, and the data of each sample records the time t before the fault occurs in the simulation sample and the real-time measurement data 1 To the time t after the fault occurs 2 The number of samples m is a positive integer greater than 1,
the physical quantity in the data of each sample includes: bus line voltage u of power grid inverter station ab (t)、u bc (t)、u ca (t) and a grid inverter station arc-quenching angle gamma (t);
the simulation sample sets the following faults in the electromagnetic transient and electromechanical-electromagnetic transient hybrid simulation of the alternating current-direct current power grid:
the types of faults include: single-phase grounding short circuit, two-phase short circuit, three-phase short circuit and two-phase grounding short circuit on the AC system line;
the position where the fault occurs: selecting positions with different electrical distances from a bus of the power grid inversion station to set faults;
timing of the fault: setting fault angles as different angles with intervals of theta degrees from 0 degrees to 180 degrees; wherein theta is more than 0 and less than 30;
the fault duration is: setting corresponding fault duration according to the actual relay protection action time of the simulated alternating current power grid;
the size of the fault: setting faults of different severity according to the scale of the simulated alternating current power grid.
Further, in the present invention,
carrying out the uninterrupted acquisition of the bus line voltage u of the power grid inverter station ab (t)、u bc (t)、u ca The acquisition frequency at (t) is 2000 Hz-10000 Hz;
the number m of the samples satisfies that m is more than or equal to 100 and less than or equal to 10000.
Further, in the present invention,
the acquisition frequency is 5000Hz;
the number of samples m =3000;
the fault duration is 0.1s, s for seconds.
Further, in the present invention,
the step B comprises the following steps:
step b1: extracting data characteristics;
step b2: selecting a model, namely selecting a convolutional neural network method on a software platform with a machine learning function to train a prediction model;
step b3: determining model input quantity hin and model output quantity hout;
step b4: defining the architecture, parameters and training options of the convolutional neural network;
step b5: splitting a sample to obtain a model training set and a model verification set;
step b6: model training, namely training a prediction model by adopting the convolutional neural network and a model training set to obtain the online secondary commutation failure intelligent prediction model;
step b7: and the model verification is passed.
Further, in the present invention,
the step b1 comprises the following steps:
for each sample data in the multi-source training data set, the following processing is carried out:
the bus line voltage u of the grid inverter station in the time period of the fundamental frequency cycle before the fault occurs ab (t)、u bc (t) and u ca (t) Fourier transform to obtain an amplitude spectrum (U) of the line voltage of 1 xn ab ) 1×n 、(U bc ) 1×n 、(U ca ) 1×n And 1 xn line voltage phase spectrum
Wherein the content of the first and second substances,
(U ab ) 1×n ＝[U ab0 U ab1 U ab2 ... U abn-1 ]；
(U bc ) 1×n ＝[U bc0 U bc1 U bc2 ... U bcn-1 ]；
(U ca ) 1×n ＝[U ca0 U ca1 U ca2 ... U can-1 ]；
0 is a direct current component after the Fourier transform, 1 is a fundamental frequency, 2 and 3 … n-1 are harmonic series, n-1 is the highest harmonic series after the Fourier transform, and n-1 is the harmonic of the level n-1 obtained by the Fourier transform; u shape ab0 For said line voltage u ab (t) amplitude of the direct current component, U ab1 To U abn-1 Respectively forming the line voltage u ab (t) the magnitude of each frequency component,
for said line voltage u ab (t) the phase angle of the direct current component,
to
To form the line voltage u ab (t) the phase of each frequency component at the time origin; u shape bc0 For said line voltage u bc (t) amplitude of the direct current component, U bc1 To U bcn-1 Respectively constitute the line voltage u bc (t) the magnitude of each frequency component of (t),
for said line voltage u bc (t) the phase angle of the direct current component,
to
To form the line voltage u bc (t) the phase of each frequency component at the time origin; u shape ca0 For said line voltage u ca (t) amplitude of the DC component, U ca1 To U can-1 Respectively constitute the line voltage u ca (t) the magnitude of each frequency component of (t),
for said line voltage u ca (t) the phase angle of the direct current component,
to
To form said line voltage u ca (t) the phase of each frequency component at the time origin;
creating a 6 x n feature matrix A using the amplitude and phase spectra 6×n ：
And U is ab1 、U bc1 、U ca1 Respectively the fundamental frequency amplitude of the bus bar voltage of the power grid inverter station,
repeating the above calculation process on the bus line voltage of the power grid inverter station in the first fundamental frequency period after the fault occurs, and creating a characteristic matrix B of 6 multiplied by n 6×n ，
The feature matrix A 6×n And the feature matrix B 6×n Splicing, creating a 12 xn feature matrix C 12×n ：
Obtaining a corresponding data characteristic matrix C for each sample 12×n ，
For m sample data in the multi-source training data set, distributing fail or success labels to each sample according to the value of the extinction angle gamma (t) of the power grid inverter station from 1.5 fundamental frequency periods after the fault occurs to 5 fundamental frequency periods after the fault ends, wherein the distribution rule is that when the gamma (t) of a certain sample is [ t ], [ t ] 4 ，t 2 ]If zero value appears in the sample, the sample is marked as fail, namely the secondary commutation failure occurs; otherwise, the sample is marked as success, namely the secondary commutation failure does not occur, and finally each sample has a corresponding label t 4 Is the time at which 1.5 fundamental frequency periods are located after the fault occurs, t 2 For 5 fundamental frequency cycles after the end of the fault,
the data feature matrix C 12×n And generating characteristic pictures as a picture pixel matrix, so that each sample has one corresponding characteristic picture, putting the characteristic pictures marked as fail of the sample into a folder named fail, putting the characteristic pictures marked as success of the sample into a folder named success, and putting the m characteristic pictures into the corresponding folders according to own labels.
Further, in the present invention,
in the step b3, the model input quantity hin is the feature picture of each sample, and the model output quantity hout is a label of the feature picture, that is, a name fail or success of a folder in which the feature picture is located.
Further, in the present invention,
in the step b5, 80% of the model input amount hin and the model output amount hout of the sample are randomly extracted as a model training set, and 20% of the model input amount hin and the model output amount hout of the sample are left as a model verification set.
Further, in the present invention,
in the step b7, the online secondary commutation failure intelligent prediction model is verified by the model verification set in the step b5: inputting the model input quantity hin of the sample into the trained online secondary commutation failure intelligent prediction model, comparing the predicted model output quantity pout with the real model output quantity hout of the sample, and calculating the probability that pout and hout are consistent, namely the accuracy and the rate of missing judgment, and when the accuracy is greater than 90% and the rate of missing judgment is about 5%, determining that the online secondary commutation failure intelligent prediction model passes the verification.
Further, in the present invention,
on the basis of the steps A and B, the multi-source training data are utilized to centrally judge samples which can cause the secondary commutation failure, the intelligent agent is trained off line, and an intelligent prediction model of the transient state track of the on-line disturbed voltage is constructed, wherein the intelligent prediction model of the transient state track of the on-line disturbed voltage is respectively constructed for the bus voltage of the power grid inverter station, and u is set LV (t) any one of the grid inverter station bus line voltages, the offline training includes:
step c1: extracting data features;
and c2: selecting a model, namely selecting a regression learning method on a software platform with a machine learning function to train a prediction model;
and c3: determining a prediction factor and a prediction response of the online disturbed voltage transient trajectory intelligent prediction model;
and c4: and c, model training and cross validation, namely training a model by adopting the prediction factor and the prediction response in the step c3, and validating the accuracy of the prediction model by using a cross validation method.
Further, in the present invention, it is preferable that,
the step c1 comprises:
aiming at each sample data in the multi-source training data set, the line voltage u in the first fundamental frequency period after the fault occurs LV (t) Fourier transforming the line voltage u LV The amplitudes and phase angles of the fundamental frequency and the 2, 3 and 4 harmonics of (t) constitute the feature matrix D of each sample 1×8 ，
Wherein, U LV1 、U LV2 、U LV3 、U LV4 Are respectively the line voltage u LV (t) fundamental frequency and amplitudes of the 2, 3, 4 harmonics;
are respectively the line voltage u LV The fundamental frequency of (t) and the phases of the 2, 3, 4 harmonics.
Further, in the present invention,
in the step c3, the step of processing,
setting the moment t of one fundamental frequency period after the fault occurs 3 To the time t of 5 fundamental frequency periods after the fault is finished 2 There are k sampling instants in between, at t 3 ，t 2 ]The predictor F in a time period k*9 Including the k sampling moments and the feature matrix D of each sample extracted in step c1 1×8 Let Δ t be the sampling time interval of every two adjacent sampling instants in said k sampling instants, then
At [ t ] 3 ，t 2 ]The predicted response R over a period of time k*1 For said line voltage u LV (t) at [ t 3 ，t 2 ]The values of the k sampling instants within the time period,
further, in the present invention,
when the AC system fault occurs, in a time period t 4 ，t 2 ]The trigger angle command value alpha of which the secondary commutation failure does not occur order (t) a feasible region satisfying α order (t)≤α order_MAX (t)，
Wherein the firing angle command value alpha order (t) maximum value at time t
arccos (X) is an inverse cosine function with respect to X, cos (y) is a cosine function with respect to y, X r For the commutation reactance of a converter transformer on the inverting side, I dh Is the limit value of DC current in the low-voltage current-limiting unit, gamma min For a minimum extinction angle at which commutation failure does not occur,
U RMS (t) is the bus line voltage u of the grid inversion station at the time t ab (t)，u bc (t)，u ca (t) minimum value of effective value, i.e.
U RMS_AB (t)，U RMS_BC (t)，U RMS_CA (t) are time periods [ t ] respectively 4 ，t 2 ]The power grid inverter station bus line voltage u is obtained through prediction ab1 (t)，u bc1 (t)，u ca1 (t) an effective value of
Wherein the content of the first and second substances,
f0 is the rated frequency of the network, u ab1 (t)，u bc1 (t)，u ca1 (t) is predicted by the online disturbed voltage transient trajectory intelligent prediction model.
The direct-current commutation failure blocking and preventing control method disclosed by the invention has the advantages that the prejudgment on commutation failure meets the requirements on rapidity and effectiveness, and whether secondary commutation failure occurs in a fault period can be rapidly and accurately predicted within ms-level time; the method can quickly, conservatively and pertinently pre-judge the impending commutation failure in the system, and reserve enough time margin and safety margin for subsequent control; the method can realize the rapid prediction of the change of the short-term phase-change voltage; under the condition that the performance of the phase-locked loop is ideal and the communication delay is not counted, the secondary commutation failure during the fault (caused by the fault of the alternating current system) can be effectively avoided; the implemented control strategy can guarantee timeliness, reliability and result constrainable.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a schematic diagram illustrating a DC commutation failure blocking and prevention control method according to an embodiment of the present invention;
FIG. 2 illustrates a multi-source training data set timeline according to an embodiment of the present invention;
fig. 3 illustrates a characteristic curve of a low voltage current limiting unit according to an embodiment of the present invention;
fig. 4 shows a flow chart of a dc commutation failure blocking and prevention control method according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention provides a direct current commutation failure blocking and preventing control method based on system disturbed transient trajectory prediction, which is a method for predicting a direct current commutation transient boundary through machine learning aiming at an electric power system to generate a secondary commutation failure blocking control strategy.
Referring to fig. 1, the dc commutation failure blocking and preventing control method of the present invention includes an offline portion and an online portion. The offline part is used to build two models: an online secondary commutation failure intelligent prediction model (which can be simply referred to as a secondary commutation failure intelligent prediction model) and an online disturbed voltage transient state trajectory intelligent prediction model. The online part is used for online generation of a control strategy to block and prevent occurrence of secondary commutation failure.
(1) Off-line part
The off-line part comprises the following steps:
step a: generating a multi-source training dataset
As can be seen from fig. 1, the sources of the multi-source training data set include: the method comprises the steps of simulating samples of an electromagnetic transient state and an electromechanical-electromagnetic transient state hybrid simulation of an alternating-current and direct-current power grid, and real-time measurement data of a power grid Phasor Measurement Unit (PMU) and a power failure wave recording device (TFR).
Referring to fig. 2, m sample data included in the multi-source training data set records the simulated simulation sample and the real-time measured data before the start of the fault (t) 1 Time) to 5 fundamental frequency periods (t) after fault removal 2 Time of day), m is a positive integer greater than 1, s represents seconds, the fundamental frequency period is a period corresponding to the rated frequency of the power system (namely, one period of commutation voltage), and the rated frequency of the power system can be 50Hz or 60Hz.
Wherein the content of the first and second substances,
t start is the time of the fault occurrence;
[t 1 ，t start ]is the fundamental frequency cycle before the fault;
t end is the time of fault clearance;
[t end ，t 2 ]is 5 fundamental frequency periods after the fault is over;
[t start ，t 3 ]is the first fundamental frequency cycle after the fault occurs;
[t start ，t 4 ]is 1.5 fundamental frequency cycles after the failure occurs.
The physical quantity in each sample data in the multi-source training data set comprises: busbar voltage (line voltage for short) u of power grid inverter station ab (t)、u bc (t)、u ca (t) and a power grid inverter station arc-quenching angle gamma (t), wherein the arc-quenching angle is used only when the intelligent prediction model for the secondary commutation failure is trained, and the arc-quenching angle is not extracted in other steps, wherein u ab (t)、u bc (t)、u ca And (t) measuring voltages of an A phase and a B phase, a B phase and a C phase and an A phase in three phases of a bus of the grid inversion station respectively.
The real-time measurement data of the power grid PMU/TFR in the multi-source training data set come from the real-time measurement of the AC/DC power grid inverter station.
The simulation sample in the multi-source training data set is based on the following fault disturbance set in the electromagnetic transient state of the alternating current-direct current power grid and the electromechanical-electromagnetic transient hybrid simulation:
the type of failure: the fault of the alternating current line which may occur in an actual system such as a single-phase grounding short circuit, a two-phase short circuit, a three-phase short circuit and a two-phase grounding short circuit on the line of the alternating current system.
Location of failure occurrence: the method comprises the steps of selecting positions where line faults can occur in a simulated actual alternating current system, and generally selecting positions with different electrical distances from buses of a grid inversion station to set the faults.
Timing of the failure: setting fault angles as different angles with intervals of theta degrees from 0 degrees to 180 degrees, wherein theta is more than 0 and less than 30, and the smaller theta is, the more fault time sequences are covered;
duration of failure: setting corresponding fault duration according to the actual relay protection action time of the simulated alternating current power grid, wherein the fault duration is usually about 0.1 s;
size of the fault: and setting the fault severity from light to heavy according to the scale of the alternating current power grid.
Step b: referring to fig. 1, based on a supervised learning method, the intelligent agent is trained offline by using the multi-source training data set in step a to construct an online intelligent prediction model for secondary commutation failure. The training process mainly comprises the following steps:
step b1: data feature extraction:
for each sample data in the multi-source training data set, the following processing is carried out: for the previous fundamental frequency cycle ([ t ] before the fault occurs 1 ，t start ]) Line voltage u in the time period of (d) ab (t)、u bc (t) and u ca (t) Fourier transform to obtain an amplitude spectrum (U) of the line voltage of 1 xn ab ) 1×n 、(U bc ) 1×n 、(U ca ) 1×n And 1 xn line voltage phase spectrum
Wherein the content of the first and second substances,
(U ab ) 1×n ＝[U ab0 U ab1 U ab2 ... U abn-1 ]；
(U bc ) 1×n ＝[U bc0 U bc1 U bc2 ... U bcn-1 ]；
(U ca ) 1×n ＝[U ca0 U ca1 U ca2 ... U can-1 ]；
0 is expressed as direct current component after Fourier transform, 1 is fundamental frequency, 2 and 3 … n-1 is harmonic series, n-1 is the highest harmonic series after Fourier transform, and n-1 harmonic, U, of n-1 level are obtained by Fourier transform ab0 Is a line voltage u ab (t) amplitude of the DC component, U ab1 To U abn-1 Are respectively the component line voltage u ab (t) the magnitude of each frequency component. In the same way as above, the first and second,
is a line voltage u ab (t) the phase angle of the direct current component,
to
To form a line voltage u ab The phase that each frequency component of (t) has at the origin of time, and so on.
Using the amplitude and phase spectra described above, a 6 × n feature matrix A is created 6×n ：
Wherein, U ab1 、U bc1 、U ca1 The fundamental frequency amplitudes of the three line voltages, respectively.
For the first fundamental frequency cycle after the fault occurs (i.e., [ t ] start ，t 3 ]) Repeating the calculation process on the bus line voltage of the internal grid inversion station to create a characteristic matrix B of 6 multiplied by n 6×n ：
Splicing the two matrixes up and down to create a 12 xn characteristic matrix C 12×n ：
Finally, each sample corresponds to a data characteristic matrix C 12×n 。
Aiming at m sample data in the multi-source training data set, according to 1.5 fundamental frequency periods after the fault occurs to the fault ending period ([ t) 4 ，t 2 ]) The value of the extinction angle gamma (t) is assigned to each sample a fail or success label, when the gamma (t) of a certain sample is in [ t ] 4 ，t 2 ]If zero value appears in the sample, the sample is marked as fail, namely, the second commutation failure occurs; otherwise, the sample is marked as success, i.e., no secondary commutation failure occurs. Finally, each sample has a corresponding label.
The feature matrix C 12×n Generating a characteristic picture with the format png as a picture pixel matrix, enabling each sample to have a corresponding characteristic picture, placing the characteristic picture of the sample marked as 'fail' into a folder named fail, and marking the characteristic picture as successThe characteristic picture of the sample is put into a folder named success. Thus, the m characteristic pictures are put into the corresponding folders according to the labels of the m characteristic pictures.
Step b2: model selection
And selecting a Convolutional Neural Network (CNN) method on a software platform containing a machine learning function to train the prediction model.
Step b3: determining model input quantities and model output quantities
The model input quantity hin is the feature picture of each sample in step b1, and the model output quantity hout is the label of the feature picture, i.e. the name (fail or success) of the folder in which the picture is located.
Step b4: defining architecture, parameters and training options for CNN networks
Step b5: sample splitting
And randomly extracting 80% of the model input quantity and the model output quantity of the sample as a model training set, and remaining 20% of the model input quantity and the model output quantity of the sample as a model verification set.
Step b6: model training
And (c) training a prediction model by using the model training set in the step b5 and the CNN network built in the step b4 to obtain an online secondary commutation failure intelligent prediction model, wherein the online secondary commutation failure intelligent prediction model can be programmed by self, each parameter of the CNN network can be defined by self, and the existing CNN network of the platform can also be used.
Step b7: model validation pass
Verifying the online secondary commutation failure intelligent prediction model by using the model verification set in the step b5: and inputting the model input quantity hin of the sample into the trained online intelligent prediction model for secondary commutation failure, comparing the predicted model output quantity pout with the real model output quantity hout of the sample, and calculating the probability (namely accuracy) that pout and hout are consistent and the miss rate, wherein when the accuracy is greater than 90% and the miss rate is about 5%, the online intelligent prediction model for secondary commutation failure is determined to pass verification. The miss-judging rate is the percentage of cases which actually have secondary commutation failure but are judged not to have commutation failure. The values of the accuracy rate and the misjudgment rate are different for different alternating current power grids and different model training sets and model verification sets. And if the accuracy rate or the missing judgment rate does not reach the set target, increasing the number of samples and retraining.
Step c: on the basis of a supervised learning method, on the basis of the steps a and b, samples which can cause secondary commutation failure are judged in a concentrated mode by utilizing multi-source training data, an intelligent body is trained off line, and an online disturbed voltage transient state track intelligent prediction model is constructed. On-line disturbed voltage transient state track intelligent prediction models are respectively constructed for three line voltages of a power grid inversion station bus, and the power grid inversion station bus line voltage u ab (t) taking the training process of the prediction model as an example, the training process mainly comprises the following steps:
step c1: data feature extraction
Setting u for each sample data in multi-source training data set LV (t) is any one of the bus line voltages of the grid inversion station, and the first fundamental frequency period after the fault occurs ([ t) start ，t 3 ]) Bus line voltage u of internal grid inversion station LV (t) performing Fourier transform, and forming the amplitude and phase angle of the fundamental frequency of the line voltage and the 2, 3 and 4 harmonics into a characteristic matrix D of each sample 1×8 。
Wherein U is LV1 、U LV2 、U LV3 、U LV4 Are respectively a component line voltage u LV (t) fundamental frequency and amplitudes of the 2, 3, 4 harmonics;
are respectively the component line voltage u LV The fundamental frequency of (t) and the phases of the 2, 3, 4 harmonics.
And c2: model selection
And (3) selecting a regression learning method on a software platform containing a machine learning function to train the prediction model.
And c3: determining a predictor and a predicted response of a predictive model
Predictor F of a prediction model k*9 Including one fundamental frequency cycle after the fault occurs to 5 fundamental frequency cycles after the fault ends ([ t ] 3 ，t 2 ]) K sampling time points in the sample list and the feature matrix D of each sample extracted in step c1 1×8 Wherein k is [ t ] 3 ，t 2 ]The number of internal samples, i.e. the number of samples of the signal which are extracted from the continuous signal and constitute the discrete signal every second, the magnitude of the k value depends on the sampling frequency f set in the simulation model or on the real-time measuring instrument. Predictor F k*9 Satisfy the requirement of
Wherein, Δ t is a sampling time interval, and the value of Δ t is determined according to the simulation platform or PMU/TFR real-time measurement sampling frequency.
Predicted response R of a prediction model k*1 For line voltage u of power grid inverter station LV (t) at [ t 3 ，t 2 ]The values of k sample points within the time period.
And c4: model training and cross validation
Using predictor F in step c3 k*9 And predicted response R k*1 And training the model, and verifying the accuracy of the prediction model by using a cross-validation method.
And c5: if the accuracy of the prediction model does not meet the requirement, repeating the steps c1 to c4, training and obtaining the bus line voltage u of the power grid inverter station bc (t) and u ca (t) intelligent prediction model of transient trajectory of online disturbed voltage.
(2) On-line part
The system is started in an online part after disturbance occurs, and the online part comprises the following steps:
step d1: referring to fig. 1, the ([ t ] is obtained by real-time measurement or sampling of, for example, grid PMU/TFR 1 ，t 3 ]) Line voltage u of power grid inverter station bus in time period ab (t)、u bc (t) and u ca (t), because the sampled line voltage is mainly used for Fourier transform to extract signal characteristics, the sampling frequency can be lower, 2000 Hz-10000 Hz, preferably 5000Hz;
step d2: referring to FIG. 1, the measured value of the line voltage in step d1 is input to the 'prediction model' created in step b to obtain [ t 4 ，t 2 ]Judging whether a secondary commutation failure occurs in a time period, and entering a step d3 if the secondary commutation failure occurs;
step d3: the bus line voltage [ t ] in the step d1 is processed start ，t 3 ]Inputting the measured value in the time period into the online disturbed voltage transient state track intelligent prediction model established in the step c, and instantly predicting to obtain the bus voltage of the power grid inverter station in the future time period [ t ] 3 ，t 2 ]Internal voltage u ab1 (t)，u bc1 (t)，u ca1 (t)，
And d4: d3, utilizing the predicted bus line voltage u of the power grid inverter station in the step d3 ab1 (t)，u bc1 (t)，u ca1 (t) calculating [ t 4 ，t 2 ]Bus voltage effective value U of power grid inversion station corresponding to time RMS_AB (t)，U RMS_BC (t)，U RMS_CA (t)；
Step d5: referring to FIG. 1, according to the phase commutation failure blocking and preventive control strategy constraint conditions, the [ t ] is obtained by on-line instant solution 4 ，t 2 ]And the trigger angle instruction value in the time period can be in a feasible range and a maximum value, and is sent to an execution unit controlled by a direct current inverter of an alternating current-direct current power grid, so that control parameters are adjusted on line, and the occurrence of subsequent secondary commutation failure is avoided.
Wherein, the commutation failure blocking and preventing control strategy needs to consider the following 3 constraint conditions:
constraint 1:
due to the quantitative relationship between commutation voltage, dc current (i.e. inverter dc side current), firing angle and arc-quenching angle:
wherein i d (alpha) and i d (α + μ) is the DC current at the start and end of commutation, X r For the commutation reactance, U, of a converter transformer on the inverting side L The effective value of the bus voltage of the power grid inversion station is shown, alpha is a trigger angle, and gamma is an arc extinguishing angle.
When the system has phase change failure, the effective values of the direct current and the bus voltage of the power grid inversion station are not constant any more, and the phase change time is very short, so that i d (α + μ) approximately equal to i d (α)，U L D, taking the predicted and calculated bus line voltage u of the power grid inverter station at the time t in the step d4 ab1 (t)，u bc1 (t)，u ca1 (t) minimum value of effective value, i.e.
U RMS (t)＝min[U RMS_AB (t)，U RMS_BC (t)，U RMS_CA (t)]，
Wherein the content of the first and second substances,
f0 is the rated frequency of the power grid, specifically 50Hz or 60Hz, and T is one period of the line voltage in the power grid.
Then, a constraint equation when commutation fails is obtained:
where α is order And (t) is a trigger angle command value.
Constraint 2:
the constraint conditions of the extinction angle when no commutation failure occurs are as follows:
γ≥γ min (3)，
wherein, gamma is min The minimum extinction angle is the minimum extinction angle at which no commutation failure occurs, and the minimum extinction angle is the minimum time to satisfy the restoration blocking capability of the converter valve. If the extinction angle is smaller than the value, the converter valve does not have enough time to recover the blocking capability, and phase commutation failure can occur.
Constraint 3:
referring to fig. 3, the constraint is obtained according to the relationship between the dc voltage and the dc current in the low voltage current limiting unit (VDCOL):
wherein i d 、U d Is a direct current, I dl 、I dh 、U dl 、U dh The upper limit value and the lower limit value of the direct current (namely, the direct current side current of the inverter) and the direct constant voltage (namely, the direct current side voltage of the inverter) are respectively. Obtaining i from the constraint (2) d The value range is as follows:
I dl ≤i d ≤I dh (5)，
substituting the formulas (3) and (5) into the formula (2) to obtain the value range of the extinction angle instruction value without commutation failure:
it can be seen that at time t, α order (t) at γ = γ min ，i d ＝I dh The maximum value is then taken, which is:
referring to fig. 1, in actual work, the time for prejudging the secondary commutation failure and the time for setting the feasible domain of the trigger angle instruction value are both negligible, the transmission switching delay of the trigger angle instruction value is within 0.01 second, [ t ] t 4 ，t 2 ]Trigger angle instruction value in [ t ] in time period 3 ，t 4 ]The value in a feasible domain where secondary commutation failure does not occur can be switched within time, and the requirement of preventing and controlling timeliness of the secondary commutation failure during the failure period is met.
Referring to fig. 4, fig. 4 shows an embodiment of basic steps executed when the dc commutation failure (which may be simply referred to as commutation failure) blocking and preventing control method of the present invention is adopted, including:
when detecting that the AC system fault occurs, recording the line voltage u between the fundamental frequency period before the fault occurs and the fundamental frequency period after the fault occurs ab (t)、u bc (t)、u ca (t)；
Data extraction for the first time: extracting the line voltage u from the fundamental frequency period before the fault occurs to the fundamental frequency period after the fault occurs ab (t)、u bc (t)、u ca (t)；
Inputting data obtained by first data extraction into an online secondary commutation failure intelligent prediction model, judging whether secondary commutation failure occurs, and if the secondary commutation failure does not occur, ending the execution of the method;
and if the second commutation failure occurs, performing second data extraction: extracting the line voltage u from the fault to the period of the fundamental frequency after the fault ab (t)、u bc (t)、u ca (t)；
Inputting data obtained by the second data extraction into an online disturbed voltage transient state trajectory intelligent prediction model:
calculating to obtain a line voltage effective value by an online disturbed voltage transient state track intelligent prediction model;
and calculating based on the constraint of the line voltage effective value and the trigger angle instruction value to obtain a condition, namely a maximum value and a feasible region of the trigger angle instruction value, which are met by the trigger angle instruction value without commutation failure under the symmetric fault and the asymmetric fault so as to block and prevent the occurrence of secondary commutation failure, wherein the trigger angle instruction value is input into an inverter control unit, and the value of the trigger angle is adjusted in time so as to avoid the secondary commutation failure.
The embodiment can show that: the method solves the problems that the prior commutation failure prejudging method only considers the disturbance severity degree and neglects the influence of the characteristics of a direct current control system on the commutation failure state, the control parameter design is lack of design universality depending on experience, and the starting response speed is influenced by a characteristic extraction method and the disturbance characteristics, and the judging performance is different under different transient conditions.
The invention also solves the following problems of the current existing control strategy:
1) Only the commutation physical process is controlled qualitatively, so that the control result is sometimes conservative and sometimes aggressive, the commutation feasible domain boundary cannot be clearly defined accurately according to different transient conditions, and the control economy is more difficult to further improve on the basis;
2) The current control optimization is mainly based on the adjustment of parameters and structures by existing controls such as constant current, minimum extinction angle and the like, the original control characteristics of direct current can be changed, and the adaptability is insufficient under different transient conditions and operating conditions;
3) The parameter design mode lacks universality;
4) Under the existing improvement scheme aiming at a single control strategy, the control feasible region is limited, and the phase commutation failure suppression effect cannot be ensured under the serious fault.
The establishment of the commutation control strategy needs to adopt the combination of various control strategies according to the physical mechanism of the commutation process to realize accurate and economic control on commutation failure. The invention combines the constant current control of the rectifying side and the trigger angle control of the inverting side which are common in the disturbance period, and fully excavates the control feasible region of the trigger angle of the inverting side. In the calculation of the trigger angle, the commutation voltage value is a real-time predicted value in disturbance, so that the trigger angle calculation is more accurate.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (11)
1. A direct current commutation failure blocking and prevention control method is characterized by comprising the following steps:
uninterrupted acquisition of bus line voltage u of power grid inversion station ab (t)、u bc (t)、u ca (t) when detecting the occurrence of an AC system fault, extracting the time t before said fault occurred 1 To the time t after the fault occurs 3 The bus line voltage u of the grid inverter station ab (t)、u bc (t)、u ca (t), t represents time;
data extraction for the first time: extracting the time t before the fault occurs 1 To the time t after the fault occurs 3 The bus line voltage u of the power grid inversion station ab (t)、u bc (t)、u ca (t)；
Inputting the data obtained by the first data extraction into an online secondary commutation failure intelligent prediction model, judging whether secondary commutation failure occurs, and if the secondary commutation failure does not occur, ending the execution of the direct current commutation failure blocking and preventing control method; if the secondary commutation failure is judged to occur, performing secondary data extraction: extracting the fault occurrence time t start To the time t after the fault occurs 3 The bus line voltage u of the power grid inversion station ab (t)、u bc (t)、u ca (t)；
Inputting the data obtained by the second data extraction into an online disturbed voltage transient state track intelligent prediction model;
predicting the bus voltage of the power grid inverter station at [ t ] by the online disturbed voltage transient trajectory intelligent prediction model 3 ，t 2 ]Is calculated to obtain [ t ] 4 ，t 2 ]The effective value of the bus voltage of the power grid inversion station is obtained;
based on the bus line voltage effective value of the power grid inversion station and a trigger angle instruction value constraint formula, calculating to obtain [ t 4 ，t 2 ]The condition satisfied by the trigger angle command value for which no commutation failure occurs is satisfied to block and prevent the occurrence of the secondary commutation failure,
wherein, let t end If it is the fault clearing time, the time t 1 、t 2 、t 3 And the time t of the fault occurrence start And a fault clearing time t end The relationship between (A) and (B) is:
[t 1 ，t start ]is the fundamental frequency cycle prior to the occurrence of the fault;
[t end ，t 2 ]is 5 fundamental frequency periods after the fault is over;
[t start ，t 3 ]is the first fundamental period after the occurrence of the fault;
t 4 the time of 1.5 fundamental frequency periods after the fault occurs;
the online secondary commutation failure intelligent prediction model is obtained by the following method, and comprises the following steps:
A. generating a multi-source training data set;
B. on the basis of a supervised learning method, performing offline training on an agent by using the multi-source training data set to construct the online secondary commutation failure intelligent prediction model;
wherein the step B comprises the step B1: data feature extraction, specifically comprising:
for each sample data in the multi-source training data set, the following processing is carried out:
the bus line voltage u of the grid inverter station in the time period of the fundamental frequency cycle before the fault occurs ab (t)、u bc (t) and u ca (t) carrying outThe Fourier transform yields a 1 n amplitude spectrum (U) of the line voltage ab ) 1×n 、(U bc ) 1×n 、(U ca ) 1×n And 1 xn line voltage phase spectrum
Wherein the content of the first and second substances,
(U ab ) 1×n ＝[U ab0 U ab1 U ab2 ...U abn-1 ]；
(U bc ) 1×n ＝[U bc0 U bc1 U bc2 ...U bcn-1 ]；
(U ca ) 1×n ＝[U ca0 U ca1 U ca2 ...U can-1 ]；
0 is expressed as a direct current component after the Fourier transform, 1 is a fundamental frequency, 2 and 3 … n-1 is a harmonic series, n-1 is the highest harmonic series after the Fourier transform, and n-1 harmonic is obtained by the Fourier transform; u shape ab0 For said line voltage u ab (t) amplitude of the direct current component, U ab1 To U abn-1 Respectively constitute the line voltage u ab (t) the magnitude of each frequency component,
for said line voltage u ab (t) the phase angle of the direct current component,
to
To form the line voltage u ab (t) the phase of each frequency component at the time origin; u shape bc0 For said line voltage u bc (t) amplitude of the direct current component, U bc1 To U bcn-1 Respectively constitute the line voltage u bc (t) the magnitude of each frequency component,
for said line voltage u bc (t) the phase angle of the direct current component,
to
To form the line voltage u bc (t) the phase of each frequency component at the time origin; u shape ca0 For said line voltage u ca (t) amplitude of the direct current component, U ca1 To U can-1 Respectively forming the line voltage u ca (t) the magnitude of each frequency component,
for said line voltage u ca (t) the phase angle of the direct current component,
to
To form said line voltage u ca (t) the phase of each frequency component at the time origin;
creating a 6 x n feature matrix A using the amplitude and phase spectra 6×n ：
And U is ab1 、U bc1 、U ca1 Respectively the fundamental frequency amplitude of the bus bar voltage of the power grid inverter station,
repeating the calculation process on the bus line voltage of the power grid inverter station in the first fundamental frequency period after the fault occurs, and creating a characteristic matrix B of 6 x n 6×n ，
The feature matrix A is divided into 6×n And the feature matrix B 6×n Splicing, creating a 12 xn data feature matrix C 12×n ：
Obtaining a corresponding data feature matrix C for each sample 12×n ；
Wherein A is 11 Representing a feature matrix A 6×n First data of the first row, A 12 Representation feature matrix A 6×n The first line of the first row and the second line of the second row 1n Representing a feature matrix A 6×n The first row n data, a 6n Representing a feature matrix A 6×n The nth data of the sixth line; b is 11 Represent a feature matrix B 6×n First data of the first row, B 12 Represent a feature matrix B 6×n The first row of the second row of the first row, B 1n Representing a feature matrix B 6×n The first row n data, B 6n Represent a feature matrix B 6×n The nth data of the sixth line;
aiming at m sample data in the multi-source training data set, distributing fail or success labels to each sample according to the value of the extinction angle gamma (t) of the power grid inverter station from 1.5 fundamental frequency periods after the fault occurs to 5 fundamental frequency periods after the fault is finished, wherein the distribution rule is that when the gamma (t) of a certain sample is in the range of) At [ t ] 4 ，t 2 ]If zero value is found in the samples, the samples are marked as fail, namely the secondary commutation failure occurs; otherwise, the sample is marked as success, namely the secondary commutation failure does not occur, and finally each sample has a corresponding label t 4 For the time instant t of 1.5 fundamental frequency cycles after the occurrence of the fault 2 For the 5 th fundamental frequency cycle instant after the end of the fault,
the data feature matrix C 12×n Generating characteristic pictures as a picture pixel matrix, enabling each sample to have one corresponding characteristic picture, placing the characteristic pictures marked as fail of the samples into a folder named fail, placing the characteristic pictures marked as success of the samples into a folder named success, and enabling m characteristic pictures to be placed into the corresponding folders according to labels of the m characteristic pictures;
on the basis of the steps A and B, judging samples which can cause the secondary commutation failure in a centralized manner by using the multi-source training data, performing off-line training on the intelligent agent, and constructing an intelligent prediction model of the transient trajectory of the online disturbed voltage;
the off-line training comprises the steps of c1: data feature extraction, which specifically comprises:
aiming at each sample data in the multi-source training data set, carrying out fault detection on the line voltage u in the first fundamental frequency period after the fault occurs LV (t) Fourier transforming the line voltage u LV The amplitude and phase angle of the fundamental frequency and the 2, 3, 4 harmonics of (t) constitute the feature matrix D of each sample 1×8 ，
Wherein, U LV1 、U LV2 、U LV3 、U LV4 Are respectively the line voltage u LV The fundamental frequency of (t) and the amplitudes of the 2, 3, 4 harmonics;
are respectively the line voltage u LV The fundamental frequency of (t) and the phases of the 2, 3, 4 harmonics.
2. The method according to claim 1, wherein the DC commutation failure blocking and preventing control method,
the sources of the multi-source training data set include: the method comprises the following steps of (1) carrying out mixed simulation on an electromagnetic transient state and an electromechanical-electromagnetic transient state of an alternating-current and direct-current power grid, and real-time measurement data obtained by carrying out real-time measurement on an alternating-current and direct-current power grid inverter station by a power grid phasor measurement device and a power fault recording device;
the multi-source training data set comprises m samples, and the data of each sample records the time t before the fault occurs in the simulation sample and the real-time measurement data 1 To the time t after the fault occurs 2 The number of samples m is a positive integer greater than 1,
the physical quantities in the data of each sample include: bus line voltage u of power grid inverter station ab (t)、u bc (t)、u ca (t) and a grid inverter station arc-quenching angle gamma (t);
the simulation sample sets the following faults in the electromagnetic transient and electromechanical-electromagnetic transient hybrid simulation of the alternating current-direct current power grid:
the types of faults include: a single-phase grounding short circuit, a two-phase short circuit, a three-phase short circuit and a two-phase grounding short circuit on the AC system line;
the position of the fault occurrence: selecting positions with different electrical distances from a bus of the power grid inversion station to set faults;
timing of the fault: setting fault angles to be different angles with intervals of theta degrees from 0 degrees to 180 degrees; wherein theta is more than 0 and less than 30;
the fault duration is: setting corresponding fault duration according to the actual relay protection action time of the simulated alternating current power grid;
the size of the fault: setting faults of different severity according to the scale of the simulated alternating current power grid.
3. The method according to claim 2, wherein the DC commutation failure blocking and preventing control method,
carrying out the uninterrupted acquisition of the bus line voltage u of the power grid inverter station ab (t)、u bc (t)、u ca The acquisition frequency at (t) is 2000 Hz-10000 Hz;
the number m of the samples satisfies that m is more than or equal to 100 and less than or equal to 10000.
4. The DC commutation failure blocking and preventing control method according to claim 3,
the acquisition frequency is 5000Hz;
the number of samples m =3000;
the fault duration is 0.1s, s for seconds.
5. The method according to claim 2, wherein the DC commutation failure blocking and preventing control method,
the step B further comprises the following steps:
step b2: selecting a model, namely selecting a convolutional neural network method on a software platform with a machine learning function to train a prediction model;
step b3: determining model input quantity hin and model output quantity hout;
step b4: defining the architecture, parameters and training options of the convolutional neural network;
and b5: splitting a sample to obtain a model training set and a model verification set;
step b6: model training, namely training a prediction model by adopting the convolutional neural network and a model training set to obtain the online secondary commutation failure intelligent prediction model;
step b7: the model verification is passed.
6. The DC commutation failure blocking and preventing control method according to claim 5,
in the step b3, the model input quantity hin is the feature picture of each sample, and the model output quantity hout is a label of the feature picture, that is, a name fail or success of a folder in which the feature picture is located.
7. The DC commutation failure blocking and preventing control method according to claim 5,
in the step b5, 80% of the model input amount hin and the model output amount hout of the sample are randomly extracted as a model training set, and 20% of the model input amount hin and the model output amount hout of the sample are left as a model verification set.
8. The DC commutation failure blocking and preventing control method according to claim 5,
in the step b7, the online secondary commutation failure intelligent prediction model is verified by the model verification set in the step b5: inputting the model input quantity hin of the sample into the trained online secondary commutation failure intelligent prediction model, comparing the predicted model output quantity pout with the real model output quantity hout of the sample, and calculating the probability that pout and hout are consistent, namely the accuracy and the rate of missing judgment, and when the accuracy is greater than 90% and the rate of missing judgment is about 5%, determining that the online secondary commutation failure intelligent prediction model passes the verification.
9. The method according to any one of claims 2-8, wherein the DC commutation failure blocking and preventing control method,
on the basis of the steps A and B, the multi-source training data are utilized to centrally judge samples which can cause the secondary commutation failure, the intelligent agent is trained off line, and an intelligent prediction model of the transient state track of the on-line disturbed voltage is constructed, wherein the intelligent prediction model of the transient state track of the on-line disturbed voltage is respectively constructed for the bus voltage of the power grid inverter station, and u is set LV (t) any one of the grid inverter station busbar voltages, then the offline training further comprises:
step c2: selecting a model, namely selecting a regression learning method on a software platform with a machine learning function to train a prediction model;
and c3: determining a prediction factor and a prediction response of the online disturbed voltage transient trajectory intelligent prediction model;
and c4: and c, model training and cross validation, namely training a model by adopting the prediction factor and the prediction response in the step c3, and validating the accuracy of the prediction model by using a cross validation method.
10. The method according to claim 9, wherein the DC commutation failure blocking and preventing control method,
in the step c3, the step of processing,
setting the moment t of one fundamental frequency period after the fault occurs 3 To the time t of 5 fundamental frequency periods after the fault is finished 2 With a total of k sampling instants, at t 3 ，t 2 ]The predictor F in a time period k*9 Including the k sampling moments and the feature matrix D of each sample extracted in step c1 1×8 Let Δ t be the sampling time interval of every two adjacent sampling instants in said k sampling instants, then
At [ t ] 3 ，t 2 ]The predicted response R over a period of time k*1 For said line voltage u LV (t) at [ t 3 ，t 2 ]The values of the k sampling instants within the time period,
11. the method for blocking and preventing DC commutation failure according to any one of claims 1-8 and 10,
when the AC system fault occurs, in a time period t 4 ，t 2 ]The trigger angle command value alpha of which the secondary commutation failure does not occur order (t) a feasible region satisfying α order (t)≤α order _ MAX (t)，
Wherein the firing angle command value alpha order (t) maximum value at time t
arccos (X) is an inverse cosine function with respect to X, cos (y) is a cosine function with respect to y, X r For the commutation reactance of a converter transformer on the inverting side, I dh Is the limit value of DC current in the low-voltage current-limiting unit, gamma min For a minimum extinction angle at which commutation failure does not occur,
U RMS (t) bus line voltage u of grid inverter station at time t ab (t)，u bc (t)，u ca (t) minimum value of effective value, i.e.
U RMS (t)＝min[U RMS_AB (t)，U RMS_BC (t)，U RMS_CA (t)]，
U RMS_AB (t)，U RMS_BC (t)，U RMS_CA (t) are time periods [ t ] respectively 4 ，t 2 ]The power grid inverter station bus line voltage u is obtained through prediction ab1 (t)，u bc1 (t)，u ca1 (t) a significant value of
Wherein, the first and the second end of the pipe are connected with each other,
f0 is the rated frequency of the network, u ab1 (t)，u bc1 (t)，u ca1 (t) predicting by the online disturbed voltage transient trajectory intelligent prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010997688.0A CN112234641B (en) | 2020-09-21 | 2020-09-21 | Direct-current commutation failure blocking and preventing control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010997688.0A CN112234641B (en) | 2020-09-21 | 2020-09-21 | Direct-current commutation failure blocking and preventing control method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112234641A CN112234641A (en) | 2021-01-15 |
| CN112234641B true CN112234641B (en) | 2022-12-23 |
Family
ID=74108466
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010997688.0A Active CN112234641B (en) | 2020-09-21 | 2020-09-21 | Direct-current commutation failure blocking and preventing control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112234641B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112993994B (en) * | 2021-03-31 | 2022-07-08 | 南方电网科学研究院有限责任公司 | Control method and device for first commutation failure of high-voltage direct current in alternating-current fault |
| CN113258670B (en) * | 2021-04-25 | 2023-12-19 | 国网江苏省电力有限公司营销服务中心 | Power grid flying line state discrimination method based on semi-supervised learning |
| CN113410862B (en) * | 2021-06-17 | 2022-04-26 | 国网河南省电力公司电力科学研究院 | Commutation failure prevention control method considering multi-feed-in direct current coupling effect |
| CN116961072B (en) * | 2023-07-12 | 2024-06-18 | 山东大学 | Multi-DC chain commutation failure recognition method and system based on space-time convolution network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109088425B (en) * | 2018-09-17 | 2022-04-01 | 国网上海市电力公司 | Commutation failure prevention method based on alternating current bus voltage disturbance quantity |
| CN109873443B (en) * | 2019-04-12 | 2020-06-30 | 重庆大学 | Method for predicting direct-current continuous commutation failure under power grid fault based on critical voltage |
| CN111327082B (en) * | 2020-03-09 | 2021-09-03 | 国网山东省电力公司济南供电公司 | Commutation failure prevention control optimization method and system based on SVM-PSO |
-
2020
- 2020-09-21 CN CN202010997688.0A patent/CN112234641B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN112234641A (en) | 2021-01-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112234641B (en) | Direct-current commutation failure blocking and preventing control method | |
| Vaiman et al. | Risk assessment of cascading outages: Methodologies and challenges | |
| EP2346132B1 (en) | Fast three-phase reclosing method in shunt reactor compensated transmission lines | |
| CN103986385B (en) | Method for restraining no-load closing magnetizing inrush current of transformer | |
| Jang et al. | Development of a logical rule-based islanding detection method for distributed resources | |
| Mestas et al. | Implementation and performance evaluation of a reclosing method for shunt reactor-compensated transmission lines | |
| Wei et al. | Variable extinction angle control strategy based on virtual resistance to mitigate commutation failures in HVDC system | |
| CN109742786A (en) | The electrically continuous preventive control for commutation failure method of high-voltage dc transmission based on fuzzy control | |
| CN111146784B (en) | Continuous commutation failure suppression method and system based on dynamic current deviation control | |
| Sun et al. | Optimal allocation of smart substations in a distribution system considering interruption costs of customers | |
| Peralta et al. | Automatic restoration of large-scale distribution networks with distributed generators, voltage control devices and heating loads | |
| CN105140973A (en) | High-voltage direct-current power transmission system reliability assessment method considering alternating-current side fault | |
| CN106253321B (en) | A kind of optimization method of DC control and protection system commutation failure PREDICTIVE CONTROL | |
| Robak et al. | Transient stability improvement by generator tripping and real-time instability prediction based on local measurements | |
| CN111769585B (en) | Commutation failure prediction method and device based on energy accumulation characteristics of inverter | |
| Kumar et al. | An adaptive fuzzy based relay for protection of distribution networks | |
| Villarroel-Gutiérrez et al. | A novel methodology for dynamic voltage support with adaptive schemes in photovoltaic generators | |
| Santoso et al. | Power quality data analytics: Tracking, interpreting, and predicting performance | |
| Ouyang et al. | An improved prediction method of subsequent commutation failure of an LCC-HVDC considering sequential control response | |
| Cannas et al. | Neural networks for power system condition monitoring and protection | |
| Nguyen et al. | Safety-assured collaborative load management in smart grids | |
| Ma et al. | An adaptive DC line fault recovery strategy for LCC-HVDC system based on voltage gradient | |
| CN112701676B (en) | Arc extinguishing angle prediction and commutation failure suppression method | |
| Song et al. | Mechanism Analysis and Prevention Methods of Commutation Failure in LCC-HVDC Transmission System | |
| CN115117915B (en) | Comprehensive commutation failure risk assessment method for hierarchical access extra-high voltage direct current transmission system |
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 |