CN116760055A - Dynamic reactive compensation method based on neural network - Google Patents
Dynamic reactive compensation method based on neural network Download PDFInfo
- Publication number
- CN116760055A CN116760055A CN202310670249.2A CN202310670249A CN116760055A CN 116760055 A CN116760055 A CN 116760055A CN 202310670249 A CN202310670249 A CN 202310670249A CN 116760055 A CN116760055 A CN 116760055A
- Authority
- CN
- China
- Prior art keywords
- representing
- reactive compensation
- feature
- features
- output
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 36
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000004364 calculation method Methods 0.000 claims abstract description 25
- 239000011159 matrix material Substances 0.000 claims description 28
- 230000004913 activation Effects 0.000 claims description 12
- 239000013598 vector Substances 0.000 claims description 9
- 238000009825 accumulation Methods 0.000 claims description 8
- 230000004044 response Effects 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 3
- 206010027175 memory impairment Diseases 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 230000010354 integration Effects 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 3
- 239000010410 layer Substances 0.000 description 71
- 230000006870 function Effects 0.000 description 15
- 230000001105 regulatory effect Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 239000002356 single layer Substances 0.000 description 1
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
- H02J3/18—Arrangements for adjusting, eliminating or compensating reactive power in networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/30—Reactive power compensation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Power Engineering (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Control Of Electrical Variables (AREA)
Abstract
The invention relates to the technical field of reactive power compensation, and discloses a dynamic reactive power compensation method based on a neural network, which comprises the following steps of: step 101, generating initial characteristics and operation parameter characteristics for reactive compensation equipment; step 102, generating unified features with dimension N based on the initial features; step 103, inputting the unified characteristic and the operation parameter characteristic into a compensation distribution calculation model to calculate the compensation power or output proportion of each reactive compensation device; the compensation distribution calculation model of the invention demands the connection of the reactive compensation equipment and the reactive compensation equipment from the reactive compensation network and the integral characteristic of the reactive compensation network through the attention mechanism, realizes the unified dispatching of the reactive compensation equipment of the power grid system and improves the stability of the power grid system.
Description
Technical Field
The invention relates to the technical field of reactive power compensation, in particular to a dynamic reactive power compensation method based on a neural network.
Background
A representative reactive compensation device includes: reactive power regulating equipment of a static voltage regulating means is difficult to meet the requirement of the system in a rapid change of the operation mode due to discontinuous regulation and slow response speed. The static reactive compensation device has a very fast response speed, but cannot provide required reactive support when the voltage is low due to the constant impedance characteristic, so that the capability of coping with emergency is weak, a filter is required to be arranged for suppressing harmonic waves, the occupied area is large, and in addition, excessive SVC devices are easy to cause system oscillation. A single reactive compensation device is not generally used in a power grid system, and in the prior art, dynamic adjustment is generally performed on a single reactive compensation device or reactive compensation unit, and systematic unified scheduling is not performed on a power grid system comprising reactive compensation devices of different types.
Disclosure of Invention
The invention provides a dynamic reactive power compensation method based on a neural network, which solves the technical problem that a power grid system comprising reactive power compensation equipment of different types lacks systematic unified scheduling in the related art.
The invention provides a dynamic reactive power compensation method based on a neural network, which comprises the following steps:
step 101, generating initial characteristics and operation parameter characteristics for reactive compensation equipment;
step 102, generating unified features with dimension N based on the initial features;
step 103, inputting the unified characteristic and the operation parameter characteristic into a compensation distribution calculation model to calculate the compensation power or output proportion of each reactive compensation device;
the compensation allocation calculation model includes: the system comprises an accumulation layer, an attention layer, a second hiding layer, a third hiding layer, a fourth hiding layer, a sequence hiding layer and a second classifier, wherein the accumulation layer inputs unified characteristics and operation parameter characteristics and outputs first characteristics;
the first feature inputs attention layers, the attention layers comprise a first attention unit, a second attention unit and a third attention unit, the first attention unit, the second attention unit and the third attention unit respectively input first features, the first attention unit, the second attention unit and the third attention unit respectively output second features, third features and fourth features, the second features and the third features are input into second hidden layers, the second hidden layers output fifth features, the fifth features and the fourth features are input into third hidden layers, the third hidden layers output sixth features, the sixth features and the operating parameter features are input into fourth hidden layers, the fourth hidden layers output seventh features, the seventh features are input into sequence hidden layers, the sequence hidden layers comprise M LSTM layers, each LSTM layer comprises H LSTM units connected in series, the LSTM units of the Mth layer of the sequence hidden layers are respectively connected with a second classifier, and the labels of the j second classifier represent the compensation power or output proportion of the j reactive compensation equipment in the next control period.
Further, the dimensions of the initial characteristics include reactive compensation equipment ID, power capacity, rated voltage, lower regulation range limit, upper regulation range limit, response speed, direct link rated voltage.
Further, the operation parameter characteristics comprise the power factor of a direct connection line (reactive compensation equipment), the content of a voltage fundamental wave component, the content of 3-h time voltage harmonic wave components of the direct connection line, the content of a current fundamental wave component of the direct connection line and the content of 3-h time current harmonic wave components of the direct connection line; wherein h is more than or equal to 3.
Further, the method for generating the unified feature comprises the following steps:
the initial features are input into a first neural network, the first neural network comprises a first hidden layer, the first hidden layer inputs the initial features, and the output of the first hidden layer unifies the features.
Further, when training the first neural network, the output of the first hidden layer of the first neural network is connected to a first classifier, the number of classification labels of the first classifier is the same as the number of reactive compensation devices, and one classification label of the first classifier corresponds to the reactive compensation device.
Further, training the first neural network by constructing a training set of initial characteristics of all reactive compensation equipment, wherein the weight parameters of the trained first neural network are fixed.
Further, the integration layer mixes the unified feature and the operation parameter feature, and the method for mixing the unified feature and the operation parameter feature is to splice the unified feature and the operation parameter feature, for example splice the tail end of the unified feature vector into the operation parameter feature vector.
Further, the set of classification labels of the jth second classifier is denoted as a= { a 1 ,A 2 …A k Value range [0, s ] representing the output ratio]And performing discretization to generate k point values, and mapping the k point values with k classification labels of the set A respectively.
Further, the internal operation of the first attention unit includes:
c i =σ(x i W xc +b c ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xc Representing weight parameters, b c Represents the bias parameter, σ represents the activation function, c i Representing a second feature;
the internal operations of the second attention unit include:
d i =σ(x i W xd +b d ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xd Representing weight parameters, b d Represents the bias parameter, σ represents the activation function, d i Representing a third feature;
the internal operations of the third attention unit include:
e i =σ(x i W xe +b e ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xe Representing weight parameters, b d Represents the bias parameter, sigma represents the activation function, e i Representing a fourth feature;
the internal operation of the first hidden layer includes:
wherein c i Representing the ith reactive compensationSecond feature of the apparatus, d j Indicating the third characteristic of the jth reactive compensation device, +. j Dimension e of j A fourth feature representing a j-th reactive compensation device, M being equal to the number of reactive compensation devices, the summation representing the addition of vectors, a i A sixth feature representing an ith reactive compensation equipment;
wherein c i Representing a second characteristic, d, of the ith reactive compensation equipment q Representing a third characteristic, d, of the q-th reactive compensation device j A third feature representing a j-th reactive compensation device, M being equal to the number of reactive compensation devices;
the internal operations of the fourth hidden layer include:
wherein a is i A sixth feature representing an ith reactive compensation device,/->An operating parameter characteristic representing the nth control period of the ith reactive compensation equipment, +.>A seventh feature representing an nth control period of the ith reactive compensation equipment.
Further, the internal operations of the sequence hiding layer include:
the operation process of the nth layer and the t-th LSTM unit is as follows:
definition:i=t,h t-1 representing the status of delivery->h 0 =0,/> Output of the t-th LSTM cell representing the n-1 th hidden layer, +.>An output of the t-1 th LSTM cell representing the nth hidden layer;
forgetting door f t The calculation formula of (2) is as follows:
wherein W is fx Representation->Transfer to f t The corresponding weight matrix is used to determine the weight matrix,representing the transfer state h t-1 Transfer to f t Corresponding weight matrix, b f Representing bias terms, σ representing a sigmoid function;
input gate i t The calculation formula of (2) is as follows:
wherein W is xi Representation input +.>Transfer to i t Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to i t Corresponding weight matrix,b i Representing bias terms, σ representing a sigmoid function;
intermediate stateCan be expressed as follows:
wherein W is xC Representation input +.>Transfer to->Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to->Corresponding weight matrix, b C Representing a bias term, tanh representing an activation function tanh;
output stateExpressed by the following formula:
wherein the method comprises the steps ofIs the output state transmitted by the t-1 LSTM of the nth layer, f t 、i t 、/>Is the result of calculation of forget gate, input gate, intermediate state, < >>Door f for indicating forgetfulness t And the output state of the t-1 th LSTM of the nth layer +.>Point-by-point multiplication is performed to make->Indicating input/output gate i t And intermediate state->Performing point-by-point multiplication;
the output gate is expressed as:
wherein W is xo Representation input +.>Transfer to o t Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to o t Corresponding weight matrix, b o Representing bias terms, σ representing a sigmoid function;
output ofCan be expressed as follows:
will output door o t And tan h (C) t ) Multiplying point by point to obtain the output of the current LSTM unit
The invention has the beneficial effects that:
the compensation distribution calculation model of the invention demands the connection of the reactive compensation equipment and the reactive compensation equipment from the reactive compensation network and the integral characteristic of the reactive compensation network through the attention mechanism, realizes the unified dispatching of the reactive compensation equipment of the power grid system and improves the stability of the power grid system. Meanwhile, the unified characteristics are not updated when the reactive compensation network is unchanged, so that the output of the attention layer of the compensation distribution calculation model is not updated after being initialized, the whole reactive compensation network has good calculation efficiency under the condition of retaining a complex expression capability framework, the millisecond-level calculation speed can be achieved, and the low-interval dynamic reactive compensation control is provided.
Drawings
Fig. 1 is a flow chart of a dynamic reactive power compensation method based on a neural network of the present invention.
Detailed Description
The subject matter described herein will now be discussed with reference to example embodiments. It is to be understood that these embodiments are merely discussed so that those skilled in the art may better understand and implement the subject matter described herein and that changes may be made in the function and arrangement of the elements discussed without departing from the scope of the disclosure herein. Various examples may omit, replace, or add various procedures or components as desired. In addition, features described with respect to some examples may be combined in other examples as well.
As shown in fig. 1, a dynamic reactive power compensation method based on a neural network includes the following steps:
step 101, generating initial characteristics and operation parameter characteristics for reactive compensation equipment;
the dimension of the initial characteristic comprises reactive compensation equipment ID, power capacity, rated voltage, lower limit of an adjusting range, upper limit of the adjusting range, response speed and rated voltage of a direct-connection line;
the operation parameter characteristics comprise the power factor of a direct connection line (reactive compensation equipment), the content of a voltage fundamental wave component, the content of 3-h time voltage harmonic wave components of the direct connection line, the content of a current fundamental wave component of the direct connection line and the content of 3-h time current harmonic wave components of the direct connection line;
wherein h is more than or equal to 3;
in one embodiment of the invention, 30 is greater than or equal to h is greater than or equal to 3, and h is an odd number; this is because the grid harmonics are essentially odd harmonics.
If there is a missing value in the operation parameter characteristics, interpolation operation needs to be carried out, and the interpolation value defaults to 0.
Step 102, generating unified features with dimension N based on the initial features;
the method for generating the unified feature comprises the following steps:
the initial features are input into a first neural network, the first neural network comprises a first hidden layer, the first hidden layer inputs the initial features, and the output of the first hidden layer unifies the features.
When the first neural network is trained, the output of the first hidden layer of the first neural network is connected with a first classifier, the number of classification labels of the first classifier is the same as the number of reactive compensation devices, and one classification label of the first classifier corresponds to the reactive compensation device.
Training the first neural network by constructing a training set of initial characteristics of all reactive compensation equipment, wherein the weight parameters of the trained first neural network are fixed.
The first neural network is of a single-layer network structure, and the calculation speed is very short.
Step 103, inputting the unified characteristic and the operation parameter characteristic into a compensation distribution calculation model to calculate the compensation power or output proportion (the proportion of output to power capacity) of each reactive compensation device;
the compensation allocation calculation model includes: the system comprises an accumulation layer, an attention layer, a second hiding layer, a third hiding layer, a fourth hiding layer, a sequence hiding layer and a second classifier, wherein the accumulation layer inputs unified characteristics and operation parameter characteristics and outputs first characteristics;
in one embodiment of the present invention, the accumulation layer mixes the unifying feature and the operating parameter feature, and the method of mixing the unifying feature and the operating parameter feature is to splice the unifying feature and the operating parameter feature, for example splice the tail end of the unifying feature vector into the operating parameter feature vector. In this embodiment, the accumulation layer does not have trainable parameters.
The first feature inputs attention layer, attention layer includes first attention unit, second attention unit, third attention unit, first attention unit, second attention unit and third attention unit input first feature respectively, first attention unit, second attention unit and third attention unit output second feature respectively, third feature and fourth feature respectively, wherein second feature and third feature input second hidden layer, second hidden layer output fifth feature, fifth feature and fourth feature input third hidden layer, third hidden layer output sixth feature, sixth feature and operating parameter feature input fourth hidden layer, fourth hidden layer output seventh feature, seventh feature input sequence hidden layer, the sequence hidden layer includes M LSTM (long short term memory network) layers, each LSTM (long term memory network) layer includes H LSTM units in series connection, the LSTM unit of the M th layer of sequence hidden layer is connected with a second classifier respectively, the reactive power classification label of j second classifier indicates the reactive power compensation ratio of the j-th reactive compensation equipment under the next control cycle.
In one embodiment of the present invention, the set of class labels of the j-th second classifier is denoted as a= { a 1 ,A 2 …A k Value range [0, s ] representing the output ratio]Discretizing to generate k point values, and mapping the k point values with k classification labels of the set A respectively;
the internal operations of the first attention unit include:
c i =σ(x i W xc +b c ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xc Representing weight parameters, b c Represents the bias parameter, σ represents the activation function, c i Representing a second feature;
the internal operations of the second attention unit include:
d i =σ(x i W xd +b d ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xd Representing weight parameters, b d Represents the bias parameter, σ represents the activation function, d i Representing a third feature;
the internal operations of the third attention unit include:
e i =σ(x i W xe +b e ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xe Representing weight parameters, b d Represents the bias parameter, sigma represents the activation function, e i Representing a fourth feature;
the internal operations of the second hidden layer include:
wherein c i Representing a second characteristic, d, of the ith reactive compensation equipment j Indicating the third characteristic of the jth reactive compensation device, +. j Dimension e of j A fourth feature representing a j-th reactive compensation device, M being equal to the number of reactive compensation devices, the summation representing the addition of vectors, a i A sixth feature representing an ith reactive compensation equipment;
wherein c i Representing a second characteristic, d, of the ith reactive compensation equipment q Representing a third characteristic, d, of the q-th reactive compensation device j A third feature representing a j-th reactive compensation device, M being equal to the number of reactive compensation devices;
the internal operations of the fourth hidden layer include:
wherein a is i A sixth feature representing an ith reactive compensation device,/->An operating parameter characteristic representing the nth control period of the ith reactive compensation equipment, +.>A seventh feature representing an nth control period of the ith reactive compensation equipment;
the internal operations of the sequence hiding layer include:
the operation process of the nth layer and the t-th LSTM unit is as follows:
definition:i=t,h t-1 representing the status of delivery->h 0 =0,/> Output of the t-th LSTM cell representing the n-1 th hidden layer, +.>An output of the t-1 th LSTM cell representing the nth hidden layer;
forgetting door f t The calculation formula of (2) is as follows:
wherein W is fx Representation->Transfer to f t Corresponding weightThe matrix is formed by a matrix of,representing the transfer state h t-1 Transfer to f t Corresponding weight matrix, b f Representing bias terms, σ representing a sigmoid function;
input gate i t The calculation formula of (2) is as follows:
wherein W is xi Representation input +.>Transfer to i t Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to i t Corresponding weight matrix, b i Representing bias terms, σ representing a sigmoid function;
intermediate stateCan be expressed as follows:
wherein W is xC Representation input +.>Transfer to->Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to->Corresponding weight matrix, b C Representing a bias term, tanh representing an activation function tanh;
output stateExpressed by the following formula:
wherein the method comprises the steps ofIs the output state transmitted by the t-1 LSTM of the nth layer, f t 、i t 、/>Is the result of calculation of forget gate, input gate, intermediate state, < >>Door f for indicating forgetfulness t And the output state of the t-1 th LSTM of the nth layer +.>Point-by-point multiplication is performed to make->Indicating input/output gate i t And intermediate state->Performing point-by-point multiplication;
the output gate is expressed as:
wherein W is xo Representation input +.>Delivery ofTo o t Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to o t Corresponding weight matrix, b o Represents the bias term, σ represents the sigmoid function.
Output ofCan be expressed as follows:
will output door o t And tan h (C) t ) Multiplying point by point to obtain the output of the current LSTM unit
The compensation distribution calculation model of the invention requires the connection of the reactive compensation equipment and the integral characteristic of the reactive compensation network from the reactive compensation network through the attention mechanism, and the unified characteristic is not updated when the reactive compensation network is unchanged, so that the output of the attention layer of the compensation distribution calculation model is not updated after initialization, the whole reactive compensation network has good calculation efficiency under the condition of keeping complex expression capability framework, can achieve millisecond-level operation speed, and provides low-interval dynamic reactive compensation control.
The embodiment has been described above with reference to the embodiment, but the embodiment is not limited to the above-described specific implementation, which is only illustrative and not restrictive, and many forms can be made by those of ordinary skill in the art, given the benefit of this disclosure, are within the scope of this embodiment.
Claims (10)
1. The dynamic reactive power compensation method based on the neural network is characterized by comprising the following steps of:
step 101, generating initial characteristics and operation parameter characteristics for reactive compensation equipment;
step 102, generating unified features with dimension N based on the initial features;
step 103, inputting the unified characteristic and the operation parameter characteristic into a compensation distribution calculation model to calculate the compensation power or output proportion of each reactive compensation device;
the compensation allocation calculation model includes: the system comprises an accumulation layer, an attention layer, a second hiding layer, a third hiding layer, a fourth hiding layer, a sequence hiding layer and a second classifier, wherein the accumulation layer inputs unified characteristics and operation parameter characteristics and outputs first characteristics;
the first feature inputs attention layers, the attention layers comprise a first attention unit, a second attention unit and a third attention unit, the first attention unit, the second attention unit and the third attention unit respectively input first features, the first attention unit, the second attention unit and the third attention unit respectively output second features, third features and fourth features, the second features and the third features are input into second hidden layers, the second hidden layers output fifth features, the fifth features and the fourth features are input into third hidden layers, the third hidden layers output sixth features, the sixth features and the operating parameter features are input into fourth hidden layers, the fourth hidden layers output seventh features, the seventh features are input into sequence hidden layers, the sequence hidden layers comprise M LSTM layers, each LSTM layer comprises H LSTM units connected in series, the LSTM units of the Mth layer of the sequence hidden layers are respectively connected with a second classifier, and the labels of the j second classifier represent the compensation power or output proportion of the j reactive compensation equipment in the next control period.
2. A method of dynamic reactive compensation based on neural networks according to claim 1, characterized in that the dimensions of the initial characteristics include reactive compensation equipment ID, power capacity, rated voltage, lower regulation range limit, upper regulation range limit, response speed, direct link rated voltage.
3. The neural network-based dynamic reactive compensation method according to claim 1, wherein the operation parameter characteristics include a power factor of a direct connection line (reactive compensation device), a voltage fundamental component content, a content of 3-h voltage harmonic components of the direct connection line, a current fundamental component content of the direct connection line, and a content of 3-h current harmonic components of the direct connection line; wherein h is more than or equal to 3.
4. The method for dynamic reactive power compensation based on a neural network according to claim 1, wherein the method for generating the unified feature comprises:
the initial features are input into a first neural network, the first neural network comprises a first hidden layer, the first hidden layer inputs the initial features, and the output of the first hidden layer unifies the features.
5. The method of claim 4, wherein the output of the first hidden layer of the first neural network is connected to a first classifier when the first neural network is trained, the number of classification labels of the first classifier is the same as the number of reactive compensation devices, and one classification label of the first classifier corresponds to the representation of one reactive compensation device.
6. The neural network-based dynamic reactive power compensation method of claim 5, wherein the first neural network is trained by constructing a training set of initial feature of all reactive power compensation devices, and the weight parameters of the trained first neural network are fixed.
7. The method for dynamic reactive power compensation based on a neural network according to claim 1, wherein the integration layer mixes the unified feature and the operation parameter feature, and the method for mixing the unified feature and the operation parameter feature is to splice the unified feature and the operation parameter feature, and comprises splicing the tail ends of the unified feature vectors into the operation parameter feature vectors.
8. The neural network-based dynamic reactive compensation method of claim 1, wherein the set of classification labels of the j-th second classifier is denoted as a= { a 1 ,A 2 …A k Value range [0, s ] representing the output ratio]And performing discretization to generate k point values, and mapping the k point values with k classification labels of the set A respectively.
9. A method of dynamic reactive power compensation based on neural networks according to claim 1, characterized in that the internal operation of the first attention unit comprises:
c i =σ(x i W xc +b c ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xc Representing weight parameters, b c Represents the bias parameter, σ represents the activation function, c i Representing a second feature;
the internal operations of the second attention unit include:
d i =σ(x i W xd +b d ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xd Representing weight parameters, b d Represents the bias parameter, σ represents the activation function, d i Representing a third feature;
the internal operations of the third attention unit include:
e i =σ(x i W xe +b e ) Wherein x is i Representing a first characteristic, W, corresponding to the ith reactive compensation equipment xe Representing weight parameters, b d Represents the bias parameter, sigma represents the activation function, e i Representing a fourth feature;
the internal operations of the second hidden layer include:
wherein c i Second bit representing the ith reactive compensation equipmentSign, d j Indicating the third characteristic of the jth reactive compensation device, +. j Dimension e of j A fourth feature representing a j-th reactive compensation device, M being equal to the number of reactive compensation devices, the summation representing the addition of vectors, a i A sixth feature representing an ith reactive compensation equipment;
wherein c i Representing a second characteristic, d, of the ith reactive compensation equipment q Representing a third characteristic, d, of the q-th reactive compensation device j A third feature representing a j-th reactive compensation device, M being equal to the number of reactive compensation devices;
the internal operations of the fourth hidden layer include:
wherein a is i A sixth feature representing an ith reactive compensation device,/->An operating parameter characteristic representing the nth control period of the ith reactive compensation equipment, +.>A seventh feature representing an nth control period of the ith reactive compensation equipment.
10. The method for dynamic reactive power compensation based on a neural network according to claim 9, wherein the internal operation of the sequence hiding layer comprises:
the operation process of the nth layer and the t-th LSTM unit is as follows:
definition:i=t,h t-1 representing the status of delivery->h 0 =0,/> Output of the t-th LSTM cell representing the n-1 th hidden layer, +.>An output of the t-1 th LSTM cell representing the nth hidden layer;
forgetting door f t The calculation formula of (2) is as follows:
wherein W is fx Representation->Transfer to f t The corresponding weight matrix is used to determine the weight matrix,representing the transfer state h t-1 Transfer to f t Corresponding weight matrix, b f Representing bias terms, σ representing a sigmoid function;
input gate i t The calculation formula of (2) is as follows:
wherein W is xi Representation input +.>Transfer to i t The corresponding weight matrix is used to determine the weight matrix,representing the transfer state h t-1 Transfer to i t Corresponding weight matrix, b i Representing bias terms, σ representing a sigmoid function;
intermediate stateCan be expressed as follows:
wherein W is xC Representation input +.>Transfer to->Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to->Corresponding weight matrix, b C Representing a bias term, tanh representing an activation function tanh;
output stateExpressed by the following formula:
wherein the method comprises the steps ofT-1 th LSTM transfer, which is the nth layerOutput state f t 、i t 、/>Is the result of calculation of forget gate, input gate, intermediate state, < >>Door f for indicating forgetfulness t And the output state of the t-1 th LSTM of the nth layer +.>Point-by-point multiplication is performed to make->Indicating input/output gate i t And intermediate state->Performing point-by-point multiplication;
the output gate is expressed as:
wherein W is xo Representation input +.>Transfer to o t Corresponding weight matrix, < >>Representing the transfer state h t-1 Transfer to o t Corresponding weight matrix, b o Representing bias terms, σ representing a sigmoid function;
output ofCan be expressed as follows:
will output door o t And tan h (C) t ) Multiplying point by point to obtain the output of the current LSTM unit
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310670249.2A CN116760055B (en) | 2023-06-07 | 2023-06-07 | Dynamic reactive compensation method based on neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310670249.2A CN116760055B (en) | 2023-06-07 | 2023-06-07 | Dynamic reactive compensation method based on neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116760055A true CN116760055A (en) | 2023-09-15 |
CN116760055B CN116760055B (en) | 2024-03-12 |
Family
ID=87960123
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310670249.2A Active CN116760055B (en) | 2023-06-07 | 2023-06-07 | Dynamic reactive compensation method based on neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116760055B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117154906A (en) * | 2023-11-01 | 2023-12-01 | 维屿(深圳)科技有限公司 | Wireless charging regulation and control method with refrigeration function |
Citations (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101247046A (en) * | 2007-12-19 | 2008-08-20 | 湖南大学 | Harmonic wave dynamic managing and reactive-load dynamic compensation compound control method and implementing device thereof |
CN101420128A (en) * | 2008-12-05 | 2009-04-29 | 湖南大学 | Dynamic passive compensation system and controlling method thereof |
CN102856910A (en) * | 2012-07-31 | 2013-01-02 | 上海交通大学 | Static compensator (STATCOM) control method based on multi-model fuzzy neural network PI |
US20170070135A1 (en) * | 2015-09-08 | 2017-03-09 | Lsis Co., Ltd. | Static var compensator apparatus and operating method thereof |
CN107230987A (en) * | 2017-07-27 | 2017-10-03 | 广东电网有限责任公司惠州供电局 | A kind of distribution network var compensation Regional Synergetic control system based on mixing reactive power compensator |
CN108199383A (en) * | 2018-02-06 | 2018-06-22 | 北京天润新能投资有限公司 | The joint debugging control method and system of a kind of automatic voltage control system and dynamic passive compensation system |
CN108767869A (en) * | 2018-06-14 | 2018-11-06 | 华北水利水电大学 | A kind of static var compensator voltage adjusting method based on artificial neural network |
CN110738344A (en) * | 2018-07-20 | 2020-01-31 | 中国农业大学 | Distributed reactive power optimization method and device for load prediction of power system |
CN110943461A (en) * | 2019-11-25 | 2020-03-31 | 国网辽宁省电力有限公司本溪供电公司 | Intelligent integrated power distribution device applied to low-voltage transformer area and method thereof |
CN112600219A (en) * | 2020-12-31 | 2021-04-02 | 国网河南省电力公司南阳供电公司 | Dynamic reactive power standby optimization method for alternating current-direct current power grid |
US20210201155A1 (en) * | 2019-12-30 | 2021-07-01 | Dalian University Of Technology | Intelligent control method for dynamic neural network-based variable cycle engine |
CN113839396A (en) * | 2021-10-10 | 2021-12-24 | 费莱(浙江)科技有限公司 | Dynamic reactive power compensation system and method for repairing line loss of power grid |
CN115313510A (en) * | 2022-07-23 | 2022-11-08 | 国网内蒙古东部电力有限公司 | Adaptive reactive compensation photovoltaic inverter control method and system |
CN115759321A (en) * | 2022-09-14 | 2023-03-07 | 国网山西省电力公司大同供电公司 | Intelligent reactive power compensation method taking minimum line loss as optimization target |
CN115828768A (en) * | 2022-12-28 | 2023-03-21 | 华南理工大学 | Data-driven power distribution network reactive power prediction method |
-
2023
- 2023-06-07 CN CN202310670249.2A patent/CN116760055B/en active Active
Patent Citations (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101247046A (en) * | 2007-12-19 | 2008-08-20 | 湖南大学 | Harmonic wave dynamic managing and reactive-load dynamic compensation compound control method and implementing device thereof |
CN101420128A (en) * | 2008-12-05 | 2009-04-29 | 湖南大学 | Dynamic passive compensation system and controlling method thereof |
CN102856910A (en) * | 2012-07-31 | 2013-01-02 | 上海交通大学 | Static compensator (STATCOM) control method based on multi-model fuzzy neural network PI |
US20170070135A1 (en) * | 2015-09-08 | 2017-03-09 | Lsis Co., Ltd. | Static var compensator apparatus and operating method thereof |
CN107230987A (en) * | 2017-07-27 | 2017-10-03 | 广东电网有限责任公司惠州供电局 | A kind of distribution network var compensation Regional Synergetic control system based on mixing reactive power compensator |
CN108199383A (en) * | 2018-02-06 | 2018-06-22 | 北京天润新能投资有限公司 | The joint debugging control method and system of a kind of automatic voltage control system and dynamic passive compensation system |
CN108767869A (en) * | 2018-06-14 | 2018-11-06 | 华北水利水电大学 | A kind of static var compensator voltage adjusting method based on artificial neural network |
CN110738344A (en) * | 2018-07-20 | 2020-01-31 | 中国农业大学 | Distributed reactive power optimization method and device for load prediction of power system |
CN110943461A (en) * | 2019-11-25 | 2020-03-31 | 国网辽宁省电力有限公司本溪供电公司 | Intelligent integrated power distribution device applied to low-voltage transformer area and method thereof |
US20210201155A1 (en) * | 2019-12-30 | 2021-07-01 | Dalian University Of Technology | Intelligent control method for dynamic neural network-based variable cycle engine |
CN112600219A (en) * | 2020-12-31 | 2021-04-02 | 国网河南省电力公司南阳供电公司 | Dynamic reactive power standby optimization method for alternating current-direct current power grid |
CN113839396A (en) * | 2021-10-10 | 2021-12-24 | 费莱(浙江)科技有限公司 | Dynamic reactive power compensation system and method for repairing line loss of power grid |
CN115313510A (en) * | 2022-07-23 | 2022-11-08 | 国网内蒙古东部电力有限公司 | Adaptive reactive compensation photovoltaic inverter control method and system |
CN115759321A (en) * | 2022-09-14 | 2023-03-07 | 国网山西省电力公司大同供电公司 | Intelligent reactive power compensation method taking minimum line loss as optimization target |
CN115828768A (en) * | 2022-12-28 | 2023-03-21 | 华南理工大学 | Data-driven power distribution network reactive power prediction method |
Non-Patent Citations (2)
Title |
---|
AVIK BHATTACHARYA等: "Harmonic elimination and reactive power compensation through a shunt active power filter by twin neural networks with predictive and adaptive properties", IEEE * |
王育飞 等: "基于神经网络预测的电弧炉供电系统动态无功混合补偿方法", 《电力电容器与无功补偿》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117154906A (en) * | 2023-11-01 | 2023-12-01 | 维屿(深圳)科技有限公司 | Wireless charging regulation and control method with refrigeration function |
CN117154906B (en) * | 2023-11-01 | 2024-04-02 | 维屿(深圳)科技有限公司 | Wireless charging regulation and control method with refrigeration function |
Also Published As
Publication number | Publication date |
---|---|
CN116760055B (en) | 2024-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Consensus-based distributed economic dispatch control method in power systems | |
CN116760055B (en) | Dynamic reactive compensation method based on neural network | |
Li et al. | An intelligent transient stability assessment framework with continual learning ability | |
Liu et al. | Model‐free adaptive control method for a class of unknown MIMO systems with measurement noise and application to quadrotor aircraft | |
CN111507464B (en) | Equation solver based on memristor array and operation method thereof | |
Balakrishnan et al. | On-line emission and economic load dispatch using adaptive Hopfield neural network | |
Migoni et al. | Efficient simulation of hybrid renewable energy systems | |
Baranyi et al. | SVD-based complexity reduction to TS fuzzy models | |
Kumar et al. | Clamped state solution of artificial neural network for real-time economic dispatch | |
Agrawal et al. | Hybrid deep neural network-based generation rescheduling for congestion mitigation in spot power market | |
Naderi et al. | Circuit implementation of high-resolution rational-powered membership functions in standard CMOS technology | |
Alfredo et al. | Time series forecasting of significant wave height using gru, cnn-gru, and lstm | |
Malik et al. | Comparison of optimization by response surface methodology with neurofuzzy methods | |
Yang et al. | Finite‐time control of periodic systems with event‐triggering mechanisms | |
CN115660385B (en) | Method and device for decomposing and parallelly solving economic operation domain of convex hull of power grid | |
CN114400675B (en) | Active power distribution network voltage control method based on weight mean value deep double-Q network | |
CN115186884A (en) | Ultra-short-term wind power prediction method based on wavelet decomposition and storage medium | |
Karad et al. | Enhanced control of doubly fed induction generator based wind turbine system using soft computing assisted fractional order controller | |
Heimes et al. | Traditional and evolved dynamic neural networks for aircraft simulation | |
Parabhane et al. | Artificial neural networks based power management scheme with enhanced stability for a solar panel/wind turbine generator/fuel cell/battery/power supply designed for industrial loads | |
Hassan-Moghaddam et al. | Data-driven proximal algorithms for the design of structured optimal feedback gains | |
Wang | Application of flywheel energy storage system to enhance transient stability of power systems | |
US11604911B2 (en) | Simulation of gas dynamics of different gas channel geometries in fuel cells | |
Wang et al. | Reconstruction of complex network from time series data based on graph attention network and Gumbel Softmax | |
Yaniv et al. | State solutions for distribution systems and switching event using a neural network: State Solution Using Neural Network |
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 | ||
CB03 | Change of inventor or designer information | ||
CB03 | Change of inventor or designer information |
Inventor after: Xiao Yicheng Inventor after: Wang Xiaohan Inventor before: Xiao Yicheng Inventor before: Huang Xiebinliang Inventor before: Wang Xiaohan Inventor before: Hu Yanglei |
|
GR01 | Patent grant | ||
GR01 | Patent grant |