CN114154688A - Short-term power prediction method for photovoltaic power station - Google Patents

Short-term power prediction method for photovoltaic power station Download PDF

Info

Publication number
CN114154688A
CN114154688A CN202111366036.8A CN202111366036A CN114154688A CN 114154688 A CN114154688 A CN 114154688A CN 202111366036 A CN202111366036 A CN 202111366036A CN 114154688 A CN114154688 A CN 114154688A
Authority
CN
China
Prior art keywords
photovoltaic power
power generation
generation system
gins
twin 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.)
Pending
Application number
CN202111366036.8A
Other languages
Chinese (zh)
Inventor
曹利蒲
曾凡春
张彬
杨继明
田长风
陈岩磊
张澈
王传鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Huaneng Xinrui Control Technology Co Ltd
Original Assignee
Beijing Huaneng Xinrui Control Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Beijing Huaneng Xinrui Control Technology Co Ltd filed Critical Beijing Huaneng Xinrui Control Technology Co Ltd
Priority to CN202111366036.8A priority Critical patent/CN114154688A/en
Publication of CN114154688A publication Critical patent/CN114154688A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • 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
    • 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
    • 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Power Engineering (AREA)
  • Molecular Biology (AREA)
  • Development Economics (AREA)
  • Evolutionary Computation (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Biomedical Technology (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention provides a short-term power prediction method for a photovoltaic power station, and belongs to the technical field of photovoltaic power generation. The short-term power prediction method of the photovoltaic power station specifically comprises the following steps: constructing a twin model of a photovoltaic power generation power prediction system based on the operation data of the photovoltaic power generation system; and the twin model carries out knowledge interactive learning with the photovoltaic power generation system through GINs so as to transfer the relation knowledge from the photovoltaic power generation system to the twin model and guide the photovoltaic power generation system to carry out short-term power prediction. The method applies a digital twinning technology to the photovoltaic power station, constructs a digital twinning body by mapping field operation data to a virtual space, can reflect the operation condition of the photovoltaic power station with high fidelity, adopts an information interaction network to complete information interaction learning between the field of the photovoltaic power generation system and a digital twinning model of the photovoltaic power generation system, completes real-time updating of the digital twinning model, and accurately predicts the ultra-short-term power of the photovoltaic power station in real time.

Description

Short-term power prediction method for photovoltaic power station
Technical Field
The invention belongs to the technical field of photovoltaic power generation, and particularly relates to a short-term power prediction method for a photovoltaic power station.
Background
In order to meet the energy-saving requirement of an energy system, a photovoltaic power generation system is an efficient solution as a green and renewable energy source. The photovoltaic power prediction plays a significant role in operation and maintenance optimization of the distributed photovoltaic power station, but the randomness of photovoltaic power generation may have negative effects on the stability and reliability of a power grid. Therefore, improving the ultra-short-term power prediction accuracy of the photovoltaic power generation system is a precondition for realizing the merging of a larger-scale distributed photovoltaic power generation system into a power grid. At present, algorithms for ultra-short-term power prediction of a photovoltaic power generation system are infinite, but a prediction model of the photovoltaic power generation system is rarely updated in real time, and in addition, transferable information research between the photovoltaic power generation system and the power prediction model is rare, namely information interaction between a photovoltaic power generation system operation site and a twin model of the photovoltaic power generation system operation site is lacked, so that the precision of short-term power prediction is reduced to a great extent.
Therefore, it is necessary to propose a new method for short-term power prediction of photovoltaic power plants.
Disclosure of Invention
The invention aims to at least solve one of the technical problems in the prior art and provides a short-term power prediction method for a photovoltaic power station.
The invention provides a short-term power prediction method of a photovoltaic power station, which specifically comprises the following steps:
constructing a twin model of a photovoltaic power generation power prediction system based on the operation data of the photovoltaic power generation system;
and the twin model carries out knowledge interactive learning with the photovoltaic power generation system through GINs so as to transfer the relation knowledge from the photovoltaic power generation system to the twin model and guide the photovoltaic power generation system to carry out short-term power prediction.
Optionally, the constructing a twin model of the photovoltaic power generation power prediction system based on the operation data of the photovoltaic power generation system includes:
and mapping the operation data of the photovoltaic power generation system to a virtual space, and constructing a twin model of the photovoltaic power generation power prediction system by adopting a digital twin technology.
Optionally, the twin model performs knowledge interactive learning with the photovoltaic power generation system through the GINs to transfer the relationship knowledge from the photovoltaic power generation system to the twin model, including:
determining a symmetric weight matrix through GINs;
learning a non-grid structure through GINs;
an objective function is constructed from the GINs.
Optionally, the convolutional network node feature of the photovoltaic power generation system is represented as G1(X1,A1)，X1＝{x1,1,x1,2，…，x1N }; wherein G is1(X1,A1) Knowledge of network relationships for the previous target; a. the1The relation between different nodes is captured as the symmetrical weight matrix of the previous target, X1Is a convolutional network node characteristic.
Optionally, the determining a symmetric weight matrix through the GINs includes:
mixing X1Embedding into a set of potential vectors
Figure BDA0003360642050000021
And calculating X1 emAnd transpose thereof (X)1 em)TTo obtain said symmetric weight momentsMatrix of
Figure BDA0003360642050000022
Optionally, the performing non-grid structure learning by GINs includes:
using GINs to assign a symmetric weight matrix A1And convolution network node characteristic X1Convolutional layers sent together to the target do non-mesh learning:
Figure BDA0003360642050000023
to obtain A1And
Figure BDA0003360642050000024
wherein, W1Is a matrix of weights that can be learned,
Figure BDA0003360642050000025
output convolutional network node characteristics for the previous target;
and, using GINs to weight the symmetric weight matrix A2And convolution network node characteristic X2Convolutional layers sent together to the target do non-mesh learning:
Figure BDA0003360642050000026
to obtain A2And
Figure BDA0003360642050000027
wherein, W2Is a learnable weight matrix, A2Is a symmetric weight matrix for the latter object,
Figure BDA0003360642050000028
and (4) outputting the convolutional network node characteristics for the later target.
Optionally, the GINs is a multi-input and multi-output module, and the specific relation is as follows:
Figure BDA0003360642050000029
wherein,
Figure BDA00033606420500000210
respectively outputting convolution network node characteristics of a front target and a rear target;
A1,A2and the symmetric weight matrixes are respectively used for transferring the relation knowledge between the photovoltaic power generation system and the digital twin model.
Optionally, the constructing an objective function through the GINs includes:
designing a loss function and enabling the network to transfer knowledge of the relation across the weight matrix, wherein the method comprises the following specific steps:
Figure BDA0003360642050000031
wherein J represents the loss function in the GINs network;
Figure BDA0003360642050000032
represents the original loss of task O;
Figure BDA0003360642050000033
a matrix A representing the transfer loss of information transfer and forcing the photovoltaic power generation system to twin models1,A2Close to each other.
Optionally, the instructing the photovoltaic power generation system to perform short-term power prediction includes:
and judging the magnitude of the power prediction error delta u of the photovoltaic power generation system and the reason causing the error, and optimizing the photovoltaic power generation system according to the error reason so as to complete short-term power prediction of the photovoltaic power generation system.
Optionally, the twin model performs knowledge interactive learning with the photovoltaic power generation system through the GINs to transfer the relationship knowledge from the photovoltaic power generation system to the twin model and guide the photovoltaic power generation system to perform short-term power prediction, and further includes:
enabling the twin model and the photovoltaic power generation system to perform knowledge interactive learning through GINs so as to update the twin model in real time; and the number of the first and second groups,
and correcting and updating the twin model according to the error reason.
The invention provides a short-term power prediction method for a photovoltaic power generation system, which specifically comprises the following steps: constructing a twin model of a photovoltaic power generation power prediction system based on the operation data of the photovoltaic power generation system; and the twin model carries out knowledge interactive learning with the photovoltaic power generation system through GINs so as to transfer the relation knowledge from the photovoltaic power generation system to the twin model and guide the photovoltaic power generation system to carry out short-term power prediction. The method applies a digital twinning technology to the photovoltaic power station, constructs a digital twinning body by mapping field operation data to a virtual space, can reflect the operation condition of the photovoltaic power station with high fidelity, adopts an information interaction network to complete information interaction learning between the field of the photovoltaic power generation system and a digital twinning model of the photovoltaic power generation system, completes real-time updating of the digital twinning model, and accurately predicts the ultra-short-term power of the photovoltaic power station in real time.
Drawings
Fig. 1 is a schematic diagram of a short-term power prediction method for a photovoltaic power plant according to an embodiment of the present invention;
FIG. 2 is a block flow diagram of a short term power prediction method for a photovoltaic power plant according to another embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a digital twinning technique according to another embodiment of the present invention;
fig. 4 is a diagram illustrating a GINs internal information transfer model according to another embodiment of the present invention.
Detailed Description
In order to make the technical solutions of the present invention better understood, the present invention will be described in further detail with reference to the accompanying drawings and specific embodiments. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the described embodiments of the invention without any inventive step, are within the scope of protection of the invention.
Unless otherwise specifically stated, technical or scientific terms used herein shall have the ordinary meaning as understood by those of ordinary skill in the art to which this invention belongs. The use of "including" or "comprising" and the like in this disclosure does not limit the presence or addition of any number, step, action, operation, component, element, and/or group thereof or does not preclude the presence or addition of one or more other different numbers, steps, actions, operations, components, elements, and/or groups thereof. Furthermore, the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number and order of the indicated features.
As shown in fig. 1 and fig. 2, the present invention provides a short-term power prediction method S100 for a photovoltaic power station, which specifically includes the following steps S110 to S120:
s110, constructing a twin model of the photovoltaic power generation power prediction system based on the operation data of the photovoltaic power generation system.
Specifically, the embodiment maps the operation data of the photovoltaic power generation system to a virtual space, and adopts a digital twin technology to construct a twin model of the photovoltaic power generation power prediction system.
It should be noted that, in the context of rapid development of the internet of things, the digital twin technology is successfully applied in a plurality of industrial fields because the digital twin technology can reflect the operation condition of a physical entity with high fidelity. Therefore, it is necessary to introduce an information interaction network and a digital twin technology into ultra-short-term power prediction of a photovoltaic power station, comprehensively consider all physical factors of photovoltaic power generation, and update a prediction model in real time so as to reduce the large deviation between the generated power of the photovoltaic power station and the predicted value thereof. The digital twin technology can accelerate the understanding of people on physical entities, has strong prediction capability, and is successfully applied to the aspects of regular digital power grid regulation of large-scale power grid models, high-efficiency inverter modeling methods and the like. Therefore, the embodiment has practical significance in introducing the digital twinning technology into the photovoltaic power station power prediction research.
It should be further noted that the digital twinning technology adopted in this embodiment fully utilizes the interactive simulation among the physical entities, the sensors, and the field monitoring database of the system to implement high-fidelity virtual-real mapping between the digital virtual space and the actual physical system, thereby implementing the full-life-cycle synchronous evolution between the digital twins and the physical system.
Specifically, the structural architecture of the digital twinning technique adopted in the present embodiment is shown in fig. 3. The structure of the digital twin technology mainly comprises a physical layer, a perception layer, a data transmission layer, a data processing layer and a decision layer. Each layer is composed of physical equipment, is related layer by layer and is progressive from bottom to top.
And S120, the twin model and the photovoltaic power generation system perform knowledge interactive learning through GINs, so that the relation knowledge is transferred from the photovoltaic power generation system to the twin model, and the photovoltaic power generation system is guided to perform short-term power prediction, namely, information interactive learning between the photovoltaic power station site and the digital twin model of the photovoltaic power station is realized through GINs.
It should be noted that, in the present embodiment, the information interaction networks (GINs) are different from the conventional knowledge transfer method, and the transfer of the internal relationship between different targets is emphasized, so that the transferred relationship information can be better stored. First, the GINs generate a "self-learned" weight matrix for each target, which is a higher-level representation of the input data. The GINs then send the weight matrix along with the Convolutional Network (CN) node characteristics to the target convolutional layer, whose output characteristics are available for the original task. Finally, in the optimization process, the GINs design the loss function specific to the task and introduce the network to transmit knowledge across the weight matrix, so that the relationship information is well stored in the transmission process. In this way, GINs can leverage knowledge of relationships in one more advantageous target (e.g., with supervisory signals or higher quality data) to guide the learning process of another target for better performance.
Specifically, the internal information transfer of the GINs algorithm is shown in fig. 4, and as can be seen from fig. 4, the present embodiment aims to transfer the relationship knowledge from the front target to the rear target. For the pre-target G1(X1,A1)，X1＝{x1,1,x1,2，…，x1，NDenotes the convolutional network nodePerforming sign; wherein G is1(X1,A1) Network knowledge of the previous target; a. the1The relation between different nodes is captured as the symmetrical weight matrix of the previous target, X1Is a convolutional network node characteristic. A. the1A weight matrix is represented that captures the relationship between the different nodes. Mathematically, element AijCoding the connection weight between the ith node and the jth node can be represented as A1，ij＝f(x1,i,x1,j)。
Further, determining the symmetry weights by the GINs includes: to obtain a symmetric weight matrix A1Is mixing X1Embedding into a set of potential vectors
Figure BDA0003360642050000061
Such as CN in fig. 4, and then non-mesh representation learning is performed and X is calculated1 emAnd transpose thereof (X)1 em)TTo obtain a symmetric weight matrix
Figure BDA0003360642050000062
Wherein, W1Is a matrix of weights that can be learned,
Figure BDA0003360642050000063
the output convolution network node characteristics of the previous target.
In addition, for the rear target G2(X2,A2)，X2＝{x2,1,x2,2，…，x2，NTo X2The same treatment is carried out to obtain the corresponding A2And
Figure BDA0003360642050000064
specifically, the symmetric weight matrix A is divided by GINs2And convolution network node characteristic X2Convolutional layers sent together to the target do non-mesh learning:
Figure BDA0003360642050000065
to obtain A2And
Figure BDA0003360642050000066
wherein, W2Is a learnable weight matrix, A2Is a symmetric weight matrix for the latter object,
Figure BDA0003360642050000067
and (4) outputting the convolutional network node characteristics for the later target.
In summary, the GINs core of the present embodiment is a multi-input/output module, and the specific relationship is as follows:
Figure BDA0003360642050000068
wherein,
Figure BDA0003360642050000069
Figure BDA00033606420500000610
the output convolutional network node characteristics of the front target and the rear target respectively. A. the1,A2And respectively obtaining the symmetrical weight matrixes of the front target and the rear target adopted for the relation knowledge transfer between the photovoltaic power generation system and the digital twin model.
Further, constructing an objective function by the GINs includes: designing a loss function and enabling the network to transfer knowledge of the relation across the weight matrix, wherein the method comprises the following specific steps:
Figure BDA00033606420500000611
wherein J represents an objective function, i.e. a loss function in the GINs network;
Figure BDA00033606420500000612
representing an original loss (e.g., recognition loss) of task O;
Figure BDA00033606420500000613
a matrix A representing the transfer loss of information transfer and forcing the photovoltaic power generation system to twin models1,A2Close to each other.
Specifically, referring to fig. 1, the guidance of the photovoltaic power generation system for short-term power prediction includes: and judging the magnitude of the power prediction error delta u of the photovoltaic power generation system and the reason causing the error, and optimizing the photovoltaic power generation system according to the error reason so as to complete short-term power prediction of the photovoltaic power generation system.
For example, as shown in fig. 1, when Δ u > ∈, it is described that an error occurs, and further, the cause of the error is determined, which includes problems of shading and aging of the photovoltaic system, and the operation and maintenance of the photovoltaic power station is rapidly adjusted by the policy layer, so as to complete real-time accurate prediction of the ultra-short-term power of the photovoltaic power station. Of course, the error reason can also be the problem that the virtual model is not accurate, and the twin model is corrected and updated at the moment, so that the problem that the power prediction accuracy of the distributed photovoltaic power station is low is solved.
It should be understood that, in the embodiment, knowledge interactive learning is performed on the twin model and the photovoltaic power generation system through the GINs, so that the twin model is updated and corrected in real time, and the field operation condition of the photovoltaic power station can be reflected in real time in a fidelity manner, so that the problem of real-time updating of the photovoltaic power generation power prediction model is solved.
Based on the above specific control method, the ultra-short-term power prediction algorithm of the photovoltaic power station provided by this embodiment specifically works according to the following principle: firstly, a digital twin body of a photovoltaic power generation power prediction system is constructed by mapping field operation data of a photovoltaic power station to a virtual space. The digital twin model of the photovoltaic power generation power prediction system can reflect the field operation condition of the photovoltaic power station in real time and fidelity. And secondly, performing information interactive learning on the digital twin model of the photovoltaic power generation power prediction system and the photovoltaic power station entity through GINs to complete real-time updating of the digital twin model. And thirdly, judging the cause of the error according to the power prediction error delta u, and rapidly adjusting the operation and maintenance of the photovoltaic power station by a decision layer so as to complete the real-time accurate prediction of the ultra-short-term power of the photovoltaic power station.
The method and the device realize the ultra-short term power prediction algorithm of the photovoltaic power station by combining the interactive network and the digital twin technology, and solve the problem of information transfer learning between the photovoltaic power generation system and the ultra-short term power prediction model thereof. And by observing the power prediction error of the photovoltaic power station, the reason causing the prediction error is judged in advance, and the operation and maintenance of the photovoltaic power station on site are guided, so that the power prediction precision is improved.
The ultra-short term power prediction method of the photovoltaic power station is further described by combining the specific embodiment as follows:
the embodiment performs example analysis on a certain 20MW photovoltaic power station, and comprises the following steps:
1. establishing twin model of photovoltaic power generation system
And mapping the field operation data of the 20MW photovoltaic power station to Matlab2019b simulation software through a sensor to construct a twin model of the photovoltaic power station.
2. Information interactive learning is carried out on photovoltaic power generation system and twin model thereof through GINs
And the twin model of the photovoltaic power generation system and the photovoltaic power generation system thereof performs knowledge interactive learning according to the information transfer flow of the figure 4. The main goal is to shift the knowledge of relationships from the photovoltaic power generation system to the twin model. The convolutional network node characteristic of the photovoltaic power generation system is represented as G1(X1,A1)，X1＝{x1,1,x1,2，…，x1，N}。
(1) Determining a symmetric weight matrix A1
To obtain a symmetric weight matrix A1We first start with X1Embedding into a set of potential vectors
Figure BDA0003360642050000081
Is fully connected to the layer. Then calculate X1 emAnd transpose thereof (X)1 em)TDot product of (a):
Figure BDA0003360642050000082
(2) learning of non-grid structures
Combining the weight matrix A1Node feature X1And (3) sending the convolution layer to perform non-grid structure representation learning:
Figure BDA0003360642050000083
wherein, W1Is a learnable weight matrix. For rear target G2(X2,A2)，X2＝{x2,1,x2,2，…，x2，NTo X2The same treatment is carried out to obtain the corresponding A2And
Figure BDA0003360642050000084
in summary, the GINs core part of the photovoltaic power generation system and the digital twin model can be expressed as a multi-input multi-output module:
Figure BDA0003360642050000085
output characteristic
Figure BDA0003360642050000086
The method is used for power prediction of the photovoltaic power generation system. The relation transfer between the photovoltaic power generation system and the digital twin model adopts a weight matrix A1,A2And contains relationship information.
(3) Object function construction
The objective function is for the GINs for photovoltaic power generation systems and twin models. One is learning the discriminative representation of the photovoltaic power generation system. And secondly, transferring knowledge from the photovoltaic power generation system to a twin model. To achieve these goals, the following loss function is minimized in the optimization process:
Figure BDA0003360642050000087
wherein the first item
Figure BDA0003360642050000088
Representing an original loss (e.g., recognition loss) of task O; second item
Figure BDA0003360642050000089
Representing the transfer loss of information transfer, which forces the matrix A of the photovoltaic power generation system with the twin model1,A2Close to each other.
3. Photovoltaic power plant power prediction
And executing the power prediction scheme shown in the figure 1, and performing real-time information intercommunication learning on a digital twin model of the photovoltaic power generation power prediction system and a photovoltaic power station site through the GINs, so as to optimize the digital twin model and further guide the power prediction of the photovoltaic power generation system. And observing the power prediction error delta u of the photovoltaic power generation system, judging the cause of the error, and quickly adjusting the operation and maintenance of the photovoltaic power station by a decision layer so as to complete the real-time accurate prediction of the ultra-short-term power of the photovoltaic power station.
The invention provides a short-term power prediction method for a photovoltaic power station, which has the following beneficial effects compared with the prior art:
firstly, the digital twin technology is applied to the photovoltaic power station, and the operation condition of the photovoltaic power station can be reflected with high fidelity by mapping the field operation data to the virtual space to construct the digital twin body.
Secondly, the timeliness and the accuracy of the photovoltaic power generation system prediction model are considered, the information interaction learning between the photovoltaic power station site and the digital twin model of the photovoltaic power station site is completed by adopting the information interaction network, and the prediction model is updated and corrected in real time.
Thirdly, the combination of the digital twin technology and the information interaction network greatly improves the ultra-short-term power prediction precision of the photovoltaic power station.
It will be understood that the above embodiments are merely exemplary embodiments taken to illustrate the principles of the present invention, which is not limited thereto. It will be apparent to those skilled in the art that various modifications and improvements can be made without departing from the spirit and substance of the invention, and these modifications and improvements are also considered to be within the scope of the invention.

Claims (10)

1. A short-term power prediction method for a photovoltaic power station is characterized by specifically comprising the following steps:
constructing a twin model of a photovoltaic power generation power prediction system based on the operation data of the photovoltaic power generation system;
and the twin model carries out knowledge interactive learning with the photovoltaic power generation system through GINs so as to transfer the relation knowledge from the photovoltaic power generation system to the twin model and guide the photovoltaic power generation system to carry out short-term power prediction.
2. The method of claim 1, wherein constructing a twin model of a photovoltaic power generation power prediction system based on operational data of the photovoltaic power generation system comprises:
and mapping the operation data of the photovoltaic power generation system to a virtual space, and constructing a twin model of the photovoltaic power generation power prediction system by adopting a digital twin technology.
3. The method of claim 1, wherein the twin model is knowledge interactive learning with the photovoltaic power generation system through GINs to transfer relational knowledge from the photovoltaic power generation system to the twin model, comprising:
determining a symmetric weight matrix through GINs;
learning a non-grid structure through GINs;
an objective function is constructed from the GINs.
4. The method of claim 3, wherein the convolutional network node characteristic of the photovoltaic power generation system is represented as G1(X1,A1)，X1＝{x1,1,x1,2，…，x1N }; wherein G is1(X1,A1) Network knowledge of the previous target; a. the1The relation between different nodes is captured as the symmetrical weight matrix of the previous target, X1Is a convolutional network node characteristic.
5. The method of claim 4, wherein determining the symmetric weight matrices by the GINs comprises:
mixing X1Embedding into a set of potential vectors
Figure FDA0003360642040000021
And calculating X1 emAnd transpose thereof (X)1 em)TTo obtain the symmetric weight matrix
Figure FDA0003360642040000022
6. The method of claim 5, wherein the non-mesh structure learning by GINs comprises:
using GINs to assign a symmetric weight matrix A1And convolution network node characteristic X1Convolutional layers sent together to the target do non-mesh learning:
Figure FDA0003360642040000023
to obtain A1And
Figure FDA0003360642040000024
wherein, W1Is a matrix of weights that can be learned,
Figure FDA0003360642040000025
output convolutional network node characteristics for the previous target;
and, using GINs to weight the symmetric weight matrix A2And convolution network node characteristic X2Convolutional layers sent together to the target do non-mesh learning:
Figure FDA0003360642040000026
to obtain A2And
Figure FDA0003360642040000027
wherein, W2Is a learnable weight matrix, A2Is a symmetric weight matrix for the latter object,
Figure FDA0003360642040000028
and (4) outputting the convolutional network node characteristics for the later target.
7. The method of claim 6 wherein the GINs are a multiple input multiple output module, and the relationship is as follows:
Figure FDA0003360642040000029
wherein,
Figure FDA00033606420400000210
respectively outputting convolution network node characteristics of a front target and a rear target;
A1,A2and respectively obtaining the symmetrical weight matrixes of the front target and the rear target adopted for the relation knowledge transfer between the photovoltaic power generation system and the digital twin model.
8. The method of claim 3, wherein the constructing the objective function from the GINs comprises:
designing a loss function and enabling the network to transfer knowledge of the relation across the weight matrix, wherein the method comprises the following specific steps:
Figure FDA00033606420400000211
wherein J represents the loss function in the GINs network;
Figure FDA00033606420400000212
represents the original loss of task O;
Figure FDA00033606420400000213
a matrix A representing the transfer loss of information transfer and forcing the photovoltaic power generation system to twin models1,A2Close to each other.
9. The method of any one of claims 1 to 8, wherein said directing the photovoltaic power generation system to make a short term power prediction comprises:
and judging the magnitude of the power prediction error delta u of the photovoltaic power generation system and the reason causing the error, and optimizing the photovoltaic power generation system according to the error reason so as to complete short-term power prediction of the photovoltaic power generation system.
10. The method of claim 9, wherein the twin model learns knowledge interaction with the photovoltaic power generation system through GINs to transfer relational knowledge from the photovoltaic power generation system to the twin model and to direct the photovoltaic power generation system to make short term power predictions, further comprising:
enabling the twin model and the photovoltaic power generation system to perform knowledge interactive learning through GINs so as to update the twin model in real time; and the number of the first and second groups,
and correcting and updating the twin model according to the error reason.
CN202111366036.8A 2021-11-18 2021-11-18 Short-term power prediction method for photovoltaic power station Pending CN114154688A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111366036.8A CN114154688A (en) 2021-11-18 2021-11-18 Short-term power prediction method for photovoltaic power station

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111366036.8A CN114154688A (en) 2021-11-18 2021-11-18 Short-term power prediction method for photovoltaic power station

Publications (1)

Publication Number Publication Date
CN114154688A true CN114154688A (en) 2022-03-08

Family

ID=80456854

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111366036.8A Pending CN114154688A (en) 2021-11-18 2021-11-18 Short-term power prediction method for photovoltaic power station

Country Status (1)

Country Link
CN (1) CN114154688A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115983494A (en) * 2023-02-10 2023-04-18 广东工业大学 Short-term wind power prediction method and system for newly-built small-sample wind power plant
CN116961575A (en) * 2023-09-21 2023-10-27 中国电建集团贵阳勘测设计研究院有限公司 Tea light complementary photovoltaic power station monitoring system and method based on digital twin

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020202184A1 (en) * 2019-03-29 2020-10-08 Helios Iot Systems Private Limited A system employing electrical digital twin for solar photovoltaic power plant
CN112183441A (en) * 2020-10-13 2021-01-05 黄日光 Photovoltaic power station guardrail integrity detection method and detection system based on image processing
CN113075940A (en) * 2021-03-24 2021-07-06 阳光电源（上海）有限公司 Photovoltaic string tracking support control method and related device
CN113092115A (en) * 2021-04-09 2021-07-09 重庆大学 Digital twin model construction method of digital-analog combined drive full-life rolling bearing

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020202184A1 (en) * 2019-03-29 2020-10-08 Helios Iot Systems Private Limited A system employing electrical digital twin for solar photovoltaic power plant
CN112183441A (en) * 2020-10-13 2021-01-05 黄日光 Photovoltaic power station guardrail integrity detection method and detection system based on image processing
CN113075940A (en) * 2021-03-24 2021-07-06 阳光电源（上海）有限公司 Photovoltaic string tracking support control method and related device
CN113092115A (en) * 2021-04-09 2021-07-09 重庆大学 Digital twin model construction method of digital-analog combined drive full-life rolling bearing

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
PAN H, DOU Z, CAI Y, ET AL.: "Digital twin and its application in power system", 2020 5TH INTERNATIONAL CONFERENCE ON POWER AND RENEWABLE ENERGY (ICPRE), 31 December 2020 (2020-12-31), pages 21 - 26 *
孙荣富等: "基于数字孪生的光伏发电功率超短期预测", 电网技术, vol. 45, no. 4, 20 January 2021 (2021-01-20), pages 1258 - 1264 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115983494A (en) * 2023-02-10 2023-04-18 广东工业大学 Short-term wind power prediction method and system for newly-built small-sample wind power plant
CN115983494B (en) * 2023-02-10 2023-09-12 广东工业大学 Short-term wind power prediction method and system for newly-built small-sample wind power plant
CN116961575A (en) * 2023-09-21 2023-10-27 中国电建集团贵阳勘测设计研究院有限公司 Tea light complementary photovoltaic power station monitoring system and method based on digital twin
CN116961575B (en) * 2023-09-21 2023-12-01 中国电建集团贵阳勘测设计研究院有限公司 Tea light complementary photovoltaic power station monitoring system and method based on digital twin

Similar Documents

Publication Publication Date Title
CN109871995B (en) Quantum optimization parameter adjusting method for distributed deep learning under Spark framework
CN108694467B (en) Method and system for predicting line loss rate of power distribution network
CN114154688A (en) Short-term power prediction method for photovoltaic power station
CN111242282A (en) Deep learning model training acceleration method based on end edge cloud cooperation
CN113537514A (en) High-energy-efficiency federal learning framework based on digital twins
CN110070228B (en) BP neural network wind speed prediction method for neuron branch evolution
CN113988449B (en) Wind power prediction method based on transducer model
CN110738344A (en) Distributed reactive power optimization method and device for load prediction of power system
CN111523648B (en) Neural network pulse synchronization method and system containing clustering topological coupling
WO2023000807A1 (en) Fast and flexible holomorphic embedding optimal power flow evaluation method for power system
CN109858798B (en) Power grid investment decision modeling method and device for correlating transformation measures with voltage indexes
CN111799808B (en) Voltage distributed control method and system based on multi-agent deep reinforcement learning
CN114970351A (en) Power grid flow adjustment method based on attention mechanism and deep reinforcement learning
CN114358520A (en) Method, system, device and medium for economic dispatching decision of power system
CN116957698A (en) Electricity price prediction method based on improved time sequence mode attention mechanism
CN112101626A (en) Distributed photovoltaic power generation power prediction method and system
Dong et al. Automatic design of arithmetic operation spiking neural P systems
CN113807040A (en) Optimal design method for microwave circuit
Morales-Hernández et al. Online learning of windmill time series using Long Short-term Cognitive Networks
CN116579375A (en) Data-driven unit combination decision method, system, equipment and medium
CN114859847B (en) Reliable optimal control system and method suitable for interconnection nonlinear system
CN115719113A (en) Intelligent power grid economic dispatching distributed accelerated optimization method based on directed imbalance topology
CN116256970A (en) Data-driven cloud edge cooperative control method and system based on disturbance observer
CN113132482B (en) Distributed message system parameter adaptive optimization method based on reinforcement learning
CN112016684B (en) Electric power terminal fingerprint identification method of deep parallel flexible transmission 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