CN111262238B - Machine learning-based method for predicting short-circuit current of power distribution network containing IIDG - Google Patents
Machine learning-based method for predicting short-circuit current of power distribution network containing IIDG Download PDFInfo
- Publication number
- CN111262238B CN111262238B CN202010089399.0A CN202010089399A CN111262238B CN 111262238 B CN111262238 B CN 111262238B CN 202010089399 A CN202010089399 A CN 202010089399A CN 111262238 B CN111262238 B CN 111262238B
- Authority
- CN
- China
- Prior art keywords
- iidg
- short
- circuit current
- sample
- fault
- 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
- 238000010801 machine learning Methods 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 37
- 238000012549 training Methods 0.000 claims abstract description 40
- 238000004364 calculation method Methods 0.000 claims abstract description 25
- 238000004088 simulation Methods 0.000 claims abstract description 12
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 9
- 238000012360 testing method Methods 0.000 claims description 20
- 238000007781 pre-processing Methods 0.000 claims description 9
- 239000000203 mixture Substances 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 4
- 238000002790 cross-validation Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 4
- 230000016507 interphase Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000011160 research Methods 0.000 abstract description 6
- 230000035699 permeability Effects 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 5
- 238000011161 development Methods 0.000 description 3
- 238000005259 measurement Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000007637 random forest analysis Methods 0.000 description 2
- 241001272567 Hominoidea Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000035515 penetration Effects 0.000 description 1
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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02H—EMERGENCY PROTECTIVE CIRCUIT ARRANGEMENTS
- H02H1/00—Details of emergency protective circuit arrangements
- H02H1/0007—Details of emergency protective circuit arrangements concerning the detecting means
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02H—EMERGENCY PROTECTIVE CIRCUIT ARRANGEMENTS
- H02H1/00—Details of emergency protective circuit arrangements
- H02H1/0092—Details of emergency protective circuit arrangements concerning the data processing means, e.g. expert systems, neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a machine learning-based method for predicting short-circuit current of an IIDG-containing power distribution network. At present, a physical modeling method is mainly adopted for calculating the short-circuit current of a power distribution network containing an inverter type distributed power supply (IIDG), and the application requirement is difficult to meet under the high-permeability condition of the IIDG. According to the invention, by analyzing the characteristics of the IIDG-containing power distribution network related to the short-circuit current, a sample characteristic combination mode reflecting the short-circuit current is provided, and the short-circuit current of the power distribution network when the IIDG is not accessed is analyzedI f As a key feature. A simulation model built on MATLAB/Simulink is operated to automatically accumulate sample sets of 10 fault types, and an XGboost algorithm in machine learning is used for model training. The method has better applicability in a larger-scale distribution network with high IIDG permeability, ensures the rapidity and the accuracy of short-circuit current calculation, and provides support for research in the directions of setting and the like of a relay protection device of the distribution network.
Description
Technical Field
The invention belongs to the field of electric power systems, and particularly relates to a machine learning-based method for predicting short-circuit current of an IIDG-containing power distribution network.
Background
With the continuous development of global economy and society, the demand for energy is increasing day by day. Fossil energy is gradually exhausted, and renewable energy sources such as solar energy, wind energy and the like have the advantages of large reserves, small environmental pollution, capability of improving energy structures and the like, so that more and more attention is paid. Distributed power supplies are rapidly evolving in this context, with more and more distributed power supplies being incorporated into the distribution grid by means of power electronic inverters, such power supplies being referred to as inverter-based distributed generators (IIDG). The IIDG has strong nonlinear characteristics, and when a power distribution network is short-circuited, the output short-circuit current is greatly different from other power supplies and generally does not exceed 2 times of the rated current. Therefore, the traditional symmetric component method is still adopted for calculating the short-circuit current in many applications, namely the IIDG is supposed to quit running in case of fault, and the error of the short-circuit current calculation is really small under the condition that the permeability of the IIDG of the power distribution network is not high, but the error of the traditional method is more and more difficult to meet the application requirement along with the improvement of the permeability of the IIDG, and the development requirement of new calculation software is more and more urgent. Therefore, the research on the short-circuit current calculation of the IIDG high-permeability power distribution network has important theoretical and application significance.
The current research mainly uses a physical modeling method, firstly, an equivalent model of the IIDG is established when a fault occurs, and then the short-circuit current is solved. However, the IIDG equivalent model is very complex, and the model is often simplified to some extent according to an application scenario, so that the applicability and accuracy of the model are limited. Meanwhile, because the output current of the IIDG equivalent model is influenced by the factors such as the voltage of a grid-connected point, and the like, an iterative algorithm is required to be continuously corrected when the short-circuit current is solved until the precision requirement is met, and therefore, the calculation efficiency is low. With the improvement of IIDG permeability and the complexity of a control strategy, the defects of the physical modeling method in the aspects of calculation speed, accuracy, universality, development difficulty and the like become more and more obvious. Therefore, how to increase the speed and versatility of the short-circuit current calculation deserves further investigation.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects in the prior art and provide a machine learning-based method for predicting the short-circuit current of the power distribution network containing the IIDG. According to the method, corresponding sample combinations are provided by analyzing the characteristics of the power distribution network related to the short-circuit current, and MATLAB/Simulink is utilized to perform power grid modeling and simulation calculation, so that an original sample set is automatically accumulated. The short-circuit current calculation can be accurately realized by adopting an off-line training machine learning model and an on-line application mode, and the short-circuit calculation speed is greatly improved.
The main technical concept of the invention is as follows: taking the fault type as a sample type; using the same operation mode to measure the short-circuit current I flowing through the measuring point when the same fault occurs but the IIDG is not connectedfAs key characteristics, the characteristic is formed by reflecting other steady state characteristics and fault characteristics of the power distribution networkSample characteristics; measuring the short-circuit current I flowing through the point after the IIDG is connectedf_DGAs a sample label; randomly generating an operation mode and a fault condition, realizing power grid modeling and simulation calculation by utilizing MATLAB/Simulink, and automatically accumulating a sample set; and performing machine learning modeling by using an XGboost algorithm in machine learning, and selecting a corresponding model according to the fault type to calculate the short-circuit current.
The invention adopts the following specific steps:
step 1): determining sample composition for training;
the daily requirements for considering the short circuit current calculation are: when different types of short-circuit faults occur at different positions in the power distribution network, the short-circuit current flowing through a certain device or relay protection installation part is required to be calculated so as to judge whether the device is safe or whether the relay protection device can act correctly. Therefore, the invention aims at the problems that: in the power distribution network containing the IIDG, when a certain type of fault occurs at any position, the short-circuit current flowing through a specified measuring point is calculated and used as a data driving model.
First, for 10 fault types, the present invention proposes to build 10 machine learning models for prediction, so there are relatively 10 different samples. f _ type is a fault type and is used to indicate different sample types.
Secondly, the invention provides a method for measuring the short-circuit current I flowing through a measuring point when the distribution network has the same fault but is not connected to the IIDG under the same operation modefAs one of the sample characteristics. Except that IfBesides, other characteristics reflect other steady-state characteristics and fault characteristics of the power distribution network, including the situation alpha of the IIDG input into the power distribution networkjIIDG input capacity SDGjLine _ cut of the cut line, f _ line of the fault line, and f _ loc of the fault position. Measuring point flowing short-circuit current I after IIDG is connectedf_DGIs a sample label. The characteristics and the sample labels jointly form a sample of a short-circuit current calculation model under the corresponding fault type of the power distribution network.
Step 2): randomly generating an operation mode, and accumulating a sample set;
in order to make the obtained samples more diversified, the method is applied to a system comprising c rotary power supplies, s IIDG nodes and l load nodesM basic operation modes (M is more than or equal to 1) are considered for the power distribution network, wherein each basic operation mode can be used for setting basic configuration parameters of a rotary power supply, an IIDG and a load according to requirements. Meanwhile, considering the input operation condition of the IIDG, a vector α ═ α is randomly generated for each operation mode1 α2… αs]In which α isj0 or 1(j ═ 1,2, …, s) respectively indicates that the jth DG is not to be thrown into the system. Each mode of operation randomly generates 1 line as an open line _ cut, taking into account the topology change of the network. In addition, fault conditions are randomly generated, including a fault line f _ line and a fault position f _ loc.
When generating a sample, firstly, the I removal is obtained through the random and automatic combination of the operation mode and the fault conditionfSteady state characteristics and fault characteristics of other power distribution networks, and short-circuit current I flowing through the measuring point obtained by simulation after the IIDG is connectedf_DGAnd (4) a label. Then, under the condition of each simulated operation mode and fault, respectively obtaining the short-circuit current characteristics I flowing through the measuring point under the condition of not connecting any IIDG for 10 fault types through simulation calculationf. Thus, a sample set of 10 machine learning models can be accumulated.
Step 3): preprocessing the obtained samples to obtain a training sample set, and finishing machine learning model training;
by preprocessing the sample, such as one-hot encoding (one-hot encoding) for specific features, a sample set more suitable for machine learning model training is obtained. And training by adopting an XGboost algorithm in machine learning and using a training sample set, and selecting a proper hyper-parameter for training through cross validation to obtain a final machine learning model.
Step 4): calculating the short-circuit current by using a trained machine learning model;
when the method is applied, a corresponding mature training model is selected according to the fault type of the sample, then the sample meeting the requirements of the model is obtained, and finally the model is called to calculate the short-circuit current to obtain the final short-circuit current calculation result.
The invention has the beneficial effects that: the invention provides a machine learning-based short-circuit current calculation method for an IIDG-containing power distribution network in consideration of the characteristic of the short-circuit current of the IIDG-containing power distribution network, and can be used for calculating the short-circuit current flowing through a designated measurement point when any point in the power distribution network fails. The method does not need complex physical modeling analysis, improves the speed and the universality of short-circuit current calculation, and provides support for research in the directions of power distribution network relay protection and the like.
Drawings
Fig. 1 is an overall framework of a machine learning-based prediction method for short-circuit current of a power distribution network with an IIDG.
Fig. 2 is a flowchart of a machine learning-based method for predicting short-circuit current of a power distribution network including an IIDG.
Fig. 3 is a wiring diagram of an IEEE34 node test network.
Fig. 4 is a wiring diagram of an IEEE13 node test network.
Fig. 5 is a wiring diagram of an IEEE69 node test network.
Detailed Description
The invention is further described below with reference to the accompanying drawings, comprising the steps of:
considering the smaller feature quantity and the higher feature quality, a sample feature composition mode aiming at the problem is provided. In the off-line training process, data are obtained through simulation, and a sample set is established. And then, acquiring machine learning models of different fault types offline by using the training sample set. In online applications, data is first collected for the desired sample characteristics. Then according to the fault type, a corresponding model is selected to directly predict the short-circuit current, and the overall framework is shown in figure 1.
As shown in fig. 2, the present invention comprises the steps of:
step 1): determining sample composition for training;
the daily requirements for considering the short circuit current calculation are: when different types of short-circuit faults occur at different positions in the power distribution network, the short-circuit current flowing through a certain device or relay protection installation part is required to be calculated so as to judge whether the device is safe or whether the relay protection device can act correctly. Therefore, the invention aims at the problems that: in the power distribution network containing the IIDG, when a certain type of fault occurs at any position, the short-circuit current flowing through a specified measuring point is calculated and used as a data driving model. In contrast, the following characteristic composition modes were determined by analyzing the characteristics of the power distribution network including the IIDG, as shown in table 1.
First, for 10 fault types, the present invention proposes to build 10 machine learning models for prediction, so there are relatively 10 different samples. f _ type is a fault type and is used to indicate different sample types.
Secondly, the distribution network has more characteristic quantity and most of the distribution network has lower quality, and in order to improve the characteristic quality of the sample and reduce the characteristic quantity, the invention provides a method for measuring the short-circuit current I flowing through a measuring point when the distribution network has the same fault but is not connected with the IIDG in the same operation modefAs one of the sample characteristics. Due to IfThe short-circuit current is calculated by a power distribution network without IIDG through a fault analysis physical model, so that the reliability and the interpretability of a machine learning model can be improved; and because the power distribution network information such as the operation mode, the fault characteristics and the like is hidden, the number of characteristics can be greatly reduced.
In the feature, except IfBesides the key characteristics, the other characteristics reflect other steady-state characteristics and fault characteristics of the power distribution network, and the method specifically comprises the following steps: situation of IIDG investing in distribution network αjIIDG input capacity SDGjIs a characteristic reflecting the influence of IIDG on the short-circuit current, alphaj0 means that the jth DG is not connected to the grid, α j1 means that the jth DG is put into the grid, SDGjIs the capacity of the jth DG into the distribution network; the line _ cut of the cut line is a characteristic representing the change of the topological structure of the power distribution network; the fault line f _ line and the fault position f _ loc reflect fault information. Wherein the value range of f _ loc is [0,1 ]]0 and 1 indicate that the fault occurred at the line head end and terminal location, respectively, and the other values are percentages from the head end. Measuring point flowing short-circuit current I after IIDG is connectedf_DGIs a sample label. The characteristics and the sample labels jointly form a sample of a short-circuit current calculation model under the corresponding fault type of the power distribution network.
TABLE 1 sample compositions
Step 2): randomly generating an operation mode, and accumulating a sample set;
in order to make the obtained samples more diversified, M basic operation modes (M is more than or equal to 1) are considered for a power distribution network comprising c rotary power supplies, s IIDG nodes and l load nodes, wherein the rotary power supplies, the IIDGs and the load basic configuration parameters can be set according to requirements in each basic operation mode. Tau isi(i=1,2,…,c)、βj(j ═ 1,2, …, s) and ρ Pk、ρQk(k ═ 1,2, …, l) for the corresponding rotary power supply, IIDG and load [ -0.2, respectively]Random quantities independently generated in the range can be superposed on the basic value to generate different equivalent impedances, IIDG output forces and load requirements of the rotary power supply, so that different operation modes of the system can be obtained.
Meanwhile, considering the input operation condition of the IIDG, a vector α ═ α is randomly generated for each operation mode1 α2… αs]In which α isj0 or 1(j ═ 1,2, …, s) respectively indicates that the jth DG is not to be thrown into the system. Considering the topology change of the network, from the point of view of the N-1 principle, each operation mode randomly generates 1 line from the line set as the line _ cut which is disconnected. The line set includes all lines in the distribution network and is represented by a line number, wherein the number 0 represents that no line is cut off, i.e. a complete network structure. In addition, fault conditions are randomly generated, including a fault line f _ line and a fault position f _ loc. Wherein, the fault line is also selected from the line set, but does not contain the number 0; and the fault type f _ type automatically sets 10 combinations of four fault types, namely single-phase grounding, two-phase interphase and three-phase short circuit, in three ABC phases in turn, and the combinations are respectively represented by the numbers 1-10.
When generating a sample, firstly, the I removal is obtained through the random and automatic combination of the operation mode and the fault conditionfOther arrangements thanThe steady state characteristic and the fault characteristic of the power grid and the short-circuit current I flowing through the measuring point are obtained through simulation after the IIDG is connectedf_DGAnd (4) a label. Then, under the condition of each simulated operation mode and fault, respectively obtaining the short-circuit current characteristics I flowing through the measuring point under the condition of not connecting any IIDG for 10 fault types through simulation calculationf. Thus, a sample set of 10 machine learning models can be accumulated.
Step 3): preprocessing the obtained samples to obtain a training sample set, and finishing machine learning model training;
by preprocessing the sample, such as one-hot encoding (one-hot encoding) for specific features, a sample set more suitable for machine learning model training is obtained. And dividing the sample set into a training set and a testing set, and respectively using the training set and the testing set for training the machine learning model and testing the prediction effect. And training by adopting an XGboost algorithm in machine learning and using a training sample set, and selecting a proper hyper-parameter for training through cross validation to obtain a final machine learning model.
Step 4): calculating the short-circuit current by using a trained machine learning model;
when the method is applied, a corresponding training mature model is selected according to the fault type of a sample, and then the I under the condition that the IIDGs are not connected is calculated by a traditional short-circuit current calculation program according to a given operation mode and a fault positionfAnd combining the IIDG access condition and the like to form initial characteristics of the sample, obtaining the sample meeting the requirements of the model through the data preprocessing process which is the same as that in the training stage, and finally calling the corresponding model to calculate the short-circuit current to obtain the final short-circuit current calculation result.
Application example
In order to verify the effectiveness and the practicability of the machine learning-based prediction method for the short-circuit current of the power distribution network containing the IIDG of the present invention, the following description will take the IEEE34 node power distribution system shown in fig. 3 as a research object.
The voltage class of the power distribution network is 24.9kV, wherein the voltage class comprises 31 lines in total, the system power supply is connected to the 800 nodes, namely the head end of the power distribution network, and the IIDG is incorporated into the system of the embodiment. The network comprises unbalanced feeders including single-phase feeders and three-phase feeders, and the three-phase feeders have different interphase mutual impedances and different phase loads, so that the network can be used as an unbalanced system for analysis. Considering that M is 2, i.e. the maximum and minimum 2 basic operation modes of the system, the equivalent impedances of the system are j0.5 Ω and j1 Ω respectively. The load and IIDG capacity are the same in both basic modes of operation. Wherein, the load is set according to the standard of an IEEE34 node power distribution system; IIDG in a power distribution network includes: the 844 node is connected with IIDG1 with the basic capacity of 500kW and IIDG4 with the basic capacity of 400kW, the 832 node is connected with IIDG2 with the basic capacity of 200kW, the 852 node is connected with IIDG3 with the basic capacity of 500kW, the IIDGs have fault ride-through capacity, and when a power distribution network fails or is abnormal, the maximum output current of an inverter is 2 times of the rated current of the inverter. According to the method, on the basis of a basic operation mode, the equivalent impedance, the load and the IIDG capacity of the system are randomly generated in the range of [ -0.2,0.2], and meanwhile, the IIDG is randomly generated to be put into a situation matrix A and a line _ cut of a cut line, so that a new operation mode is generated. And setting short-circuit faults in the network, and generating fault samples. Since different fault types have the same flow, the analysis is performed by taking a three-phase short circuit as an example. Faults can occur anywhere in the network, including at the line and bus nodes. The short circuit current is calculated as flowing through the short circuit current at a fixed measurement point, where the 858 node is selected as the measurement point, i.e., the R1 position in fig. 3. 35000 operation modes, namely 17500 groups of each basic operation mode are set. Through simulation, the features and the labels under each operation mode are extracted to obtain 35000 groups of samples to form an original sample set. The original sample set is divided into training set and test set according to 8:2 ratio, i.e. training set 28000 group and test set 7000 group. Since the IEEE34 node system is a three-phase asymmetric system, the following example only selects the c-phase current as a sample label for analysis, and the a and b phases are similar.
To measure the accuracy of the short-circuit current prediction, a Mean Absolute Percentage Error (MAPE) is selected as an index for measuring the accuracy of machine learning. The indicators are defined as follows:
wherein N is the number of samples participating in accuracy evaluation, ytIs the label of the t-th sample,is the predicted value of the t-th sample. MAPE is used for measuring the relative error between the predicted value and the actual value of the short-circuit current, and the smaller the MAPE value is, the more accurate the short-circuit current prediction is.
Pre-processing a sample, comprising: and carrying out independent heat treatment on the characteristic line cut. And (3) using an integration method in machine learning, modeling based on a training sample set, determining the optimal hyper-parameter of the model through cross validation, and testing by using the test sample set to realize the calculation of the short-circuit current of the power distribution network containing the IIDG.
In order to verify the accuracy of the method, partial test data and the prediction result of the model are selected, and the sample value is compared with the prediction value. Some predicted results of the machine learning model on the test set are shown in table 2, including actual values, predicted values and predicted Absolute Percent Error (APE) for the test samples. It can be observed that all APEs predicted in table 2 are less than 2%, which confirms the accuracy of the proposed method for predicting short circuit currents in power distribution networks containing IIDG.
TABLE 2 test set prediction results
Serial number | Actual value/A | Predicted value/A | APE/% |
1 | 351.476 | 350.193 | 0.365 |
2 | 67.772 | 66.706 | 1.573 |
3 | 25.682 | 25.906 | 0.876 |
4 | 48.369 | 47.879 | 1.013 |
5 | 322.865 | 322.978 | 0.035 |
In addition, the advantages of the XGboost algorithm are verified, and the prediction results of different machine learning models are compared. Support Vector Regression (SVR), Random Forest (RF), GBDT methods were selected and compared to XGBoost methods for prediction results, as shown in table 3. As can be seen from the table, the XGBoost algorithm performs better on both the training set and the test set than other algorithms.
TABLE 3 comparison of prediction results for different machine learning models
In order to analyze the influence of the network scale on the required sample number, an IEEE13 node network and an IEEE69 node network test example are added and compared with an IEEE34 node network example. The voltage class of an IEEE13 node power distribution network is 4.16kV, the power distribution network comprises 10 lines in total, loads are arranged according to a standard, IIDGs with the basic capacity of 200kW are respectively connected to nodes 633 and 680, the position of R2 is selected as a fixed measuring point, and the structure diagram of the network is shown in figure 4; an IEEE69 node power distribution network has a voltage class of 12.66kV and comprises 68 lines in total, loads are set according to a standard, nodes 19, 25, 32, 45, 54 and 65 are connected to an IIDG with a basic capacity of 200kW respectively, the position of a node 7, namely R3 is selected as a fixed measuring point, and the structure diagram of the network is shown in figure 5. The simulation is still performed according to the procedure described above. Table 4 compares the prediction effects of the XGBoost model in three power distribution networks, including the prediction error of the test set and the average prediction speed of the samples.
TABLE 4 comparison of prediction results for different distribution networks
As can be seen from the table, for the same power distribution network, with the increase of the number of samples, the prediction effect of the model on the short-circuit current is improved, and the average prediction speed of the model on the samples is basically kept stable. Meanwhile, as the network scale and the IIDG number are increased, the prediction difficulty of the machine learning model on the short-circuit current is increased, the characteristic quantity is increased, the prediction error and the prediction time of the model are increased to a certain extent, but in general, the error can meet the use requirement, and the prediction speed of each model is high. Therefore, with the enlargement of the scale of the power distribution network, the requirement of the model on the number of samples does not increase in a geometric progression on the premise of meeting the requirements of an error level and a prediction speed. Therefore, the method provided by the invention has better applicability in a power distribution network with larger scale and high IIDG penetration.
Therefore, the prediction method of the short-circuit current containing the IIDG based on the machine learning is high in precision, high in calculation speed and good in applicability, and provides support for research in the directions of relay protection of the power distribution network and the like.
Claims (1)
1. The method for predicting the short-circuit current of the power distribution network containing the IIDG based on machine learning is characterized by comprising the following steps of: the method comprises the following steps:
step 1): determining sample composition for training;
firstly, 10 machine learning models are respectively established for 10 fault types f _ types for prediction, so that 10 different samples exist relatively; f _ type is a fault type and is used for representing different sample types, wherein 10 fault types are 10 combinations of four fault types including single-phase grounding, two-phase interphase and three-phase short circuit in three ABC phases;
secondly, measuring the short-circuit current I flowing through a point when the power distribution network has the same fault in the same operation mode but is not connected to the inverter type distributed power source IIDGfAs one of the sample characteristics; the remaining sample characteristics include: situation of IIDG investing in distribution network αjIIDG input capacity SDGjJ (th) capacity S of DG to be put into the power distribution networkDGjLine _ cut of the cut line, f _ line of the fault line and f _ loc of the fault position;
measuring point flowing short-circuit current I after IIDG is connectedf_DGAs a sample label;
the sample characteristics and the sample labels jointly form a sample of a short-circuit current calculation model under the corresponding fault type of the power distribution network;
step 2): randomly generating an operation mode, and accumulating a sample set;
in order to make the obtained samples have more diversity, M basic operation modes are considered for a power distribution network comprising c rotary power supplies, s IIDG nodes and l load nodes, wherein each basic operation mode can be used for setting rotary power supplies, IIDG nodes and load basic configuration parameters according to requirements;
meanwhile, considering the input operation condition of the IIDG, a vector α ═ α is randomly generated for each operation mode1 α2…αj]In which α isj0 or 1, which means that the jth DG is not put into the system or put into the system, respectively; considering the topological structure change of the network, from the perspective of an N-1 principle, each operation mode randomly generates 1 line from a line set as a line _ cut of a cut line; the line set comprises all lines in the power distribution network and is represented by line numbers, wherein the number 0 represents that no line is cut off, namely a complete network structure; in addition, randomly generating fault conditions including a fault line f _ line and a fault position f _ loc; wherein, the fault line is also selected from the line set, but does not contain the number 0;
when a sample is generated, the I removal rate is obtained by the random and automatic combination of the operation mode and the fault conditionfSteady state characteristics and fault characteristics of other power distribution networks, and short-circuit current I flowing through the measuring point obtained by simulation after the IIDG is connectedf_DGA label;
then, under the condition of each simulated operation mode and fault, respectively obtaining the short-circuit current characteristics I flowing through the measuring point under the condition of not connecting any IIDG for 10 fault types through simulation calculationfThereby accumulating a sample set of 10 machine learning models;
step 3): preprocessing the obtained samples to obtain a training sample set, and finishing machine learning model training;
preprocessing a sample to obtain a sample set more suitable for machine learning model training; dividing a sample set into a training set and a testing set, and respectively using the training set and the testing set for training of a machine learning model and testing of a prediction effect; an XGboost algorithm in machine learning is adopted, a training sample set is used for training, and proper hyper-parameters are selected through cross validation for training to obtain a final machine learning model;
step 4): calculating the short-circuit current by using a trained machine learning model;
when the method is applied, a corresponding training mature model is selected according to the fault type of a sample, and then I under the condition that the IIDGs are not accessed is calculated according to a given operation mode and fault positionfCombining with IIDG access case to form initial characteristics of sample, and thenAnd obtaining a sample meeting the requirements of the model through a data preprocessing process which is the same as that in the training stage, and finally calling the model to calculate the short-circuit current to obtain a final short-circuit current calculation result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010089399.0A CN111262238B (en) | 2020-02-12 | 2020-02-12 | Machine learning-based method for predicting short-circuit current of power distribution network containing IIDG |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010089399.0A CN111262238B (en) | 2020-02-12 | 2020-02-12 | Machine learning-based method for predicting short-circuit current of power distribution network containing IIDG |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111262238A CN111262238A (en) | 2020-06-09 |
CN111262238B true CN111262238B (en) | 2021-09-03 |
Family
ID=70954491
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010089399.0A Active CN111262238B (en) | 2020-02-12 | 2020-02-12 | Machine learning-based method for predicting short-circuit current of power distribution network containing IIDG |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111262238B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113761793B (en) * | 2021-08-16 | 2024-02-27 | 固德威技术股份有限公司 | Inverter output impedance detection device and method and inverter operation control method |
CN113991652B (en) * | 2021-10-27 | 2024-05-17 | 浙江大学 | Data-driven multi-output calculation method for short-circuit current of distribution network containing IIDG |
CN116256602B (en) * | 2023-05-15 | 2023-07-11 | 广东电网有限责任公司中山供电局 | Method and system for identifying state abnormality of low-voltage power distribution network |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014047733A1 (en) * | 2012-09-27 | 2014-04-03 | Rajiv Kumar Varma | Fault detection and short circuit current management technique for inverter based distributed generators (dg) |
CN108663601A (en) * | 2018-05-11 | 2018-10-16 | 山东理工大学 | A kind of distribution network failure current management method based on IIDG |
CN109490704A (en) * | 2018-10-16 | 2019-03-19 | 河海大学 | A kind of Fault Section Location of Distribution Network based on random forests algorithm |
CN110120669A (en) * | 2019-04-29 | 2019-08-13 | 国网河北省电力有限公司经济技术研究院 | Limit rack method of adjustment, device and the terminal device of grid short circuit electric current |
CN110323784A (en) * | 2019-07-25 | 2019-10-11 | 金华电力设计院有限公司 | Meter and the probabilistic photovoltaic power generation short circuit current appraisal procedure of low voltage crossing |
CN110635479A (en) * | 2019-10-25 | 2019-12-31 | 中国南方电网有限责任公司 | Intelligent aid decision-making method and system for limiting short-circuit current operation mode |
-
2020
- 2020-02-12 CN CN202010089399.0A patent/CN111262238B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2014047733A1 (en) * | 2012-09-27 | 2014-04-03 | Rajiv Kumar Varma | Fault detection and short circuit current management technique for inverter based distributed generators (dg) |
CN108663601A (en) * | 2018-05-11 | 2018-10-16 | 山东理工大学 | A kind of distribution network failure current management method based on IIDG |
CN109490704A (en) * | 2018-10-16 | 2019-03-19 | 河海大学 | A kind of Fault Section Location of Distribution Network based on random forests algorithm |
CN110120669A (en) * | 2019-04-29 | 2019-08-13 | 国网河北省电力有限公司经济技术研究院 | Limit rack method of adjustment, device and the terminal device of grid short circuit electric current |
CN110323784A (en) * | 2019-07-25 | 2019-10-11 | 金华电力设计院有限公司 | Meter and the probabilistic photovoltaic power generation short circuit current appraisal procedure of low voltage crossing |
CN110635479A (en) * | 2019-10-25 | 2019-12-31 | 中国南方电网有限责任公司 | Intelligent aid decision-making method and system for limiting short-circuit current operation mode |
Non-Patent Citations (1)
Title |
---|
含低电压穿越型分布式电源配电网的短路电流计算方法;杨杉等;《电力***自动化》;20160610;第40卷(第11期);第93-99、151页 * |
Also Published As
Publication number | Publication date |
---|---|
CN111262238A (en) | 2020-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Kong et al. | Fault detection and location method for mesh-type DC microgrid using pearson correlation coefficient | |
CN111262238B (en) | Machine learning-based method for predicting short-circuit current of power distribution network containing IIDG | |
Gomez et al. | Support vector machine-based algorithm for post-fault transient stability status prediction using synchronized measurements | |
CN109033702A (en) | A kind of Transient Voltage Stability in Electric Power System appraisal procedure based on convolutional neural networks CNN | |
CN104716646B (en) | A kind of node Coupling Degrees method based on Injection Current | |
CN111064182B (en) | Short-circuit current calculation method based on power grid characteristics | |
CN111159841B (en) | Power distribution network short-circuit current calculation method based on data fusion | |
CN106033894B (en) | Method for judging stability of extra-high voltage direct current multi-drop point power grid | |
CN117148212A (en) | Single-phase earth fault line selection method for small-current grounding system of micro-grid power distribution network | |
Uddin et al. | Hybrid machine learning-based intelligent distance protection and control schemes with fault and zonal classification capabilities for grid-connected wind farms | |
CN105375475B (en) | A kind of power system fast transient stability simulation method based on Real-time Dynamic Equivalence | |
CN107317326B (en) | Grid regulation current limiting method based on improved REI equivalence | |
Wang et al. | Measurement-based coherency identification and aggregation for power systems | |
CN107565547A (en) | A kind of power distribution network operation reliability evaluation and optimization system | |
CN116187082A (en) | Single-machine equivalent modeling method for wind power plant | |
CN113991652B (en) | Data-driven multi-output calculation method for short-circuit current of distribution network containing IIDG | |
Lu et al. | An optimal reactive power compensation allocation method considering the economic value affected by voltage sag | |
Liao et al. | Identification of Fault Line Selection and Section for Single-Phase Ground Fault in Small Current Grounding System | |
CN112670966B (en) | Self-adaptive current protection method for photovoltaic power distribution network | |
Massucco et al. | Evaluation of some indices for voltage stability assessment | |
CN113315100A (en) | Micro-grid protection method and system based on convolutional neural network | |
Tao et al. | Transient stability analysis of AC/DC system considering electromagnetic transient model | |
Alfieri et al. | Impact of Photovoltaic Generators on the Three Phase Short Circuit Operating Conditions | |
Magalhães et al. | Parametric regression in synchronous and induction generators | |
Zaker et al. | Measurement-based equivalent model of distribution networks considering static and dynamic loads |
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 |