CN115632406A - Reactive voltage control method and system based on digital-mechanism fusion drive modeling - Google Patents
Reactive voltage control method and system based on digital-mechanism fusion drive modeling Download PDFInfo
- Publication number
- CN115632406A CN115632406A CN202211645141.XA CN202211645141A CN115632406A CN 115632406 A CN115632406 A CN 115632406A CN 202211645141 A CN202211645141 A CN 202211645141A CN 115632406 A CN115632406 A CN 115632406A
- Authority
- CN
- China
- Prior art keywords
- reactive power
- reactive
- optimization
- power
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 62
- 230000004927 fusion Effects 0.000 title claims abstract description 36
- 238000005457 optimization Methods 0.000 claims abstract description 178
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims abstract description 9
- 238000010248 power generation Methods 0.000 claims description 66
- 238000011217 control strategy Methods 0.000 claims description 30
- 238000009826 distribution Methods 0.000 claims description 30
- 239000003990 capacitor Substances 0.000 claims description 20
- 238000013527 convolutional neural network Methods 0.000 claims description 20
- 230000006870 function Effects 0.000 claims description 20
- 238000003062 neural network model Methods 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 14
- 238000005286 illumination Methods 0.000 claims description 12
- 238000004088 simulation Methods 0.000 claims description 10
- 239000013598 vector Substances 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 241000036569 Carp sprivivirus Species 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000007619 statistical method Methods 0.000 claims description 4
- 241000288105 Grus Species 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000005096 rolling process Methods 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 10
- 230000001364 causal effect Effects 0.000 abstract description 2
- 238000010276 construction Methods 0.000 description 6
- 239000000243 solution Substances 0.000 description 6
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 241000135164 Timea Species 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000712 assembly Effects 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 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
- H02J3/12—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load
- H02J3/16—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load by adjustment of reactive power
-
- 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
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
-
- 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
- H02J2300/22—The renewable source being solar energy
- H02J2300/24—The renewable source being solar energy of photovoltaic origin
-
- 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
-
- 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
- H02J2300/40—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation wherein a plurality of decentralised, dispersed or local energy generation technologies are operated simultaneously
-
- 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
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/30—Reactive power compensation
Landscapes
- Engineering & Computer Science (AREA)
- Power Engineering (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Control Of Electrical Variables (AREA)
Abstract
The invention provides a reactive voltage control method and a system based on digital-mechanism fusion drive modeling, wherein the method comprises the following steps: in the off-line training stage, the reactive power optimization result based on the deterministic optimization model is used as input characteristics, the reactive power optimization result based on the stochastic optimization model is used as an output label, and the model containing the target neural network is trained. The method has causal knowledge with clear mechanism models and strong fitting capability of digital models, and can realize accurate and rapid online control of system voltage, thereby effectively relieving the voltage fluctuation problem caused by randomness of new energy and loads to a certain extent, and ensuring the operation safety and power supply reliability of a power grid.
Description
Technical Field
The invention belongs to the technical field of power system control, and particularly relates to a reactive voltage control method and system based on digital-mechanism fusion drive modeling.
Background
Under the influence of Distributed Generation (DG) and wide access of flexible loads, the characteristics and operation control of a power distribution network system are increasingly complex, and meanwhile, the randomness of fan power Generation, photovoltaic power Generation and loads also causes an increasingly serious voltage fluctuation problem, so that the problem becomes a key problem to be solved urgently in the construction of a novel power system. The traditional reactive voltage optimization control method is mainly based on an accurate power grid mechanism model, but the calculation precision and the calculation speed of the traditional reactive voltage optimization control method cannot meet the voltage control requirement of a novel power distribution system containing a large amount of new energy.
Disclosure of Invention
Aiming at the problems, the invention provides a reactive voltage control method based on digital-mechanism fusion drive modeling, which comprises the following steps:
in the off-line training stage, a reactive power optimization result based on a deterministic optimization model is used as an input feature, a reactive power optimization result based on a random optimization model is used as an output label, and a model containing a target neural network is trained.
Further, the reactive power optimization result based on the stochastic optimization model is generated by the following steps:
and performing reactive power optimization on historical observation data of the system by adopting a random optimization model based on a Latin hypercube sampling method to obtain a reactive power optimization control strategy corresponding to the historical observation data, and forming a first reactive power optimization strategy.
Further, a reactive power optimization result based on the deterministic optimization model is generated by the following method:
and performing reactive power optimization on historical observation data of the system by adopting a deterministic optimization model based on mixed integer second-order cone relaxation to obtain a corresponding reactive power optimization control strategy as a second reactive power optimization strategy.
Furthermore, the target neural network model adopts a model in which a convolutional neural network CNN is combined with a gating circulation unit GRU, namely a CNN-GRU model.
Further, the reactive power optimization result based on the stochastic optimization model is generated by the following steps:
according to historical observation data, estimating Beta distribution related parameters of illumination intensity, weibull distribution related parameters of wind speed and normal distribution related parameters of load power at a certain moment, taking an observation value of the load power at the moment as a mean value parameter of the normal distribution, and then obtaining probability distribution and related parameters of corresponding photovoltaic power generation output power and fan power generation output power at the moment according to an approximate relational expression between the photovoltaic power generation output power and the illumination intensity, and between the fan power generation output power and the wind speed;
obtaining random sample data of photovoltaic power generation output power, fan power generation output power and load power at the moment by adopting a Latin hypercube sampling method, sequentially carrying out reactive power optimization simulation on each group of sample data, carrying out statistical analysis on simulation results, and taking the gear of each on-load tap-changing transformer
Switchable capacitor bank gear
And SVC compensation power
As a reactive power control strategy derived from a stochastic optimization model at that moment
；
Will system
Respectively adopting a Latin hypercube sampling method to carry out random reactive power optimization on photovoltaic power generation output power, fan power generation output power and load power at each moment, storing a reactive power optimization control strategy obtained by solving at each moment into a historical strategy library to form a first reactive power optimization strategy
，
And representing a reactive power control strategy obtained by a random optimization model at the t-th moment.
Further, a reactive power optimization result based on the deterministic optimization model is generated by the following method:
system for connecting a plurality of mobile communication terminals
Historical observation data of photovoltaic power generation output power, fan power generation output power and load power at each moment are respectively subjected to reactive power optimization by adopting a deterministic optimization model based on mixed integer second-order cone relaxation to obtain a corresponding reactive power optimization control strategy
As a second reactive power optimization strategy.
Further, the method also comprises the following steps of:
acquiring prediction data of system operation;
obtaining a predictive reactive power optimization strategy based on a deterministic optimization model according to the predictive data;
and inputting the predicted reactive power optimization strategy into the target neural network model, and generating a modified reactive power optimization strategy after source load uncertainty is considered.
Further, the method comprises, in the online application phase:
step 1: forecasting data including photovoltaic power generation output power, fan power generation output power and load power of a system at a certain time in the future
, wherein ,
for the predicted value of the output power of the photovoltaic power generation,
the power generation output power of the fan is predicted value,
the predicted value of the load power is obtained;
step 2:will predict the data
Inputting a reactive power optimization model based on mixed integer second order cone relaxation to obtain the prediction data
Corresponding predictive reactive power optimization strategy
；
And step 3: forecasting reactive power optimization strategy of deterministic optimization model in step 2
Inputting the data into a target neural network model, and outputting a corrected reactive power optimization strategy under the condition of considering the uncertainty of a fan, a photovoltaic and a load
；
The target neural network model is a model CNN-GRU formed by combining a convolutional neural network CNN and a gating circulation unit GRU;
and 4, step 4: sending the modified reactive power optimization strategy to the power grid
And the method is used for simulating and deducing on line and calculating the power control parameters under the strategy.
Further, training the target neural network model includes:
the second reactive power optimization strategies output by the reactive power optimization model based on the mixed integer second-order cone relaxation are all represented by binary coding, including adopting the on-load tap-changing transformer gears
Bit binary representation and switchable capacitor bank gear adoption
Bit binary representation, SVC compensation power value adoption
Bit binary representation;
constructing a second reactive power optimization strategy in a binary coding form into a tensor form, and taking the channel number as 1, namely, inputting the data sample dimension of the CNN-GRU model as;
wherein ,
，
，
the numbers of the on-load tap changers, the switchable capacitor banks and the SVCs participating in reactive power control are respectively;
the input data is firstly subjected to multi-layer CNN extraction characteristics, then is subjected to flattening treatment to be used as input of multi-layer GRUs, and finally is output through a full connection layer, so that the relation between a second reactive power optimization strategy and a first reactive power optimization strategy is mapped and used for correcting a prediction reactive power control strategy.
Further, the CNN-GRU model is constructed in the following manner:
the CNN adopts a series structure of 3 groups of rolling layers and pooling layers;
after the input data is subjected to the operations of the convolutional layer and the pooling layer, the input flat layer is tiled into a one-dimensional vector, the extracted data abstract features are converted into global feature vectors, and the global feature vectors are used as GRU input;
GRU adopts 3 layer structure, and single GRU structure comprises update gate and reset gate, and its formula is:
in the formula ,
indicating that the current time step updated the door state,
indicating the current time step reset gate state,
indicates the state of the GRU at the current time step,
indicating the state of the GRU at the previous time step,
candidate states representing the current time step of the GRU for computation
，
An input representing the current time step of the GRU,
to input the weights of the layers to the hidden layer,
in order to imply the self-weight of the layer,
for input into the refresh door
The weight of (a) is calculated,
hiding the layer to the refresh gate for the previous time step
The weight of (a) is calculated,
to input to a reset gate
The weight of (a) is calculated,
hiding a layer to reset gate for a previous time step
Weight and sign of
Representing a Hadamard product;
and adding a discarding layer behind each GRU layer to prevent over-fitting of the network, and finally outputting a corrected reactive power optimization strategy considering the uncertainty of the fan, the photovoltaic and the load through a full connection layer.
Further, the CNN-GRU training adopts an Adam algorithm, and the weight updating formula is as follows:
in the formula ,
and
for the network weight parameter to be updated in the adjacent time step,
in order to smooth out the parameters of the image,
in order to obtain the learning rate of the learning,
as a function of loss
To pair
The partial derivative of (a) of (b),
and
the exponential decay rates of the first order moment estimate and the second order moment estimate respectively,
and
are respectively as
And
to the power of t of (a),
、
for gradients at t time steps respectively
The first order moment estimated value and the second order moment estimated value of the first order moment, similarly,
、
respectively corresponding values at t-1 time step,
、
taking into account the corrected deviations for the gradient at t time steps
A first moment estimate and a second moment estimate of;
the root mean square error function is a loss function of model training, and the formula is as follows:
in the formula ,
for the total number of reactive power control strategies at each time t stored in the strategy library,
as in the policy repository
A reactive power control strategy obtained by a random optimization model at a moment,
and outputting the corrected reactive power control strategy for the CNN-GRU model.
The invention also provides a reactive voltage control system based on digital-mechanism fusion drive modeling, which comprises an offline training module, wherein the offline training module is used for:
and taking a reactive power optimization result based on the deterministic optimization model as an input characteristic, taking a reactive power optimization result based on the random optimization model as an output label, and training the model containing the target neural network.
Further, the system also includes an online application module for:
acquiring prediction data of system operation;
obtaining a prediction reactive power optimization strategy based on a deterministic optimization model according to prediction data;
and inputting the predicted reactive power optimization strategy into the target neural network model, and generating a modified reactive power optimization strategy after source load uncertainty is considered.
The invention also provides a reactive voltage control system based on digital-mechanism fusion driving modeling, which comprises at least one processor and at least one memory;
the memory stores a computer program for executing the reactive voltage control method based on the digital-mechanism fusion driving modeling, and the processor calls the computer program in the memory to execute the reactive voltage control method based on the digital-mechanism fusion driving modeling.
According to the reactive voltage control method and system based on the digital-mechanism fusion drive modeling, the potential law among data is fully extracted by using an artificial intelligence method represented by deep learning, a model of a complementary fusion mechanism of a mechanism model and a digital model is established by adopting a guide mode, the causal knowledge of the mechanism model and the strong fitting capacity of the digital model are clear, the accurate and rapid online control of the system voltage is realized, the voltage fluctuation problem caused by the randomness of new energy and load is effectively relieved to a certain extent, and the operation safety and the power supply reliability of a power grid are ensured.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
FIG. 1 is a flow chart of a reactive voltage control method based on digital-mechanical fusion driving modeling according to an embodiment of the invention;
FIG. 2 shows a schematic diagram of a CNN-GRU model construction and training process according to an embodiment of the invention;
fig. 3 shows a schematic structural diagram of an on-load tap changer branch according to an embodiment of the invention;
FIG. 4 is a schematic diagram of a reactive voltage control system based on digital-mechanical fusion driving modeling according to an embodiment of the present invention;
fig. 5 shows a schematic structural diagram of another reactive voltage control system based on digital-mechanical fusion driving modeling according to an embodiment of the invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention provides a reactive voltage control method based on digital-mechanism fusion drive modeling, which can realize accurate and efficient reactive voltage control of an electric power system comprising distributed power generation equipment. Without loss of generality, the distributed power generation equipment of the embodiment of the invention mainly considers photovoltaic power generation equipment and fan power generation equipment. In other embodiments, other new energy power generation devices may be included.
As shown in fig. 1, the control method of the embodiment of the present invention includes an offline training phase and an online application phase.
In an off-line training stage, a mechanism model and a digital model are constructed by adopting a guide mode, on the basis of considering the wind turbine power generation, the photovoltaic power generation and the load uncertainty, a reactive power optimization result based on a deterministic optimization model is used as an input characteristic, a reactive power optimization result based on a stochastic optimization model is used as an output label, and a model containing a target neural network is trained.
The generated target network model is used for correcting the control strategy in the online application stage.
The deterministic reactive power optimization model is used as a mechanism driving part, and a deterministic optimization model based on mixed integer second-order cone relaxation is adopted. The random reactive power optimization model is used as a data driving part and adopts a reactive power optimization model based on a Latin hypercube sampling method. The target neural network adopts a convolutional neural network containing a cyclic structure, and adopts a CNN-GRU model in which a CNN (convolutional neural network) and a GRU (Gated Recurrent Unit) are combined with each other without loss of generality.
The training process of the off-line training phase is explained in detail below. The off-line training phase comprises:
step 1: according to a given example system, the following system parameters are input: the system comprises line parameters, a network topology structure, system node voltage and current safety limit values, an on-load tap changer access position and a maximum tap gear, a switchable capacitor bank installation position and a maximum switching gear, an SVC (static var compensator) installation position and a compensation power maximum limit value, and a photovoltaic power generation device, a fan power generation device and a load access position.
And 2, step: according to historical observation data, beta distribution related parameters of the illumination intensity at a certain time are estimated
And
weibull distribution of wind speed
And
and load power normal distribution related parameters
、
、
、
, wherein ,
、
respectively obtain the active power of the load at the momentHistorical observation values of power and reactive power; according to the approximate relational expression between the photovoltaic power generation output power and the illumination intensity as well as between the fan power generation output power and the wind speed, the probability distribution and the related parameters of the corresponding photovoltaic power generation output power and the fan power generation output power at the moment can be obtained;
and step 3: in order to fully consider the uncertainty of fan power generation, photovoltaic power generation and load, according to the related probability distribution and parameters obtained in the step 2, a random optimization model based on a Latin hypercube sampling method is adopted to carry out reactive power optimization simulation, all simulated results are subjected to statistical analysis, and the gear positions of all on-load tap-changing transformers are selected
Switchable capacitor bank gear
And SVC compensation power
As a reactive power control strategy derived from the stochastic optimization model at that time
；
And 4, step 4: according to step 3 pair system
Respectively adopting a Latin hypercube sampling method to carry out random reactive power optimization on photovoltaic power generation output power, fan power generation output power and load power at each moment, storing a reactive power optimization control strategy obtained by solving at each moment into a historical strategy library to form a first reactive power optimization strategy
，
Indicates none by random optimization at the t-th timeA power control strategy;
and 5: in the power system
Historical observation data of photovoltaic power generation output power, fan power generation output power and load data at each moment are respectively subjected to reactive power optimization by adopting a deterministic optimization model based on mixed integer second-order cone relaxation to obtain corresponding reactive power optimization control strategies
As a second reactive power optimization strategy;
step 6: establishing a CNN-GRU model structure of a target neural network;
and 7: the training modes for setting the CNN and the GRU adopt an Adam optimization algorithm, the loss function is a root mean square error function (RMSE), the learning rate is 0.001, and the input characteristic is a second reactive power optimization strategy obtained by a deterministic optimization model in the step 5
The output label is the first reactive optimization strategy at the corresponding moment in the historical strategy library
Training the CNN-GRU model parameters;
(2) The online application phase comprises:
acquiring prediction data of system operation;
obtaining a predictive reactive power optimization strategy based on a deterministic optimization model according to the predictive data;
and inputting the predicted reactive power optimization strategy into the target neural network model, and generating a corrected reactive power optimization strategy after source load uncertainty is considered.
Specifically, the method comprises the following steps:
step 1: forecasting data including photovoltaic power generation output power, fan power generation output power and load power of a system at a certain time in the future
, wherein ,
is a predicted value of the output power of the photovoltaic power generation,
the power generation output power of the fan is predicted value,
the predicted value of the load power is obtained;
step 2: will predict the data
Inputting a reactive power optimization model based on mixed integer second-order cone relaxation, (determinacy optimization model) to obtain the prediction data
Corresponding predictive reactive power optimization strategy
；
And step 3: forecasting reactive power optimization strategy of deterministic optimization model in step 2
Inputting the CNN-GRU model trained in the step 7 of the off-line training stage, and outputting a corrected reactive power optimization strategy under the condition of considering the uncertainty of the fan, the photovoltaic and the load
；
And 4, step 4: sending the corrected reactive power optimization strategy to the power grid
And performing on-line simulation deduction to calculate power control parameters including node voltage under the strategy
Branch active power
Sum branch reactive power
And fed back to the scheduling personnel for assisting decision making. The power distribution network adopts distributed computation, and online simulation deduction can be completed in a local server.
The construction and training process of the CNN-GRU model is shown in FIG. 2.
(1) Input data construction
1) The second reactive power optimization strategies output by the reactive power optimization model based on the mixed integer second-order cone relaxation are all represented by binary coding, namely the gears of the on-load tap-changing transformer adopt
Bit binary representation and switchable capacitor bank gear adoption
Bit binary representation, SVC compensation power value adoption
The representation of the bit binary is represented by a bit binary,
，
，
the values of the three-phase alternating current transformer are determined according to the maximum tap position of the on-load tap changing transformer, the maximum switching position of the switchable capacitor bank and the maximum SVC compensation power which are set in the practical calculation example;
2) In order to adapt to the input requirement of the CNN, a second reactive power optimization strategy in a binary coding form is constructed into a tensor form, the number of channels is taken as 1, and the dimensionality of a data sample input into the CNN-GRU model is taken as;
wherein ,
，
，
the numbers of the on-load tap changing transformers, the switchable capacitor banks and the SVCs participating in the reactive power control are respectively,
，
，
the meanings are as described above;
(2) CNN-GRU model construction
The method comprises the steps of firstly extracting features of input data through multiple layers of CNNs, then performing flattening processing to serve as input of multiple layers of GRUs, and finally obtaining output through a full connection layer, so that the relation between a second reactive power strategy obtained by a deterministic optimization model and a first reactive power strategy obtained by a random optimization model is mapped, and further the reactive power control strategy is corrected under the condition that uncertainty of a fan, photovoltaic and load is considered. The specific construction method of the CNN-GRU is as follows:
1) The CNN adopts a series structure of 3 groups of convolutional layers and pooling layers. Wherein, the number of convolution kernels in 3 groups of convolution layers is 16, 48 and 256 in sequence, and the sizes of the convolution kernels are all 3
1,3 groups of the pooling layers are all subjected to average pooling, and the size of each pool is 2
2；
2) After the operation of the convolution layer and the pooling layer, inputting a flat layer to be tiled into a one-dimensional vector, converting the extracted data abstract features into a global feature vector and inputting the global feature vector as GRU;
3) The GRU adopts a 3-layer structure, the number of neurons of an implicit layer in each layer of GRU is 256, 128 and 64 in sequence, and a single GRU structure consists of an update gate and a reset gate, and the formula is as follows:
in the formula ,
indicating that the current time step updated the door state,
indicating that the current time step resets the gate state,
indicating the state of the GRU at the current time step,
to representThe state of the GRU at the previous time step,
candidate states representing the current time step of the GRU for computation
，
An input representing the current time step of the GRU,
for the weights of the input layer to the hidden layer,
in order to imply the self-weight of the layer,
for input to the refresh door
The weight of (a) is calculated,
hiding a layer to the refresh gate for a previous time step
The weight of (a) is calculated,
to be input to a reset gate
The weight of (a) is calculated,
hiding a layer to reset gate for a previous time step
Weight and sign of
Representing a Hadamard product;
meanwhile, a discarding layer is added behind each GRU layer to prevent over-fitting of the network, the discarding rate is set to be 0.5, and finally a corrected reactive power optimization strategy considering the uncertainty of the fan, the photovoltaic and the load is obtained through the output of the full connection layer.
(3) CNN-GRU model training setup
The embodiment of the invention adopts an Adam algorithm to train the CNN layer and GRU layer models, and the weight updating formula is as follows:
in the formula ,
and
for the network weight parameters to be updated in the adjacent time steps (i.e. t time step and t +1 time step),
in order to smooth out the parameters of the image,
in order to obtain a learning rate,
as a function of loss
For is to
The partial derivative of (a) is,
and
the exponential decay rates of the first order moment estimate and the second order moment estimate respectively,
and
are respectively as
And
to the power of t of (a),
、
for gradients at t time steps respectively
The first order moment estimated value and the second order moment estimated value of the first order moment, similarly,
、
respectively corresponding values at t-1 time step,
、
taking into account the corrected deviations for the gradient at t time steps
First moment estimate and second moment estimate.
The root mean square error function (RMSE) is a loss function for model training, and its formula is:
in the formula ,
for the total number of reactive control strategies at each moment stored in the strategy library,
as in the policy repository
A reactive power control strategy obtained by a random optimization model at any moment, namely a reactive power optimization result after considering fan, photovoltaic and load fluctuation,
and outputting the modified reactive power control strategy for the CNN-GRU model.
The deterministic optimization model and the stochastic optimization model are explained in detail below.
The reactive power optimization model based on the mixed integer second order cone relaxation involved in the step 5 of the off-line training stage is as follows:
1.1 Determining an objective function
The reactive power optimization is mainly realized by adjusting the gear of the on-load tap changer (namely the tap gear of the on-load tap changer), the gear of the switchable capacitor bank, the compensation power of the static reactive power compensator and the like, on the premise of meeting the requirements of power flow constraint and power quality, the minimum active loss and the minimum voltage fluctuation of a network are realized, and the objective function is as follows:
in the formula ,
is a branch
The active power is lost,
is a node
The actual voltage is applied to the voltage-sensitive element,
for the reference node voltage, usually the root node is selected as the reference node;
the weighting factor of the voltage deviation is generally selected according to the magnitude relationship between the network loss and the voltage deviation term, and preferably, the two are made to have the same order of magnitude,
is the set of all nodes.
1.2 Constraint conditions
The constraint conditions comprise power balance constraint, branch load flow equation constraint, on-load tap changing transformer tap constraint, node injection power constraint, distributed power supply operation constraint, static Var Compensator (SVC) compensation power constraint, switchable capacitor bank operation constraint, safety constraint and the like. Respectively as follows:
power balance constraint
In the formula (II)
Is shown to all in
Aggregation of head-end nodes for branches to end nodes, aggregation
Is shown to all in
The set of end nodes of the branch that is the head-end node,
、
are respectively a branch
The active power and the reactive power of the head end,
、
are respectively a branch
The active power and the reactive power of the head end,
、
are respectively a branch
The resistance and the reactance of (a) is,
is a branch
The square of the magnitude of the current is,
is a node
Is the square of the voltage magnitude of (a).
Branch load flow equation constraints
A load flow equation without on-load tap changer branches:
in the formula ,
is a node
The square of the magnitude of the voltage of (c),
is a branch set which does not contain an on-load tap-changing transformer.
And (3) carrying out load flow equation constraint on the on-load tap changer branch:
at the nodes at two ends of the branch containing the on-load tap-changing transformer
、
Introducing virtual nodes between
As shown in fig. 3. The virtual node t is an access point of the on-load tap changer in the branch close to the node i, wherein the current flow direction of the branch is from the node i to the node j.
in the formula ,
for a set of branches containing an on-load tap changer,
is a transformation ratio of an on-load tap changer,
as virtual nodes
Is the square of the voltage magnitude of (a).
On-load tap changer tap restraint
in the formula ,
and
respectively the lower limit and the upper limit of the transformation ratio of the on-load tap changer;
the maximum gear of the on-load tap changing transformer tap joint;
the transformation ratio of each gear of the on-load tap changing transformer tap;
the tap is an actual gear of an on-load tap changer and belongs to an integer variable.
Because the load current equation constraint of the on-load tap changing transformer branch circuit and the on-load tap changing transformer tap constraint have non-convex constraint, the load current equation constraint and the on-load tap changing transformer tap constraint can be converted into convex constraint through accurate linear modeling of the on-load tap changing transformer tap. Considering that the tap positions of the on-load tap-changing transformer belong to integer variables, the tap positions can be represented by a binary method, as shown in formulas (19) and (20).
in the formula ,
is a 0-1 integer variable;
the maximum gear of the on-load tap changing transformer tap joint;
is constant and its value is determined by the number of binary digits required to represent the maximum tap position of the on-load tap changer.
Multiplying the equal sign of the formula (19) by
And define new variables
、
：
Further, formula (23) can be obtained:
meanwhile, the formula (24) can be obtained from the formula (16) and the formula (21)
Further, the formula (22) can be expressed by the formula (25) and the formula (26) by the large M method conversion:
in the large M method, M is an arbitrarily large (but not infinite) positive real constant that is manually assigned, and is typically determined according to the upper and lower bounds of a constraint variable.
Multiplying the two sides of the equation (23) by
And defining new variables
：
Combining formula (22), formula (24) and formula (27) to obtain formula (28)
The equation (27) is subjected to large-M conversion in the same manner, and the equations (29) to (30) are obtained.
Through accurate linear modeling of the on-load tap changing transformer, the load flow equation constraint of the on-load tap changing transformer branch and the non-convex constraint in the on-load tap changing transformer tap constraint can be converted into the convex constraint.
The node injection power constraints are:
in the formula ,
and
are respectively a node
The active and reactive power of the connected distributed power supplies,
and
is a node
The compensation reactive power of the connected discrete type and continuous type adjustable reactive power compensation devices;
and
is a node
The active power and the reactive power of the connected loads,
and
are respectively nodes
Active injected power and reactive injected power.
Distributed power supply operation constraints
in the formula ,
and
are respectively nodes
The upper limit of active power output and the upper limit of reactive power output of the connected distributed power supply;
is a collection of nodes connected to a distributed DG.
Switchable capacitor bank operation constraints
in the formula ,
is a node
The reactive power value corresponding to each grade of the connected switchable capacitor bank;
the maximum gear number of the switchable capacitor bank;
the actual gear of the switchable capacitor bank belongs to an integer variable;
is a collection of nodes connected to switchable capacitor banks.
Safety restraint
1) The node voltage constraint is:
in the formula ,
and
are respectively a node
The voltage magnitude is squared at the lower and upper limits,
is the set of all nodes.
2) The branch current constraints are:
in the formula ,
and
are respectively a branch
The square of the lower limit and the square of the upper limit of the current value,
is the set of all branches.
Cone constraint
The random reactive power optimization model based on the Latin hypercube sampling method is as follows:
(1) Photovoltaic power generation output power uncertainty model
The Beta distribution curve is adopted to fit a probability model of the illumination intensity distribution, and the probability density function can be expressed as:
in the formula ,
the ratio of the actual illumination intensity to the maximum illumination intensity in the time period is obtained;
and with
Two parameters for Beta distribution;
is a Gamma function.
The total output power of the photovoltaic array is:
in the formula ,
to account for the number of photovoltaic cell assemblies that make up the photovoltaic array,
and
respectively the area and the photoelectric conversion efficiency of a single photovoltaic cell assembly,
the total area of the photovoltaic array is,
for the overall photoelectric conversion efficiency of the photovoltaic array,
and the actual output power of the photovoltaic power generation in the time period.
According to the probability density function of the illumination intensity, the actual output power of the photovoltaic power generation in the period of time can be obtained
The probability density function is:
in the formula ,
is the same asAnd photovoltaic power generation maximum output power in a time period.
(2) Uncertainty model of fan power generation output power
Fitting actual wind speed with Weibull (Weibull) distribution
The probability model of the distribution, its probability density function, can be expressed as:
in the formula ,
and with
Respectively the shape parameter and the scale parameter of the weibull distribution,
representing the actual wind speed at the hub of the wind turbine.
The relationship between the output power and the wind speed of the wind generating set is as follows:
in the formula ,
represents the actual output power of the wind power generation,
represents the rated output power of the wind power generation,
，
，
the rated wind speed, the cut-in wind speed and the cut-out wind speed of the wind turbine generator system are respectively represented.
Actual wind speed at the hub of the fan
Is substantially maintained at
And with
Thus the actual output power of wind power generation
The probability density function is:
(3) Load power uncertainty optimization model
Load active power
And reactive power
The probability model is fitted by a normal distribution curve, and the probability density function can be expressed as:
in the formula ,
and
respectively the mean value and the standard deviation of the active power of the load;
and
mean and standard deviation of the reactive power of the load, respectively.
(4) Random reactive power optimization model based on Latin hypercube sampling method
Step 1: according to historical observation data, beta distribution related parameters of the illumination intensity at a certain time are estimated
And
weibull distribution of wind speed
And
and load power normal distribution related parameters
、
、
、
, wherein ,
、
respectively taking historical observation values of load active power and reactive power at the moment;
step 2: according to the approximate relational expression between the photovoltaic power generation output power and the illumination intensity as well as between the fan power generation output power and the wind speed, the probability distribution and the related parameters of the corresponding photovoltaic power generation output power and the fan power generation output power at the moment can be obtained;
and step 3: obtaining the relevant probability distribution and parameters obtained in the step 2 by adopting a Latin hypercube sampling method
Grouping random sample data of photovoltaic power, fan power generation output power and load power;
and 4, step 4: taking each group of random sample data obtained by sampling in the step 3 as a known scene, and performing deterministic solution under each scene by respectively adopting a reactive power optimization model based on mixed integer second-order cone relaxation to obtain a reactive power optimization strategy under the generated scene, wherein the reactive power optimization model based on mixed integer second-order cone relaxation is detailed in the part 1 in the specific implementation mode, and is not repeated here;
and 5: performing statistical analysis on the simulation results, and respectively calculating the obtained results
The mathematical expectation of on-load tap changing transformer tap position, switchable capacitor bank switching position and SVC compensation power in the group reactive power optimization strategy is as follows:
in the formula ,
the number of Latin hypercube sampling, namely the total simulation times;
(
=1,2,…, 
) Is the system one
The result obtained by the secondary random simulation;
representative sample results
Is a mathematical expectation of
Large enough to be used as an unbiased estimate of the true value of the system result. Therefore, the influence of the randomness factors on the system can be reflected.
And the tap gear of the on-load tap changing transformer, the switching gear of the switchable capacitor and the expected value of the SVC compensation power are used as a final reactive power control strategy, so that a system reactive power optimization control scheme under the consideration of the uncertainty of a fan, photovoltaic and load is realized.
Based on the same inventive concept, an embodiment of the present invention further provides a reactive voltage control system based on digital-mechanism fusion driving modeling, as shown in fig. 4, the system includes an offline training module, configured to:
and taking a reactive power optimization result based on the deterministic optimization model as an input characteristic, taking a reactive power optimization result based on the random optimization model as an output label, and training the model containing the target neural network.
Further, the system also includes an online application module for: acquiring prediction data of system operation; obtaining a predictive reactive power optimization strategy based on a deterministic optimization model according to the predictive data; and inputting the predicted reactive power optimization strategy into a target neural network model trained by an offline training module, and generating a corrected reactive power optimization strategy after source load uncertainty is considered.
The specific implementation of the offline training module and the online application module can be obtained according to the method for controlling reactive voltage based on digital-mechanism fusion drive modeling in the embodiment of the invention, and is not described again.
Based on the same inventive concept, as shown in fig. 5, an embodiment of the present invention further provides another reactive voltage control system based on digital-mechanism fusion driving modeling, which includes at least one processor and at least one memory, where the memory stores a computer program for executing the reactive voltage control method based on digital-mechanism fusion driving modeling in the embodiment of the present invention, and the processor is configured to call the computer program in the memory to execute the reactive voltage control method based on digital-mechanism fusion driving modeling. Fig. 5 illustrates an example of a memory and a processor.
The reactive voltage control method and system based on the digital-mechanism fusion drive modeling, disclosed by the invention, adopt a model of a complementary fusion mechanism of the mechanism model and the digital model, have both definite cause and effect knowledge of the mechanism model and strong fitting capability of the digital model, and realize accurate and rapid online control of system voltage, so that the problem of voltage fluctuation caused by randomness of new energy and load is effectively relieved to a certain extent, and the operation safety and the power supply reliability of a power grid are ensured.
Although the present invention has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (14)
1. A reactive voltage control method based on digital-mechanism fusion drive modeling is characterized by comprising the following steps:
in the off-line training stage, a reactive power optimization result based on a deterministic optimization model is used as an input feature, a reactive power optimization result based on a random optimization model is used as an output label, and a model containing a target neural network is trained.
2. The reactive voltage control method based on the digital-mechanism fusion driving modeling is characterized in that the reactive power optimization result based on the stochastic optimization model is generated by the following method:
and performing reactive power optimization on historical observation data of the system by adopting a random optimization model based on a Latin hypercube sampling method to obtain a reactive power optimization control strategy corresponding to the historical observation data, and forming a first reactive power optimization strategy.
3. The reactive voltage control method based on the digital-mechanical fusion driving modeling according to claim 2, wherein the reactive power optimization result based on the deterministic optimization model is generated by:
and performing reactive power optimization on the historical observation data of the system by adopting a deterministic optimization model based on mixed integer second-order cone relaxation to obtain a corresponding reactive power optimization control strategy, which is used as a second reactive power optimization strategy.
4. The reactive voltage control method based on digital-mechanical fusion driving modeling according to claim 3,
the target neural network model adopts a model that a convolutional neural network CNN is combined with a gating circulation unit GRU, namely a CNN-GRU model.
5. The reactive voltage control method based on the digital-mechanism fusion driving modeling is characterized in that the reactive power optimization result based on the stochastic optimization model is generated by the following method:
according to historical observation data, estimating Beta distribution related parameters of illumination intensity, weibull distribution related parameters of wind speed and normal distribution related parameters of load power at a certain moment, taking an observation value of the load power at the moment as a mean value parameter of the normal distribution, and then obtaining probability distribution and related parameters of corresponding photovoltaic power generation output power and fan power generation output power at the moment according to an approximate relational expression between the photovoltaic power generation output power and the illumination intensity, and between the fan power generation output power and the wind speed;
obtaining random sample data of photovoltaic power generation output power, fan power generation output power and load power at the moment by adopting a Latin hypercube sampling method, sequentially carrying out reactive power optimization simulation on each group of sample data, carrying out statistical analysis on simulation results, and taking the gear of each on-load tap-changing transformer
Switchable capacitor bank gear
And SVC compensation power
As a reactive power control strategy derived from a stochastic optimization model at that moment
；
Will system
The photovoltaic power generation output power, the fan power generation output power and the load power at each moment are respectivelyRandom reactive power optimization is carried out by adopting a Latin hypercube sampling method, and reactive power optimization control strategies obtained by solving at each moment are stored into a historical strategy library to form a first reactive power optimization strategy
，
And representing the reactive power control strategy obtained by the random optimization model at the t-th moment.
6. The reactive voltage control method based on the digital-mechanical fusion driving modeling is characterized in that the reactive power optimization result based on the deterministic optimization model is generated by the following method:
will system
Historical observation data of photovoltaic power generation output power, fan power generation output power and load power at each moment are respectively subjected to reactive power optimization by using a deterministic optimization model based on mixed integer second-order cone relaxation to obtain corresponding reactive power optimization control strategies
As a second reactive power optimization strategy.
7. The reactive voltage control method based on digital-mechanical fusion driving modeling according to any one of claims 1-6, further comprising, during an online application phase:
acquiring prediction data of system operation;
obtaining a predictive reactive power optimization strategy based on a deterministic optimization model according to the predictive data;
and inputting the predicted reactive power optimization strategy into the target neural network model, and generating a corrected reactive power optimization strategy after source load uncertainty is considered.
8. The reactive voltage control method based on digital-mechanical fusion driving modeling according to claim 7, characterized by comprising in an online application phase:
step 1: forecasting data including photovoltaic power generation output power, fan power generation output power and load power of a system at a certain time in the future
, wherein ,
for the predicted value of the output power of the photovoltaic power generation,
the power generation output power of the fan is predicted value,
the predicted value of the load power is obtained;
and 2, step: will predict the data
Inputting a reactive power optimization model based on mixed integer second order cone relaxation to obtain the prediction data
Corresponding predictive reactive power optimization strategy
；
And step 3: forecasting reactive power optimization strategy of deterministic optimization model in step 2
Inputting the data into a target neural network model, and outputting a corrected reactive power optimization strategy under the condition of considering the uncertainty of a fan, a photovoltaic and a load
；
The target neural network model is a model CNN-GRU formed by combining a convolutional neural network CNN and a gating circulation unit GRU;
and 4, step 4: sending the modified reactive power optimization strategy to the power grid
And the method is used for simulating and deducing on line and calculating the power control parameters under the strategy.
9. The reactive voltage control method based on the digital-mechanism fusion driving modeling according to claim 4, wherein training a target neural network model comprises:
the second reactive power optimization strategies output by the reactive power optimization model based on the mixed integer second-order cone relaxation are all represented by binary coding, including adopting the on-load tap-changing transformer gears
Bit binary representation and switchable capacitor bank gear adoption
Bit binary representation, SVC compensation power value adoption
Bit binary representation;
constructing a second reactive power optimization strategy in a binary coding form into a tensor form, and taking the channel number as 1, namely, inputting the data sample dimension of the CNN-GRU model as;
wherein ,
，
，
the numbers of the on-load tap changers, the switchable capacitor banks and the SVCs participating in reactive power control are respectively;
the input data is firstly subjected to multi-layer CNN extraction characteristics, then is subjected to flattening treatment to be used as input of multi-layer GRUs, and finally is output through a full connection layer, so that the relation between a second reactive power optimization strategy and a first reactive power optimization strategy is mapped and used for correcting a prediction reactive power control strategy.
10. The reactive voltage control method based on the digital-mechanism fusion driving modeling is characterized in that the CNN-GRU model is constructed in the following way:
the CNN adopts a series structure of 3 groups of rolling layers and pooling layers;
after the input data is subjected to the operations of the convolutional layer and the pooling layer, the input flat layer is tiled into a one-dimensional vector, the extracted data abstract features are converted into global feature vectors, and the global feature vectors are used as GRU input;
GRU adopts 3 layer structure, and single GRU structure comprises update gate and reset gate, and its formula is:
in the formula ,
indicating that the current time step updated the door state,
indicating that the current time step resets the gate state,
indicating the state of the GRU at the current time step,
indicating the state of the GRU at the previous time step,
candidate states representing the current time step of the GRU for computation
，
An input representing the current time step of the GRU,
for the weights of the input layer to the hidden layer,
in order to imply the weight of the layer itself,
for input to the refresh door
The weight of (a) is calculated,
hiding the layer to the refresh gate for the previous time step
The weight of (a) is calculated,
to input to a reset gate
The weight of (a) is calculated,
hiding a layer to reset gate for a previous time step
Weight and symbol of
Representing a Hadamard product;
and adding a discarding layer behind each GRU layer to prevent over-fitting of the network, and finally outputting a corrected reactive power optimization strategy considering the uncertainty of the fan, the photovoltaic and the load through a full connection layer.
11. The reactive voltage control method based on the digital-mechanism fusion driving modeling is characterized in that Adam algorithm is adopted in CNN-GRU training, and the weight updating formula is as follows:
in the formula ,
and
for the network weight parameter to be updated in the adjacent time step,
in order to smooth out the parameters of the image,
in order to obtain a learning rate,
as a function of loss
To pair
The partial derivative of (a) of (b),
and
the exponential decay rates of the first order moment estimate and the second order moment estimate respectively,
and
are respectively as
And
to the power of t of (a),
、
for gradients at t time steps respectively
The first order moment estimated value and the second order moment estimated value of the first order moment, similarly,
、
respectively corresponding values at t-1 time step,
、
taking into account the corrected deviations for the gradient at t time steps
A first moment estimate and a second moment estimate of;
the root mean square error function is a loss function of model training, and the formula is as follows:
in the formula ,
for the total number of reactive control strategies at each time t stored in the strategy library,
in a policy repository
A reactive power control strategy obtained by a random optimization model at a moment,
and outputting the modified reactive power control strategy for the CNN-GRU model.
12. A reactive voltage control system based on digital-mechanism fusion drive modeling is characterized by comprising an offline training module, a reactive voltage control module and a reactive voltage control module, wherein the offline training module is used for:
and taking a reactive power optimization result based on the deterministic optimization model as an input characteristic, taking a reactive power optimization result based on the random optimization model as an output label, and training the model containing the target neural network.
13. The digital-mechanical fusion drive modeling based reactive voltage control system of claim 12, further comprising an online application module for:
acquiring prediction data of system operation;
obtaining a prediction reactive power optimization strategy based on a deterministic optimization model according to prediction data;
and inputting the predicted reactive power optimization strategy into the target neural network model, and generating a modified reactive power optimization strategy after source load uncertainty is considered.
14. A reactive voltage control system based on digital-mechanical fusion drive modeling, the system comprising at least one processor and at least one memory;
the memory stores a computer program for performing the method of any of claims 1-11, and the processor calls the computer program in the memory to perform the method of any of claims 1-11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211645141.XA CN115632406B (en) | 2022-12-21 | 2022-12-21 | Reactive voltage control method and system based on digital-mechanism fusion driving modeling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211645141.XA CN115632406B (en) | 2022-12-21 | 2022-12-21 | Reactive voltage control method and system based on digital-mechanism fusion driving modeling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115632406A true CN115632406A (en) | 2023-01-20 |
| CN115632406B CN115632406B (en) | 2023-04-28 |
Family
ID=84911157
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211645141.XA Active CN115632406B (en) | 2022-12-21 | 2022-12-21 | Reactive voltage control method and system based on digital-mechanism fusion driving modeling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115632406B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117446670A (en) * | 2023-12-25 | 2024-01-26 | 泰安市特种设备检验研究院 | Automatic control method and system for tower crane based on man-machine co-fusion |
| CN117955121A (en) * | 2024-03-27 | 2024-04-30 | 国网吉林省电力有限公司长春供电公司 | Power distribution network reactive voltage control method based on data driving and source load uncertainty |
| CN117955121B (en) * | 2024-03-27 | 2024-06-28 | 国网吉林省电力有限公司长春供电公司 | Power distribution network reactive voltage control method based on data driving and source load uncertainty |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112039070A (en) * | 2020-09-09 | 2020-12-04 | 华北电力大学 | Hierarchical network reconstruction method and system combining data and model hybrid drive |
| CN112821424A (en) * | 2021-01-29 | 2021-05-18 | 国网辽宁省电力有限公司大连供电公司 | Power system frequency response analysis method based on data-model fusion drive |
| CN113553768A (en) * | 2021-07-27 | 2021-10-26 | 天津大学 | Method and device for rapidly calculating reliability of power grid driven by model data in hybrid mode |
| CN113972667A (en) * | 2021-11-29 | 2022-01-25 | 昆明理工大学 | Power distribution network reactive power optimization method based on one-dimensional convolutional neural network |
| CN115329911A (en) * | 2022-10-14 | 2022-11-11 | 国网江苏省电力有限公司电力科学研究院 | Safety correction method for UPFC-containing power system based on SAE two-classification model |
-
2022
- 2022-12-21 CN CN202211645141.XA patent/CN115632406B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112039070A (en) * | 2020-09-09 | 2020-12-04 | 华北电力大学 | Hierarchical network reconstruction method and system combining data and model hybrid drive |
| CN112821424A (en) * | 2021-01-29 | 2021-05-18 | 国网辽宁省电力有限公司大连供电公司 | Power system frequency response analysis method based on data-model fusion drive |
| CN113553768A (en) * | 2021-07-27 | 2021-10-26 | 天津大学 | Method and device for rapidly calculating reliability of power grid driven by model data in hybrid mode |
| CN113972667A (en) * | 2021-11-29 | 2022-01-25 | 昆明理工大学 | Power distribution network reactive power optimization method based on one-dimensional convolutional neural network |
| CN115329911A (en) * | 2022-10-14 | 2022-11-11 | 国网江苏省电力有限公司电力科学研究院 | Safety correction method for UPFC-containing power system based on SAE two-classification model |
Non-Patent Citations (3)
| Title |
|---|
| 王志杨 等: "数据与模型混合驱动的区域综合能源系统 双层优化调度决策方法" * |
| 蔡昌春等: "一种基于数据驱动的配电网无功优化" * |
| 邵美阳;吴俊勇;石琛;安然;朱孝文;黄杏;蔡蓉;: "基于数据驱动和深度置信网络的配电网无功优化" * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117446670A (en) * | 2023-12-25 | 2024-01-26 | 泰安市特种设备检验研究院 | Automatic control method and system for tower crane based on man-machine co-fusion |
| CN117446670B (en) * | 2023-12-25 | 2024-04-26 | 泰安市特种设备检验研究院 | Automatic control method and system for tower crane based on man-machine co-fusion |
| CN117955121A (en) * | 2024-03-27 | 2024-04-30 | 国网吉林省电力有限公司长春供电公司 | Power distribution network reactive voltage control method based on data driving and source load uncertainty |
| CN117955121B (en) * | 2024-03-27 | 2024-06-28 | 国网吉林省电力有限公司长春供电公司 | Power distribution network reactive voltage control method based on data driving and source load uncertainty |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115632406B (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112186743B (en) | Dynamic power system economic dispatching method based on deep reinforcement learning | |
| Li et al. | Learning to operate distribution networks with safe deep reinforcement learning | |
| Cao et al. | Model-free voltage control of active distribution system with PVs using surrogate model-based deep reinforcement learning | |
| Khan et al. | Adopting Scenario-Based approach to solve optimal reactive power Dispatch problem with integration of wind and solar energy using improved Marine predator algorithm | |
| Li et al. | An intelligent transient stability assessment framework with continual learning ability | |
| Zhang et al. | An efficient multi-agent negotiation algorithm for multi-period photovoltaic array reconfiguration with a hydrogen energy storage system | |
| CN114006370B (en) | Power system transient stability analysis and evaluation method and system | |
| CN113489015A (en) | Power distribution network multi-time scale reactive voltage control method based on reinforcement learning | |
| CN115374692B (en) | Double-layer optimization scheduling decision method for regional comprehensive energy system | |
| CN115632406B (en) | Reactive voltage control method and system based on digital-mechanism fusion driving modeling | |
| Huang et al. | Resilient power network structure for stable operation of energy systems: A transfer learning approach | |
| Du et al. | Applying deep convolutional neural network for fast security assessment with N-1 contingency | |
| CN112561273A (en) | Active power distribution network renewable DG planning method based on improved PSO | |
| CN113872213B (en) | Autonomous optimization control method and device for power distribution network voltage | |
| CN113344283B (en) | Energy internet new energy consumption capability assessment method based on edge intelligence | |
| Yan et al. | Multi-Agent Safe Graph Reinforcement Learning for PV Inverters-Based Real-Time Decentralized Volt/Var Control in Zoned Distribution Networks | |
| Vohra et al. | End-to-end learning with multiple modalities for system-optimised renewables nowcasting | |
| CN111799820A (en) | Double-layer intelligent hybrid zero-star cloud energy storage countermeasure regulation and control method for power system | |
| Yang et al. | Cloud-edge-end intelligence for fault-tolerant renewable energy accommodation in smart grid | |
| CN115018179A (en) | CNN-BiGRU-based ultra-short-term power load prediction method | |
| CN112101651B (en) | Electric energy network coordination control method, system and information data processing terminal | |
| CN115276067A (en) | Distributed energy storage voltage adjusting method adaptive to topological dynamic change of power distribution network | |
| Madeiro et al. | Gradient-based algorithms for the automatic construction of fuzzy cognitive maps | |
| Sun et al. | Reliability prediction of distribution network based on PCA-GA-BP neural network | |
| Yu et al. | Deep reinforcement learning for coordinated voltage regulation in active distribution networks |
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 |