CN116306236A - Wind power plant dynamic equivalent modeling method and system based on GRU-LSTM-FC combined network - Google Patents
Wind power plant dynamic equivalent modeling method and system based on GRU-LSTM-FC combined network Download PDFInfo
- Publication number
- CN116306236A CN116306236A CN202310081362.7A CN202310081362A CN116306236A CN 116306236 A CN116306236 A CN 116306236A CN 202310081362 A CN202310081362 A CN 202310081362A CN 116306236 A CN116306236 A CN 116306236A
- Authority
- CN
- China
- Prior art keywords
- wind
- gru
- lstm
- combined network
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000012549 training Methods 0.000 claims abstract description 65
- 238000012360 testing method Methods 0.000 claims abstract description 38
- 210000002569 neuron Anatomy 0.000 claims abstract description 28
- 230000004044 response Effects 0.000 claims abstract description 18
- 238000004088 simulation Methods 0.000 claims abstract description 15
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 12
- 230000002068 genetic effect Effects 0.000 claims abstract description 11
- 239000010410 layer Substances 0.000 claims description 108
- 230000006870 function Effects 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 12
- 239000002356 single layer Substances 0.000 claims description 12
- 210000000349 chromosome Anatomy 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 9
- 230000008859 change Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000001052 transient effect Effects 0.000 claims description 3
- 238000004590 computer program Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 230000007246 mechanism Effects 0.000 description 5
- 238000012545 processing Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/06—Wind turbines or wind farms
-
- 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]
-
- 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
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/28—The renewable source being wind energy
-
- 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
- Y02E10/00—Energy generation through renewable energy sources
- Y02E10/70—Wind energy
- Y02E10/76—Power conversion electric or electronic aspects
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Evolutionary Biology (AREA)
- Educational Administration (AREA)
- Marketing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Primary Health Care (AREA)
- Geometry (AREA)
- Water Supply & Treatment (AREA)
- Power Engineering (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Game Theory and Decision Science (AREA)
Abstract
The invention discloses a wind power plant dynamic equivalent modeling method and a system based on a GRU-LSTM-FC combined network, firstly, a mixed wind power plant containing a doubly-fed fan and a direct-driven fan is built; secondly, obtaining output power of the wind power plant under the voltage drop disturbance of different wind speeds, wind directions and grid-connected points through simulation, establishing a historical data set, and dividing the historical data set into a training set and a testing set; thirdly, training the GRU-LSTM-FC combined network by adopting training set data to obtain combined network model parameters; then, optimizing the FC layer number and the neuron number of each layer of the combined network by adopting a genetic algorithm to obtain optimized combined network model parameters; and finally, predicting the test set data by adopting the optimized combined network to obtain a wind power plant dynamic response prediction result. According to the invention, a data driving method is adopted to carry out dynamic modeling on the wind power plant, so that the method has good adaptability to wind power plants containing different types of units or wind power plants with complex terrains, and the modeling precision is improved.
Description
Technical Field
The invention belongs to the field of modeling of power systems, and particularly relates to a wind power plant dynamic equivalent modeling method and system based on a GRU-LSTM-FC combined network.
Background
The dynamic equivalent modeling of the new energy station has important significance for guaranteeing the safe and stable operation of the power system. The equivalent modeling method comprises 3 classes of mechanism modeling, non-mechanism modeling and data-driven modeling. Most of the current research focused on mechanism modeling and non-mechanism modeling is performed, however, both modeling methods depend on an operation mode and specific disturbance, and a universal equivalent model of the wind power plant with strong universality is difficult to obtain.
In recent years, with the continuous progress of big data technology, data-driven modeling methods are becoming more and more important. The method not only utilizes the response data under a certain specific disturbance, but also fully excavates the characteristic information of the historical data, and has better adaptability compared with a non-mechanism modeling method based on a transfer function. The current modeling method for data driving is mainly focused on the field of deep learning, and is characterized in that a deep learning network suitable for wind power plant dynamic modeling is determined, and meanwhile, the problems of network structure adjustment, super-parameter optimization and the like are faced.
Disclosure of Invention
The invention aims to: the invention aims to provide a high-adaptability wind power plant dynamic equivalent modeling method and system based on a gating circulation unit-long and short term memory-full connection (GRU-LSTM-FC) combined network.
The technical scheme is as follows: according to the wind power plant dynamic equivalent modeling method based on the GRU-LSTM-FC combined network, aiming at a target wind power plant, the following steps are carried out to obtain the active power and reactive power dynamic response time sequence of the target wind power plant;
step 1: constructing a target wind power plant comprising a doubly-fed wind turbine DFIG and a direct-driven wind turbine PMSG;
step 2: according to wind speed data and wind direction data measured by a wind measuring tower, calculating wind speed data of each wind turbine in a target wind power plant, obtaining a voltage time sequence of a grid-connected point of the target wind power plant according to disturbance simulation, taking the wind speed data, the wind direction data and the voltage time sequence of the grid-connected point of each wind turbine as inputs, taking a dynamic response time sequence of active power and reactive power of the target wind power plant as outputs, and constructing a training set and a test set;
step 3: training the GRU-LSTM-FC combined network by adopting data of a training set to obtain GRU-LSTM-FC combined network model parameters, optimizing the number of FC layers and the number of neurons of each layer in the GRU-LSTM-FC combined network by adopting a genetic algorithm based on the GRU-LSTM-FC combined network model parameters, and obtaining an optimized GRU-LSTM-FC combined network model;
step 4: and inputting the wind speed data and wind direction data of each wind turbine in the test set and the voltage time sequence of the grid connection point into the optimized GRU-LSTM-FC combined network model, and outputting the dynamic response time sequence of the active power and the reactive power of the target wind power plant.
In step 1, the target wind power plant is boosted in two stages and then is connected to a CEPRI-36 node system.
Further, the step 2 specifically includes the following steps:
(a) Wind turbine generator steady-state power calculation considering wake effect
According to the wind speed and wind direction measured by the wind measuring tower, calculating the input wind speed of each unit in the wind power plant under the wake effect, wherein the input wind speed is specifically as follows:
in the formula, v 0 The wind speed is measured by the wind measuring tower; v d The wind speed is the wind speed which is influenced by the wake flow of the fan at the position of the wind tower and away from the wind tower d; r is the radius of the wind wheel;
is the wake drop constant; c (C) T Is a thrust coefficient; v i The actual wind speed of the wind turbine generator is; v i0 The wind speed of the ith fan is not considered when the wake flows are not considered; v j-i To take into account the wind speed of the jth wind turbine acting at the ith when the wake; n is the total number of units, and the machine end wind speed of the fan is according to the wind direction v at the wind measuring tower wd Is changed by a change in (a);
calculating the steady-state power P of each wind turbine according to the wind speed of the wind turbine at the wind turbine end 0 The method specifically comprises the following steps:
wherein v is the wind speed at the machine end; v in And v out The cut-in wind speed and the cut-out wind speed are respectively; v N Is the rated wind speed; p (P) N Rated power of the wind turbine generator;
(b) Wind farm LVRT process simulation setting
The wind power plant is at t under different wind speeds and directions 0 Grid-connected point of wind power plant is short-circuited in three phases, t 1 Removing faults at any time, obtaining different voltage drop depths by adjusting grounding impedance, and obtaining active power and reactive power time sequences of the wind power plant through simulation;
(c) Construction of training and test sets
The input data comprise wind speed and wind direction data measured by a wind measuring tower and a voltage time sequence of a grid-connected point of the wind power plant; the output data includes a time series of active and reactive power of the wind farm.
The training set data includes: the wind speed interval is 1m/s within the range of 3m/s of cut-in wind speed and 12m/s of rated wind speed; the wind direction is [0,180 ]](due to [180 °,360 ]]And [0,180 ]]The wake effect due to the change of the wind direction in the wind direction is uniform, so that only 0,180 DEG is considered]Internal wind direction), the direction interval is 10 °. Voltage drop range of [0.1,0.9 ]]U n Drop interval of 0.1U n . The total sampling duration of the data is 1s, and the sampling interval is 0.001s, so that 1710 sets (10×19×9) of sample data are shared as a training set, and each data length is 1000.
The test set data includes: the three types of data of wind speed, wind direction and voltage drop degree of PCC point of the wind power plant are all in smaller change scale (namely 0.1m/s, 1 degree and 0.01U) n ) The random sets were combined into 100 sets, constituting test data.
Further, the step 3 specifically includes the following steps:
(a) Construction of GRU-LSTM-FC combined network
The constructed combined network comprises 3 layers, wherein the 1 st layer is a GRU layer, the 2 nd layer is an LSTM layer, the 3 rd layer is an FC layer, and data sequentially passes through the GRU layer, the LSTM layer and the FC layer;
(b) Performance evaluation index of combined network
The method adopts a mean square error MSE index to evaluate the matching degree between the output and the actual response of the GRU-LSTM-FC combined network, and uses the absolute error of each time point as another index to describe the instantaneous error in the whole transient region, and is specifically as follows:
wherein y is i The actual response value of the wind power plant at the moment i;
the output of the GRU-LSTM-FC combined network at the moment i; n=1000, where N is the total number of data sampled during the whole time domain simulation process;
(c) Determination of the number of layers of a combined network
Performing performance evaluation on five networks, namely single-layer LSTM, single-layer GRU, double-layer LSTM, double-layer GRU and single-layer GRU-single-layer LSTM, respectively from the two aspects of iteration efficiency and mean square error of a training set and a testing set, wherein under the five conditions, the number of FC full-connection layers is uniformly set to be 1, and the number of neurons of all hidden layers is set to be 50; the iteration termination error of the combined network training is set to 10 -2 The maximum iteration number is 30000, and the model learning rate I r The inverse proportion function taking the iteration number epoch as an independent variable is expressed as follows:
wherein I is r0 ＝10 -2 Is the initial learning rate of the model; epoch is the number of iterations in the model training process;
according to the mean square error of the five networks in the training set and the testing set, the minimum error is obtained when the GRU is 1 layer and the LSTM is 1 layer, and the GRU is determined to be 1 layer and the LSTM is 1 layer according to the minimum error.
Further, the step 4 specifically includes the following steps:
(a) Super-parameters to be optimized of combined network
The super parameters to be optimized comprise the number of FC layers and the number of each neuron of GRU, LSTM and FC;
(b) Super-parameter optimization based on genetic algorithm
Carrying out population chromosome coding, and constructing chromosome C structures of individuals in the population according to the number of FC layers and the number of neurons of each hidden layer, wherein the chromosome C structures are as follows:
C＝[L FC ,N GRU ,N LSTM ,N FC1 .,...,N FCn ]
wherein L is FC The number of layers for FC; n (N) GRU For the number of GRU layer neurons, N LSTM Is LSTM layer neuronNumber N FC1 ,…,N FCn For the number of neurons in each layer of the FC layer,
designing an fitness function, comprehensively measuring the performance of the combined network according to three indexes of accuracy, generalization and training efficiency, and designing the fitness function J as the sum of the three indexes, wherein the expression is as follows:
wherein J is an fitness function; loss (Low Density) train0 And Loss of train For the mean square error and the total mean square error of the training set, the ratio of 2 is used for representing the accuracy; loss (Low Density) test0 And Loss of test The mean square error and the total mean square error of the test set are respectively, and the ratio of 2 is used for representing the generalization capability of the model; t (T) 0 For training time length, the value is the average time length of multiple training, T start And T end Training start and end times, respectively, in min, ratio T 0 /(T end -T start ) The training efficiency of the network is obtained; and alpha, beta and gamma are weight coefficients of the indexes respectively, and an optimized hyper-parameter result is obtained according to iteration termination conditions.
The invention also discloses a wind farm dynamic equivalent modeling system based on the GRU-LSTM-FC combined network, which comprises:
modeling unit: based on a PSASP platform, constructing a target wind power plant containing n wind turbines, and accessing the target wind power plant into a CEPRI-36 node system after two-stage boosting;
simulation unit: simulating a target wind power plant according to wind speed data, wind direction data and grid-connected point voltage drop disturbance to obtain output power of the wind power plant, and constructing a training set and a testing set;
an optimizing unit: training the GRU-LSTM-FC combined network by adopting training set data, and determining the GRU layer number and the LSTM layer number of the GRU-LSTM-FC combined network through performance indexes; optimizing the number of FC layers and the number of neurons of each layer of the combined network by adopting a genetic algorithm to obtain optimized GRU-LSTM-FC combined network model parameters;
prediction unit: and according to input data of a certain wind speed, a certain wind direction and a certain voltage drop disturbance in the test set, predicting by adopting the optimized GRU-LSTM-FC combined network to obtain the active power and reactive power dynamic response time sequence of the wind power plant.
Furthermore, in the modeling unit, the target wind power plant is subjected to two-stage boosting and then is connected into the CEPRI-36 node system through the BUS30 BUS.
The beneficial effects are that: compared with the prior art, the invention has the following remarkable advantages: the invention provides a data driving modeling method based on a GRU-LSTM-FC combined network, which can mine and learn internal features of a large amount of historical data, wherein the more the historical data is, the higher the modeling precision is. Meanwhile, the model has strong universality and has good adaptability to wind farms with different types of units or wind farms with complex terrains.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention;
FIG. 2 is a topological structure diagram of a wind farm;
FIG. 3 is a CEPRI-36 node system;
FIG. 4 is a graph of mean square error during training for five network structures;
FIG. 5 is a graph showing the fitness ratio of the number of different FC layers in each generation of genetic population;
FIG. 6 is a graph of mean square error of test sample modeling before and after GA optimization;
fig. 7 shows the mean square error of the four networks in the test sample data.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings.
Example 1:
the implementation flow is shown in figure 1.
The structure of the hybrid wind power plant is shown in figure 2, and the wind power plant consists of 18 doubly-fed wind turbines with rated power of 1.5MW and 18 direct-drive permanent magnet wind turbines with rated power of 1.5MW, wherein the distance between the wind turbines is 500m. The wind farm is connected to BUS30 BUS of the CEPRI-36 node system through two-stage boosting of the machine end box transformer and the main transformer.
The CEPRI-36 node system structure is shown in figure 3, and detailed parameters of the wind turbine generator and the collector line are shown in table 1.
TABLE 1 wind farm model parameters
And 2, calculating the wind speed of each wind turbine generator in the target wind power plant according to the wind speed and the wind direction measured by the wind measuring tower, obtaining the voltage, the active power and the reactive power time sequence of the grid-connected point of the target wind power plant according to the disturbance simulation, obtaining a historical data set formed by { wind speed, wind direction, voltage time sequence, active power time sequence and reactive power time sequence }, and dividing the historical data set into a training set and a testing set.
(a) Wind turbine generator steady-state power calculation considering wake effect
According to the wind speed and wind direction measured by the wind measuring tower, calculating the input wind speed of each unit in the wind power plant under the wake effect, wherein the input wind speed is specifically as follows:
in the formula, v 0 The wind speed is measured by the wind measuring tower; v d The wind speed is the wind speed which is influenced by the wake flow of the fan at the position of the wind tower and away from the wind tower d; r is the radius of the wind wheel;
is the wake drop constant; c (C) T Is a thrust coefficient; v i The actual wind speed of the wind turbine generator is; v i0 The wind speed of the ith fan is not considered when the wake flows are not considered; v j-i To take into account the wind speed of the jth wind turbine acting at the ith when the wake; n is the total number of units. At the same time, wind direction v at wind tower wd Different, the machine end wind speed of fan is different.
Calculating the steady-state power P of each wind turbine according to the wind speed of the wind turbine at the wind turbine end 0 The method specifically comprises the following steps:
wherein v is the wind speed at the machine end; v in And v out The cut-in wind speed and the cut-out wind speed are respectively; v N Is the rated wind speed; p (P) N Is the rated power of the wind turbine generator.
(b) Wind farm LVRT process simulation setting
And under different wind speeds and wind directions, a three-phase short circuit fault occurs at a grid-connected point of the wind power plant when t=0.1 s, the fault is removed at the moment of 0.25s, different voltage drop depths are obtained by adjusting the grounding impedance, and active power and reactive power tracks of the wind power plant are obtained through simulation.
(c) Construction of training and test sets
The input data comprise wind speed and wind direction data measured by a wind measuring tower and a voltage time sequence of a grid-connected point of the wind power plant; the output data includes a time series of active and reactive power of the wind farm.
The training set data includes: the wind speed interval is 1m/s within the range of 3m/s of cut-in wind speed and 12m/s of rated wind speed; the wind direction is [0,180 ]](due to [180 °,360 ]]And [0,180 ]]The wake effect due to the change of the wind direction in the wind direction is uniform, so that only 0,180 DEG is considered]Internal wind direction), the direction interval is 10 °. Voltage drop range of [0.1,0.9 ]]U n Drop interval of 0.1U n . The total sampling duration of the data is 1s, and the sampling interval is 0.001s, so that 1710 sets (10×19×9) of sample data are shared as a training set, and each data length is 1000.
The test set data includes: the three types of data of wind speed, wind direction and voltage drop degree of PCC point of the wind power plant are all in smaller change scale (namely 0.1m/s, 1 degree and 0.01U) n ) The random sets were combined into 100 sets, constituting test data.
Training the GRU-LSTM-FC combined network by adopting training set data to obtain GRU-LSTM-FC combined network model parameters;
(a) Construction of GRU-LSTM-FC combined network
The constructed combined network comprises 3 layers, wherein the 1 st layer is a GRU layer, the 2 nd layer is an LSTM layer, and the 3 rd layer is an FC layer.
(b) Performance evaluation index of combined network
A Mean Square Error (MSE) indicator is used to evaluate the degree of match between the output and the actual response based on the GRU-LSTM-FC combined network. Meanwhile, the absolute error (error) at each time point is used as another index to describe the instantaneous error in the whole transient region, and the method is as follows:
wherein y is i The actual response value of the wind power plant at the moment i;
the output of the GRU-LSTM-FC combined network at the moment i; n=1000, where N is the total number of data sampled during the entire time domain simulation.
(c) Determination of the number of layers of a combined network
In order to analyze the influence of the GRU and LSTM layers in the combined network on the equivalent model performance of the wind power plant, the five networks of single-layer LSTM, single-layer GRU, double-layer LSTM, double-layer GRU and single-layer GRU-single-layer LSTM are respectively subjected to performance evaluation from the two aspects of iteration efficiency and mean square error of a training set and a testing set. In the above five cases, the number of FC full-connection layers was uniformly set to 1 layer, and the number of neurons in all hidden layers was set to 50. The iteration termination error of the combined network training is set to 10 -2 The maximum iteration number is 30000, and the model learning rate I r The inverse proportion function taking the iteration number epoch as an independent variable is expressed as follows:
wherein I is r0 ＝10 -2 Is the initial learning rate of the model; epoch is the number of iterations in the model training process.
The mean square error of the five networks in the training process is shown in fig. 4, and the mean square error of the training set and the test set is shown in table 2. As can be seen from fig. 4 and table 2: the error is minimal when the GRU is layer 1 and the LSTM is layer 1. The GRU is thus determined to be layer 1 and the LSTM is determined to be layer 1.
Table 2 mean square error of five network structures in training set and test set
(a) Super-parameters to be optimized of combined network
The super parameters to be optimized include the number of FC layers and the number of neurons for GRU, LSTM and FC.
(b) Ultra-parameter optimization based on GA algorithm
Population chromosome coding is first performed. The chromosome C structure of the individuals in the population is constructed according to the number of FC layers and the number of neurons of each hidden layer, and the chromosome C structure is as follows:
C＝[L FC ,N GRU ,N LSTM ,N FC1 .,...,N FCn ]
wherein L is FC The number of layers for FC; n (N) GRU For the number of GRU layer neurons, N LSTM For the number of LSTM layer neurons, N FC1 ,…,N FCn The number of neurons for each layer of the FC layer.
The fitness function is then designed. According to three indexes of accuracy, generalization and training efficiency, comprehensively measuring the performance of the combined network, designing an fitness function J as the sum of the three indexes, wherein the expression is as follows:
wherein J is an fitness function; loss (Low Density) train0 And Loss of train For the mean square error and the total mean square error of the training set, the ratio of 2 is used for representing the accuracy; loss (Low Density) test0 And Loss of test The mean square error and the total mean square error of the test set are respectively, and the ratio of 2 is used for representing the generalization capability of the model; t (T) 0 For training time length, the value is the average time length of multiple training, T start And T end Training start and end times, respectively, in min, ratio T 0 /(T end -T start ) The training efficiency of the network is obtained; alpha, beta and gamma are weight coefficients of the indexes respectively.
The adaptation ratio of the iteration times of the GA in the iteration process and the number of different FC layers (including the number of neurons of each layer under the corresponding FC layers) is shown in figure 5, and the GA algorithm converges when iterating to the 10 th time in the calculation example. It can be seen that: in the iteration process, the fitness ratio is highest when the number of FC layers is 6, and the fitness ratio reaches 88.684% when the iteration is 10, so that the FC is determined to be 6 layers, the chromosome optimal value is C= [6,71,66,48,79,45,26,77,49], namely the FC is 6 layers, the neuron numbers of GRU and LSTM are 71 and 66 respectively, and the neuron numbers of the FC layers are 48,79,45,26,77 and 49 respectively.
And 5, predicting the test set data by adopting the optimized GRU-LSTM-FC combined network to obtain a wind power plant dynamic response prediction result.
Based on the optimized GRU-LSTM-FC combined network, dynamic equivalent modeling is carried out on the wind power plant, the mean square error of the output power of the GRU-LSTM-FC combined network under 100 groups of test set samples is shown in figure 6, and the mean square error of the output power before the combined network is optimized is also shown in the figure. It can be seen that: the GRU-LSTM-FC combined network based on GA optimization can remarkably improve equivalent modeling precision of the wind power plant.
In order to verify the equivalent effect of the GRU-LSTM-FC combined network, the invention uses the calculation example and the training data as examples, and adopts BP neural network, RNN network and nonlinear autoregressive neural Network (NARX) with external source input to model. The mean square error of the active power and reactive power in 100 groups of test samples for the 4 models is shown in fig. 7 and table 3, respectively. Both fig. 7 and table 3 show that the combined model of the present invention not only has higher equivalent accuracy, but also training time is within an acceptable range.
Table 3 comparison of modeling performance of four models
| Network structure | MSE(P) | MSE(Q) | Average training time/s |
| BP | 4.00215 | 0.04439 | 91 |
| RNN | 1.04354 | 0.41124 | 257 |
| NARX | 0.51146 | 0.63713 | 289 |
| Model herein | 0.01941 | 0.00889 | 173 |
Example 2:
the invention provides a wind power plant dynamic equivalent modeling system based on a GRU-LSTM-FC combined network, which comprises the following components:
modeling unit: based on a PSASP platform, a wind power plant containing 36 wind turbines is constructed, and the wind power plant is subjected to two-stage boosting and then is connected into a CEPRI-36 node system through a BUS30 BUS.
Simulation unit: and simulating the target wind power plant according to the wind speed, the wind direction and the voltage drop disturbance of the grid-connected point to obtain an output power time sequence of the lower wind power plant, obtaining a historical data set formed by { wind speed, wind direction, voltage time sequence, active power time sequence and reactive power time sequence }, and dividing the historical data set into a training set and a testing set. .
An optimizing unit: training the GRU-LSTM-FC combined network by adopting training set data, and determining the GRU layer number and the LSTM layer number of the GRU-LSTM-FC combined network through performance indexes; further optimizing the number of FC layers and the number of neurons of each layer of the combined network by adopting a Genetic Algorithm (GA) to obtain optimized GRU-LSTM-FC combined network model parameters;
prediction unit: and according to input data of a certain wind speed, a certain wind direction and a certain voltage drop disturbance in the test set, predicting by adopting the optimized GRU-LSTM-FC combined network to obtain the active power and reactive power dynamic response time sequence of the wind power plant.
For a specific implementation of each module of the device according to the invention, reference is made to the specific implementation of the above-mentioned method.
It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that it will be apparent to those skilled in the art that modifications and variations can be made without departing from the technical principles of the present invention, and these modifications and variations should also be regarded as the scope of the invention.
Claims (7)
1. A wind power plant dynamic equivalent modeling method based on a GRU-LSTM-FC combined network is characterized in that for a target wind power plant, the following steps are executed to obtain the active power and reactive power dynamic response time sequence of the target wind power plant;
step 1: constructing a target wind power plant comprising a doubly-fed wind turbine DFIG and a direct-driven wind turbine PMSG;
step 2: according to wind speed data and wind direction data measured by a wind measuring tower, calculating wind speed data of each wind turbine in a target wind power plant, obtaining a voltage time sequence of a grid-connected point of the target wind power plant according to disturbance simulation, taking the wind speed data, the wind direction data and the voltage time sequence of the grid-connected point of each wind turbine as inputs, taking a dynamic response time sequence of active power and reactive power of the target wind power plant as outputs, and constructing a training set and a test set;
step 3: training the GRU-LSTM-FC combined network by adopting data of a training set to obtain GRU-LSTM-FC combined network model parameters, optimizing the number of FC layers and the number of neurons of each layer in the GRU-LSTM-FC combined network by adopting a genetic algorithm based on the GRU-LSTM-FC combined network model parameters, and obtaining an optimized GRU-LSTM-FC combined network model;
step 4: and inputting the wind speed data and wind direction data of each wind turbine in the test set and the voltage time sequence of the grid connection point into the optimized GRU-LSTM-FC combined network model, and outputting the dynamic response time sequence of the active power and the reactive power of the target wind power plant.
2. The method for dynamically modeling equivalent of a wind power plant based on a GRU-LSTM-FC combined network according to claim 1, wherein in step 1, the target wind power plant is constructed based on a PSASP platform and comprises a doubly fed wind turbine DFIG and a directly driven wind turbine PMSG, and the target wind power plant is connected to a CEPRI-36 node system after two-stage boosting.
3. The method for modeling the dynamic equivalence of the wind farm based on the GRU-LSTM-FC combined network according to claim 1, wherein the step 2 specifically comprises the following steps:
(a) Wind turbine generator steady-state power calculation considering wake effect
According to the wind speed and wind direction measured by the anemometer tower, calculating the input wind speed of each unit in the wind power plant under the wake effect, wherein the input wind speed is specifically as follows:
in the formula, v 0 The wind speed is measured by the wind measuring tower; v d The wind speed is the wind speed which is influenced by the wake flow of the fan at the position of the wind tower and away from the wind tower d; r is the radius of the wind wheel;
is the wake drop constant; c (C) T Is a thrust coefficient; v i The actual wind speed of the wind turbine generator is; v i0 The wind speed of the ith fan is not considered when the wake flows are not considered; v j-i To take into account the wind speed of the jth wind turbine acting at the ith when the wake; n is the total number of units, n=n D +n P ，n D Is the number of DFIG of the doubly-fed fans, n P The number of the PMSG directly driven fans is the number of PMSG, and the machine end wind speed of the fans is according to the wind direction v at the wind measuring tower wd Is changed by a change in (a); calculating the steady-state power P of each wind turbine according to the wind speed of the wind turbine at the wind turbine end 0 The method specifically comprises the following steps:
wherein v is the wind speed at the machine end; v in And v out The cut-in wind speed and the cut-out wind speed are respectively; v N Is the rated wind speed; p (P) N Rated power of the wind turbine generator;
(b) Wind farm LVRT process simulation setting
The wind power plant is at t under different wind speeds and directions 0 Grid-connected point of wind power plant is short-circuited in three phases, t 1 Faults are removed at moment, different voltage drop depths are obtained by adjusting grounding impedance, and active power and reactive power time of the wind power plant are obtained through simulationA sequence;
(c) Construction of training and test sets
The input data comprise wind speed and wind direction data measured by a wind measuring tower and a voltage time sequence of a grid-connected point of the wind power plant; the output data includes a time series of active and reactive power of the wind farm.
4. The method for dynamically modeling the equivalent of the wind farm based on the GRU-LSTM-FC combined network according to claim 1, wherein the step 3 is characterized in that the GRU-LSTM-FC combined network is trained by adopting data of a training set to obtain the GRU-LSTM-FC combined network model parameters, and specifically comprises the following steps:
(a) Construction of GRU-LSTM-FC combined network
The constructed combined network comprises 3 layers, wherein the 1 st layer is a GRU layer, the 2 nd layer is an LSTM layer, the 3 rd layer is an FC layer, and data sequentially passes through the GRU layer, the LSTM layer and the FC layer;
(b) Performance evaluation index of combined network
The method adopts a mean square error MSE index to evaluate the matching degree between the output and the actual response of the GRU-LSTM-FC combined network, and uses the absolute error of each time point as another index to describe the instantaneous error in the whole transient region, and is specifically as follows:
wherein y is i The actual response value of the wind power plant at the moment i;
the output of the GRU-LSTM-FC combined network at the moment i; n=1000, where N is the total number of data sampled during the whole time domain simulation process;
(c) Determination of the number of layers of a combined network
Performing performance evaluation on five networks, namely single-layer LSTM, single-layer GRU, double-layer LSTM, double-layer GRU and single-layer GRU-single-layer LSTM, respectively from the two aspects of iteration efficiency and mean square error of a training set and a testing set, wherein under the five conditions, the number of FC full-connection layers is uniformly set to be 1, and the number of neurons of all hidden layers is set to be 50; the iteration termination error of the combined network training is set to 10 -2 The maximum iteration number is 30000, and the model learning rate I r The inverse proportion function taking the iteration number epoch as an independent variable is expressed as follows:
wherein I is r0 ＝10 -2 Is the initial learning rate of the model; epoch is the number of iterations in the model training process;
according to the mean square error of the five networks in the training set and the testing set, the minimum error is obtained when the GRU is 1 layer and the LSTM is 1 layer, and the GRU is determined to be 1 layer and the LSTM is 1 layer according to the minimum error.
5. The method for dynamically modeling equivalent of wind farm based on GRU-LSTM-FC combined network according to claim 1, wherein in the step 3, the number of FC layers and the number of neurons of each layer in the GRU-LSTM-FC combined network are optimized by genetic algorithm, and an optimized GRU-LSTM-FC combined network model is obtained, which specifically comprises the following steps:
(a) Super-parameters to be optimized of combined network
The super parameters to be optimized comprise the number of FC layers and the number of each neuron of GRU, LSTM and FC;
(b) Super parameter optimization based on genetic algorithm
Carrying out population chromosome coding, and constructing chromosome C structures of individuals in the population according to the number of FC layers and the number of neurons of each hidden layer, wherein the chromosome C structures are as follows:
C＝[L FC ,N GRU ,N LSTM ,N FC1 ,...,N FCn ]
wherein L is FC The number of layers for FC; n (N) GRU For the number of GRU layer neurons, N LSTM For the number of LSTM layer neurons, N FC1 ,…,N FCn For the number of neurons in each layer of the FC layer,
designing an fitness function, comprehensively measuring the performance of the combined network according to three indexes of accuracy, generalization and training efficiency, and designing the fitness function J as the sum of the three indexes, wherein the expression is as follows:
wherein J is an fitness function; loss (Low Density) train0 And Loss of train For the mean square error and the total mean square error of the training set, the ratio of 2 is used for representing the accuracy; loss (Low Density) test0 And Loss of test The mean square error and the total mean square error of the test set are respectively, and the ratio of 2 is used for representing the generalization capability of the model; t (T) 0 For training time length, the value is the average time length of multiple training, T start And T end Training start and end times, respectively, in min, ratio T 0 /(T end -T start ) The training efficiency of the network is obtained; and alpha, beta and gamma are weight coefficients of the indexes respectively, and an optimized hyper-parameter result is obtained according to iteration termination conditions.
6. A wind power plant dynamic equivalent modeling system based on a GRU-LSTM-FC combined network is characterized by comprising:
modeling unit: based on a PSASP platform, constructing a target wind power plant containing n wind turbines, and accessing the target wind power plant into a CEPRI-36 node system after two-stage boosting;
simulation unit: simulating a target wind power plant according to wind speed data, wind direction data and grid-connected point voltage drop disturbance to obtain output power of the wind power plant, and constructing a training set and a testing set;
an optimizing unit: training the GRU-LSTM-FC combined network by adopting training set data, and determining the GRU layer number and the LSTM layer number of the GRU-LSTM-FC combined network through performance indexes; optimizing the number of FC layers and the number of neurons of each layer of the combined network by adopting a genetic algorithm to obtain optimized GRU-LSTM-FC combined network model parameters;
prediction unit: and according to input data of a certain wind speed, a certain wind direction and a certain voltage drop disturbance in the test set, predicting by adopting the optimized GRU-LSTM-FC combined network to obtain the active power and reactive power dynamic response time sequence of the wind power plant.
7. The GRU-LSTM-FC combined network-based wind farm dynamic equivalence modeling system according to claim 6, wherein in the modeling unit, a target wind farm is subjected to two-stage boosting and then is connected into a CEPRI-36 node system through a BUS30 BUS.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310081362.7A CN116306236A (en) | 2023-01-30 | 2023-01-30 | Wind power plant dynamic equivalent modeling method and system based on GRU-LSTM-FC combined network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310081362.7A CN116306236A (en) | 2023-01-30 | 2023-01-30 | Wind power plant dynamic equivalent modeling method and system based on GRU-LSTM-FC combined network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116306236A true CN116306236A (en) | 2023-06-23 |
Family
ID=86819449
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310081362.7A Pending CN116306236A (en) | 2023-01-30 | 2023-01-30 | Wind power plant dynamic equivalent modeling method and system based on GRU-LSTM-FC combined network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116306236A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117277314A (en) * | 2023-11-21 | 2023-12-22 | 深圳航天科创泛在电气有限公司 | Wind power prediction method and device, electronic equipment and readable storage medium |
-
2023
- 2023-01-30 CN CN202310081362.7A patent/CN116306236A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117277314A (en) * | 2023-11-21 | 2023-12-22 | 深圳航天科创泛在电气有限公司 | Wind power prediction method and device, electronic equipment and readable storage medium |
| CN117277314B (en) * | 2023-11-21 | 2024-03-08 | 深圳航天科创泛在电气有限公司 | Wind power prediction method and device, electronic equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109063276B (en) | Wind power plant dynamic equivalent modeling method suitable for long-time domain wind speed fluctuation | |
| CN103887815B (en) | Based on wind energy turbine set parameter identification and the Dynamic Equivalence of service data | |
| Hayes et al. | Equivalent power curve model of a wind farm based on field measurement data | |
| CN110334870B (en) | Photovoltaic power station short-term power prediction method based on gated cyclic unit network | |
| CN110009135B (en) | Wind power prediction method based on width learning | |
| CN107045574B (en) | SVR-based effective wind speed estimation method for low wind speed section of wind generating set | |
| CN109408849B (en) | Wind power plant dynamic equivalence method based on coherent unit grouping | |
| Wang et al. | Dynamic equivalent modeling for wind farms with DFIGs using the artificial bee colony with K-means algorithm | |
| CN105787592A (en) | Wind turbine generator set ultra-short period wind power prediction method based on improved RBF network | |
| CN115017787A (en) | Wind power plant voltage ride through characteristic equivalent modeling method and system based on intelligent algorithm | |
| CN112149905A (en) | Photovoltaic power station short-term power prediction method based on wavelet transformation and wavelet neural network | |
| CN116306236A (en) | Wind power plant dynamic equivalent modeling method and system based on GRU-LSTM-FC combined network | |
| CN116796644A (en) | Wind farm parameter identification method based on multi-agent SAC deep reinforcement learning | |
| CN112651576A (en) | Long-term wind power prediction method and device | |
| CN115392133A (en) | Wind power plant optimal clustering equivalence method adopting Gaussian mixture model | |
| CN111401792A (en) | Dynamic safety assessment method based on extreme gradient lifting decision tree | |
| Zhang et al. | Short-Term Power Prediction of Wind Power Generation System Based on Logistic Chaos Atom Search Optimization BP Neural Network | |
| CN109657380A (en) | A kind of double-fed fan motor field Dynamic Equivalence based on Extended Kalman filter | |
| CN115713029A (en) | Wind power plant stochastic model prediction optimization control method considering delay | |
| CN115719975A (en) | Wind power plant equivalent virtual inertia constant online evaluation method and device and storage medium | |
| CN111027816B (en) | Photovoltaic power generation efficiency calculation method based on data envelope analysis | |
| Chen et al. | Reward adaptive wind power tracking control based on deep deterministic policy gradient | |
| CN110674605A (en) | Fan power modeling method based on operation parameters | |
| CN111460596A (en) | Method for acquiring equivalent machine parameters under wind power plant multi-machine equivalence step by step | |
| CN117390418B (en) | Transient stability evaluation method, system and equipment for wind power grid-connected 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 |