CN113422371A - Distributed power supply local voltage control method based on graph convolution neural network - Google Patents

Distributed power supply local voltage control method based on graph convolution neural network Download PDF

Info

Publication number
CN113422371A
CN113422371A CN202110839908.1A CN202110839908A CN113422371A CN 113422371 A CN113422371 A CN 113422371A CN 202110839908 A CN202110839908 A CN 202110839908A CN 113422371 A CN113422371 A CN 113422371A
Authority
CN
China
Prior art keywords
distribution network
power distribution
training
power supply
neural network
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
Application number
CN202110839908.1A
Other languages
Chinese (zh)
Other versions
CN113422371B (en
Inventor
赵金利
张子麒
习伟
李鹏
冀浩然
于浩
蔡田田
邓清唐
陈波
李肖博
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.)
Tianjin University
Southern Power Grid Digital Grid Research Institute Co Ltd
Original Assignee
Tianjin University
Southern Power Grid Digital Grid Research Institute 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 Tianjin University, Southern Power Grid Digital Grid Research Institute Co Ltd filed Critical Tianjin University
Priority to CN202110839908.1A priority Critical patent/CN113422371B/en
Publication of CN113422371A publication Critical patent/CN113422371A/en
Application granted granted Critical
Publication of CN113422371B publication Critical patent/CN113422371B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/12—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load
    • H02J3/16—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load by adjustment of reactive power
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/003—Load forecast, e.g. methods or systems for forecasting future load demand
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/381—Dispersed generators
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
    • H02J2300/20—The dispersed energy generation being of renewable origin
    • H02J2300/22—The renewable source being solar energy
    • H02J2300/24—The renewable source being solar energy of photovoltaic origin
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
    • H02J2300/20—The dispersed energy generation being of renewable origin
    • H02J2300/28—The renewable source being wind energy
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E10/00—Energy generation through renewable energy sources
    • Y02E10/50—Photovoltaic [PV] energy
    • Y02E10/56—Power conversion systems, e.g. maximum power point trackers

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Power Engineering (AREA)
  • Theoretical Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

A distributed power supply local voltage control method based on a graph convolution neural network divides the administered area of each edge computing device according to a selected active power distribution network and inputs power distribution network parameters; according to the parameters of the power distribution network and historical tide data of a typical day, a training set of a power distribution network proxy model based on a graph convolution neural network is established at the cloud end, the power distribution network proxy model based on the graph convolution neural network is established, and the power distribution network proxy model is obtained after training; the trained power distribution network agent model is sent to edge computing devices on the edge side of the distribution network to serve as power distribution network sub-agent models of all the edge computing devices; setting local voltage control curve parameters of the distributed power supplies in each region according to each distribution network sub-agent model; and outputting a local voltage control curve setting result. The method fully excavates massive multi-source data information, improves the intelligent level of operation regulation and control of the power distribution network, and effectively solves the problem that accurate full-network parameters are difficult to obtain in the process of setting the Q-V curve.

Description

Distributed power supply local voltage control method based on graph convolution neural network
Technical Field
The invention relates to a distributed power supply local voltage control method. In particular to a distributed power supply local voltage control method based on a graph convolution neural network.
Background
The large-scale and high-proportion access of Distributed Generators (DGs) realizes the low carbon of the energy supply of the distribution network, but also causes the problem of voltage out-of-limit of the distribution network, and makes the optimization and control mode of the distribution network more complicated. Meanwhile, with the fact that large power distribution and power distribution data information is more and more numerous, data of multiple benefit subjects on the distribution network side exist in an isolated island mode, and the data concentration faces many problems of communication pressure, privacy disclosure and the like, so that the revolution of a power distribution network operation management mechanism is accelerated.
By scheduling various reactive devices, voltage violations can be mitigated. Fast voltage control is difficult because conventional regulation devices (e.g., on-load tap changers and capacitor banks) have slow response speeds and voltage cannot be continuously regulated. The remaining capacity of the distributed power inverter can be used for continuous regulation of the voltage, so that there is great potential to achieve fast voltage control using a DG interface inverter.
For a power distribution network with a distributed power supply with high penetration rate access, a centralized control mode is generally adopted to uniformly allocate controllable resources to realize global optimization of the system, but the method has large measurement data volume and heavy communication burden. In order to reduce the communication data volume, the distributed control mode realizes the integral optimization of the system operation by reasonably partitioning the power distribution network and interacting boundary information between adjacent areas, but in the method, a large amount of information communication is still required between area controllers, and a final optimization scheme can be determined by a plurality of iterative processes. Compared with a centralized control mode and a distributed control mode, the local control mode can complete the control of the local adjustable resources by only utilizing local information, and has the advantages of high response speed, low investment cost and small communication data volume. The Q-V curve is used for carrying out local control, so that quick response to uncertain output of the distributed power supply can be realized, and the robustness is good. Most of the current researches set the Q-V curve in a mode of solving an optimization model based on the whole network parameters and accurate prediction data, and accurate network parameters are difficult to obtain in the actual operation of a power distribution network. Meanwhile, the running state of the power distribution network changes along with the change of the output of the distributed power supply, so that the anti-interference capability of the model is poor, and the advantage of local control is reduced to a certain extent.
The rapid development of information and communication technologies has increased the level of digitization of power distribution networks. A large amount of heterogeneous operating data can be measured and acquired by using a data acquisition and monitoring device of the power distribution network. How to fully mine mass multi-source data information becomes a key for improving the intelligent level of operation regulation and control of the power distribution network. Data-driven technology has attracted a lot of attention as a method for achieving optimal control based on only historical data without requiring an accurate model. The artificial intelligence is used as a branch of the data-driven technology to analyze and learn the collected historical data, and valuable information is extracted from the historical data and used for guiding the operation of the power distribution network. The Graph Convolutional Neural network (GCN) is a new multi-layer image data processing framework capable of processing non-european space structures, and in recent years, has many application cases in many fields such as computer vision, chemical engineering and the like.
Disclosure of Invention
The invention aims to solve the technical problem of providing a distributed power supply local voltage control method based on a graph convolution neural network, which can fully mine massive multi-source data information and improve the intelligent level of operation regulation and control of a power distribution network.
The technical scheme adopted by the invention is as follows: a distributed power supply local voltage control method based on a graph convolution neural network comprises the following steps:
1) dividing the region under the jurisdiction of each edge computing device according to the selected active power distribution network, and inputting parameters of the power distribution network, including the network topology connection relation of the power distribution network, the access positions, the capacities and the parameters of the distributed power supplies; input distribution network NdHistorical power flow data of a typical day, and a predicted sunrise output curve and a predicted sunrise load output curve of the distributed power supply; setting a typical topological total number NsThe total sampling time length delta T and the sampling time interval delta T of the training object; setting parameters of a power distribution network proxy model based on a graph convolution neural network, wherein the parameters comprise a learning rate gamma, a regularization coefficient beta and times E of completely traversing a training set;
2) according to the parameters of the power distribution network in the step 1) and NdHistorical load flow data of a typical day, constructing a training set of a power distribution network agent model based on a graph convolution neural network at a cloud end, and determining the number of training objects in the training set, the composition of each training object and preprocessing the training objects; constructing a power distribution network proxy model based on a graph convolution neural network at the cloud end, and finishing the training of the power distribution network proxy model based on the graph convolution neural network to obtain the trained power distribution network proxy model;
3) the trained power distribution network agent model is sent to edge computing devices on the edge side of the distribution network to serve as power distribution network sub-agent models of all the edge computing devices;
4) according to each distribution network subagent model, on the basis of a predicted distributed power supply daily output curve and daily load output curve and voltage and net load power measurement information in the area governed by each edge computing device, on the edge side of a distribution network, setting local voltage control curve parameters of the distributed power supply in each area by taking the minimum node voltage deviation in the area governed by each edge computing device as a target function;
5) and outputting the local voltage control curve setting result in the step 4).
The invention discloses a distributed power supply local voltage control method based on a graph convolution neural network, which aims to solve the problem of local control strategy calculation of a distributed power supply under the condition that network parameters of the power distribution network are unknown or inaccurate, constructs a power distribution network proxy model based on the graph convolution neural network for simulating the dynamic characteristics of the power distribution network, and sets local control curve parameters of the distributed power supply by taking the minimum voltage deviation in an active power distribution network area as a target function to obtain a local voltage reactive power control strategy of the distributed power supply. The method fully excavates massive multi-source data information, improves the intelligent level of operation regulation and control of the power distribution network, and effectively solves the problem that accurate full-network parameters are difficult to obtain in the process of setting the Q-V curve.
Drawings
FIG. 1 is a flow chart of a distributed power supply in-situ voltage control method based on a graph convolution neural network according to the invention;
FIG. 2 is a modified IEEE 33 node distribution network example structure diagram;
FIG. 3 is a predicted distributed power, load out curve;
FIG. 4 is a comparison graph of the distribution of extreme values of system voltages for case I and case II;
FIG. 5 is a comparison graph of the distribution of extreme voltage values for the systems of case II and case III.
Detailed Description
The present invention relates to a distributed power supply local voltage control method based on a graph convolution neural network, and is described in detail below with reference to the embodiments and the accompanying drawings.
The invention discloses a distributed power supply local voltage control method based on a graph convolution neural network, which comprises the following steps as shown in figure 1:
1) dividing the region under the jurisdiction of each edge computing device according to the selected active power distribution network, and inputting parameters of the power distribution network, including the network topology connection relation of the power distribution network, the access positions, the capacities and the parameters of the distributed power supplies; input distribution network NdHistorical power flow data of a typical day, and a predicted sunrise output curve and a predicted sunrise load output curve of the distributed power supply; setting a typical topological total number NsThe total sampling time length delta T and the sampling time interval delta T of the training object; setting parameters of a power distribution network proxy model based on a graph convolution neural network, wherein the parameters comprise a learning rate gamma, a regularization coefficient beta and times E of completely traversing a training set; wherein the content of the first and second substances,
(1) the distribution network NdHistorical trend data for a typical day is: in the distribution network NdDuring a typical day of the day,
Figure BDA0003174461830000031
a typical day distributed power supply does not participate in reactive power regulation,
Figure BDA0003174461830000032
a typical daily distributed power supply is given as [1.0,1.0 ]]Adjusting reactive power for the base curve of the dead zone to obtain distribution network NdHistorical trend data for a typical day.
For this embodiment, the modified IEEE 33 node power distribution network example structure is shown in fig. 2, and the topological connection relationship of the power distribution network, the access positions, capacities, and parameters of the distributed power sources, the predicted distributed power sources, the load output curves, and the historical power flow data of the power distribution network are input, and the detailed parameters are shown in tables 1 and 2.
TABLE 1 IEEE 33 node sample load Access location and Power
Figure BDA0003174461830000033
TABLE 2 IEEE 33 node example line parameters
Figure BDA0003174461830000034
Figure BDA0003174461830000041
The node 9, the node 10, the node 11, the node 18, the node 20, the node 21, the node 23, the node 24, the node 25, the node 31, the node 32 and the node 33 are respectively connected to a group of photovoltaic systems, and the capacities are all 0.18 MVA; the node 15, the node 16, the node 17, the node 22, the node 29 and the node 30 are respectively connected to a group of fans, and the capacities are all 0.36 MVA; the predicted distributed power supply and load output curve is shown in fig. 3; the upper and lower safe operation limits of the voltage amplitude (per unit value) of each node are respectively 1.10 and 0.90; setting typical number of days NdTotal number of topologies N200 s2, the total sampling time duration Δ T of the training object is 24h, and the sampling time interval Δ T is 5 min.
2) According to the parameters of the power distribution network in the step 1) and NdHistorical load flow data of a typical day, constructing a training set of a power distribution network agent model based on a graph convolution neural network at a cloud end, and determining the number of training objects in the training set, the composition of each training object and preprocessing the training objects; constructing a power distribution network proxy model based on a graph convolution neural network at the cloud end, and finishing the training of the power distribution network proxy model based on the graph convolution neural network to obtain the trained power distribution network proxy model; wherein the content of the first and second substances,
(1) the number of training objects in the training set is based on the distribution network NdHistorical trend data of a typical day, and considering NsA typical topological structure is used for constructing a training set of a power distribution network agent model based on a graph convolution neural network, and the training set comprises
Figure BDA0003174461830000042
And (4) training the subject.
(2) The composition of each training object comprises:
each training object consists of three matrixes, namely a characteristic information matrix X, a topological information matrix A and a label matrix Z, and a power distribution network agent model based on a graph convolution neural network takes the characteristic information matrix X and the topological information matrix A as input to fit the label matrix Z;
the feature information matrix X represents the input features of the training objects, and the feature information matrix X of the nth training objectnThe system consists of net load active power and net load reactive power injected by a sampling moment node and a node voltage value of a power distribution network, and is expressed as follows:
Figure BDA0003174461830000043
in the formula (I), the compound is shown in the specification,
Figure BDA0003174461830000051
a column vector representing the net load active power injected by each node at the sampling instant,
Figure BDA0003174461830000052
a column vector representing the net load reactive power injected by each node at the sampling instant,
Figure BDA0003174461830000053
representing a column vector formed by node voltage values of the power distribution network at the sampling moment, wherein N is the number of nodes of the power distribution network;
the topology information matrix A represents the connection relationship between the nodes, and the topology information matrix An of the nth training object is represented as:
Figure BDA0003174461830000054
Figure BDA0003174461830000055
in the formula (I), the compound is shown in the specification,
Figure BDA0003174461830000056
topological information matrix A for the n-th training objectnRow i and column j elements of (1);
the label matrix Z represents the label values of the output features of the training objects, and the label matrix Z of the nth training objectnThe node voltage control value at the sampling moment is represented as:
Figure BDA0003174461830000057
in the formula (I), the compound is shown in the specification,
Figure BDA0003174461830000058
representing the label value of the output characteristic of the ith node of the nth training object for the ith row element of the matrix; and N is the number of nodes of the power distribution network.
(3) The preprocessing of the training objects is to preprocess each element in a characteristic information matrix X and a label matrix Z of the training objects in a training set before model training; wherein
Characteristic information matrix X for n-th training objectnCarrying out pretreatment, wherein the pretreatment formula is as follows:
Figure BDA0003174461830000059
in the formula (I), the compound is shown in the specification,
Figure BDA00031744618300000510
characteristic information matrix X of n training objects before preprocessingnThe kth input feature of the ith node,
Figure BDA00031744618300000511
the characteristic information matrix X of the n-th training object after preprocessingnThe kth input feature of the ith node,
Figure BDA00031744618300000512
is the mean value of all elements of the kth input feature in the feature information matrix X of the training object,
Figure BDA00031744618300000513
the variance of all elements of the kth input feature in a feature information matrix X of a training object;
label matrix Z for nth training objectnCarrying out pretreatment, wherein the pretreatment formula is as follows:
Figure BDA00031744618300000514
in the formula (I), the compound is shown in the specification,
Figure BDA00031744618300000515
label matrix Z for n training object before preprocessingnThe ith node in the tree outputs the tag value of the feature,
Figure BDA00031744618300000516
label matrix Z for n training object after pretreatmentnThe tag value of the output characteristic of the ith node, muZIs the mean value, delta, of all elements of the output features in the label matrix Z of the training objectZAnd outputting the variance of all elements of the features in the label matrix Z of the training object.
(4) The distribution network agent model based on the graph convolution neural network is represented as follows:
Figure BDA00031744618300000517
Figure BDA0003174461830000061
in the formula, H(l+1)Outputting a (l + 1) th hidden layer of the power distribution network agent model; h(l)Is the output of the l-th hidden layer; a is a topological information matrix,
Figure BDA0003174461830000062
IMis of order MAn identity matrix;
Figure BDA0003174461830000063
is a diagonal matrix and is characterized in that,
Figure BDA0003174461830000064
W(l)is the weight matrix of the l layer; y is an output matrix of the graph convolution neural network; h(k-1)The output of the last hidden layer; w(k)Is the weight of the output layer; σ (-) is a nonlinear activation function, expressed as:
σ(x)＝LeakyReLU(x,α)＝max(0,x)+α×min(0,x) (9)
wherein α is a negative slope; f (-) is the output layer function, F (-) 1 (-);
the power distribution network agent model based on the graph convolution neural network takes the minimum loss function as a target function and a batch gradient descent algorithm as a training mode, and the training process is expressed as follows:
Figure BDA0003174461830000065
Figure BDA0003174461830000066
in the formula (I), the compound is shown in the specification,
Figure BDA0003174461830000067
the method is a loss function of a power distribution network proxy model based on a graph convolution neural network when a weight coefficient is omega; yin is the output matrix Y of the nth training objectnOutputting a fitting value of the characteristic by the ith node;
Figure BDA0003174461830000069
label matrix Z for n training object after pretreatmentnThe ith node outputs the label value of the characteristic; n is a radical ofTThe number of the training objects is N, and the number of the nodes of the power distribution network is N; omega is the current weight coefficient of the distribution network proxy model based on the graph convolution neural network, and omegalIs based onAnd the ith weight coefficient of the power distribution network proxy model of the graph convolution neural network is the learning rate.
3) And issuing the trained power distribution network agent model to the edge computing devices at the edge side of the distribution network to serve as the power distribution network sub-agent models of all the edge computing devices.
4) According to each distribution network subagent model, on the basis of the predicted distributed power supply daily output curve and daily load output curve and voltage and net load power measurement information in the area governed by each edge computing device, the node voltage deviation minimum in the area governed by each edge computing device is taken as a target function, and local voltage control curve parameters of the distributed power supply in each area are respectively set. Wherein the content of the first and second substances,
(4.1) the local voltage control curve of the distributed power supply is obtained by adopting the following formula:
Figure BDA00031744618300000610
in the formula, Vt，iThe voltage amplitude of node i, g (V), for a period of tt,i) For the expression of the distributed power supply in-situ voltage reactive power control strategy, g (V)t,i) With dead zone of regulation [ V ]i q,min,Vi q,max]，Vi q,minAnd Vi q,maxThe reactive power control method is characterized in that the reactive power control method is a local voltage control curve parameter of the distributed power supply, and the reactive power generated by the distributed power supply in the regulation dead zone is 0.
(4.2) the minimum node voltage deviation in the area governed by each edge computing device is taken as an objective function, and the objective function is expressed as:
Figure BDA0003174461830000071
or
Figure BDA0003174461830000072
Wherein f is a nodeDeviation of voltage, NtIs the number of time slices, NaCalculating the total number of nodes in the area covered by the device for the a-th edge; vt,iThe voltage amplitude of the node i is t time period;
Figure BDA0003174461830000073
in order to be the maximum voltage threshold value,
Figure BDA0003174461830000074
at minimum voltage threshold, when Vt,iOut of desired voltage interval
Figure BDA0003174461830000075
The objective function is used to reduce the voltage deviation.
(4.3) the setting of the local voltage control curve parameters of the distributed power supplies in each area comprises the following steps:
(4.3.1) selecting upper and lower limits V of voltage regulation dead zone of distributed power supplyi q,minAnd Vi q,maxIs [0.9,1.1 ]]And the constraint conditions of the upper limit and the lower limit of the dead zone are expressed as follows:
Figure BDA0003174461830000076
(4.3.2) based on the upper and lower limits V of the dead zonei q,minAnd Vi q,maxAnd power distribution network measurement information in the area, and updating the characteristic information matrix to obtain an updated characteristic information matrix X';
(4.3.3) inputting the updated characteristic information matrix X' into a power distribution network subagent model to obtain system voltage distribution in an area corresponding to the upper limit and the lower limit of the dead zone, and calculating the node voltage deviation, wherein the formula is as follows:
Figure BDA0003174461830000077
or
Figure BDA0003174461830000078
In the formula, NtIs the number of time slices, NaCalculating the total number of nodes in the area covered by the device for the a-th edge; vt,iThe voltage amplitude of the node i is t time period;
Figure BDA0003174461830000079
in order to be the maximum voltage threshold value,
Figure BDA00031744618300000710
at minimum voltage threshold, when Vt,iOut of desired voltage interval
Figure BDA00031744618300000711
The objective function is used to reduce the voltage deviation.
(4.3.4) scanning each dead zone upper and lower limit combination, and outputting the dead zone upper and lower limit combination with the minimum node voltage deviation as an optimal solution, namely the local voltage control curve parameter of the distributed power supply in the region.
5) And outputting the local voltage control curve setting result in the step 4).
In order to fully verify the advancement of the distributed power supply local voltage control method based on the graph convolution neural network, in this embodiment, the following three schemes are adopted for comparative analysis:
scheme I: the reactive power output of the distributed power supply is not optimized, and the initial running state of the power distribution network is obtained;
scheme II: the method is adopted to set the local voltage control curve of the distributed power supply in a partitioning manner and optimize the output of the distributed power supply;
scheme III: and the local voltage control curve of the distributed power supply is set in a partitioning manner by adopting a centralized optimization method, and the output of the distributed power supply is optimized.
The comparison of the optimization results of the scheme I, the scheme II and the scheme III is shown in the table 3, the comparison of the extreme value distribution of the system voltage of the scheme I and the scheme II is shown in the figure 4, and the comparison of the extreme value distribution of the system voltage of the scheme II and the scheme III is shown in the figure 5.
TABLE 3 comparison of simulation results under different control strategies
Figure BDA00031744618300000712
Figure BDA0003174461830000081
The computer hardware environment for executing the optimization calculation is Intel (R) core (TM) i7-9750H CPU, the main frequency is 2.60GHz, and the memory is 16 GB; the software environment is a Windows 10 operating system.
As can be seen from the comparison of schemes I and II, when the control means is not used, the connection of the distributed power supply causes the system voltage to fluctuate dramatically. After the distributed power supply output is optimized by the distributed power supply local voltage control strategy calculation method based on the graph convolution neural network, each distributed power supply quickly adjusts reactive compensation in real time, and when the node voltage is lower, the distributed power supply supports voltage by sending out reactive power; when the node voltage is higher, the distributed power supply reduces the voltage by absorbing reactive power, so that the system voltage is maintained at a safe operation level.
As can be seen from the comparison of the schemes II and III, the result obtained by optimizing the output of the distributed power supply in a partitioning manner by using the distributed power supply in-situ voltage control method based on the graph convolution neural network is slightly inferior to the result obtained by optimizing in a partitioning manner by using a centralized setting method; compared with a centralized setting method, the distributed power supply local voltage control method based on the graph convolution neural network effectively avoids the problem that the network parameters of the power distribution network are unknown or inaccurate in the optimization process.

Claims (9)

1. A distributed power supply local voltage control method based on a graph convolution neural network is characterized by comprising the following steps:
1) according to the selected active power distribution network, dividing the dominated region of each edge computing device and inputting parameters of the power distribution network, including the network topology connection relation of the power distribution network, the access positions, the capacities and the parameters of the distributed power supplies(ii) a Input distribution network NdHistorical power flow data of a typical day, and a predicted sunrise output curve and a predicted sunrise load output curve of the distributed power supply; setting a typical topological total number NsThe total sampling time length delta T and the sampling time interval delta T of the training object; setting parameters of a power distribution network proxy model based on a graph convolution neural network, wherein the parameters comprise a learning rate gamma, a regularization coefficient beta and times E of completely traversing a training set;
2) according to the parameters of the power distribution network in the step 1) and NdHistorical load flow data of a typical day, constructing a training set of a power distribution network agent model based on a graph convolution neural network at a cloud end, and determining the number of training objects in the training set, the composition of each training object and preprocessing the training objects; constructing a power distribution network proxy model based on a graph convolution neural network at the cloud end, and finishing the training of the power distribution network proxy model based on the graph convolution neural network to obtain the trained power distribution network proxy model;
3) the trained power distribution network agent model is sent to edge computing devices on the edge side of the distribution network to serve as power distribution network sub-agent models of all the edge computing devices;
4) according to each distribution network subagent model, on the basis of a predicted distributed power supply daily output curve and daily load output curve and voltage and net load power measurement information in the area governed by each edge computing device, on the edge side of a distribution network, setting local voltage control curve parameters of the distributed power supply in each area by taking the minimum node voltage deviation in the area governed by each edge computing device as a target function;
5) and outputting the local voltage control curve setting result in the step 4).
2. The distributed power supply local voltage control method based on the graph convolution neural network as claimed in claim 1, wherein the distribution network N in the step 1) isdHistorical trend data for a typical day is:
in the distribution network NdDuring a typical day of the day,
Figure FDA0003174461820000011
a typical day distributed power supply does not participate in reactive power regulation,
Figure FDA0003174461820000012
a typical daily distributed power supply is given as [1.0,1.0 ]]Adjusting reactive power for the base curve of the dead zone to obtain distribution network NdHistorical trend data for a typical day.
3. The distributed power supply local voltage control method based on the graph convolution neural network as claimed in claim 1, wherein the number of training objects in the training set in the step 2) is based on a distribution network NdHistorical trend data of a typical day, and considering NsA typical topological structure is used for constructing a training set of a power distribution network agent model based on a graph convolution neural network, and the training set comprises
Figure FDA0003174461820000013
And (4) training the subject.
4. The distributed power supply local voltage control method based on the graph convolution neural network as claimed in claim 1, wherein the composition of each training object in step 2) comprises:
each training object consists of three matrixes, namely a characteristic information matrix X, a topological information matrix A and a label matrix Z, and a power distribution network agent model based on a graph convolution neural network takes the characteristic information matrix X and the topological information matrix A as input to fit the label matrix Z;
the feature information matrix X represents the input features of the training objects, and the feature information matrix X of the nth training objectnThe system consists of net load active power and net load reactive power injected by a sampling moment node and a node voltage value of a power distribution network, and is expressed as follows:
Figure FDA0003174461820000021
in the formula (I), the compound is shown in the specification,
Figure FDA0003174461820000022
a column vector representing the net load active power injected by each node at the sampling instant,
Figure FDA0003174461820000023
a column vector representing the net load reactive power injected by each node at the sampling instant,
Figure FDA0003174461820000024
representing a column vector formed by node voltage values of the power distribution network at the sampling moment, wherein N is the number of nodes of the power distribution network;
the topological information matrix A represents the connection relation between nodes, and the topological information matrix A of the n-th training objectnExpressed as:
Figure FDA0003174461820000025
Figure FDA0003174461820000026
in the formula (I), the compound is shown in the specification,
Figure FDA0003174461820000027
topological information matrix A for the n-th training objectnRow i and column j elements of (1);
the label matrix Z represents the label values of the output features of the training objects, and the label matrix Z of the nth training objectnThe node voltage control value at the sampling moment is represented as:
Figure FDA0003174461820000028
in the formula (I), the compound is shown in the specification,
Figure FDA0003174461820000029
representing the label value of the output characteristic of the ith node of the nth training object for the ith row element of the matrix; and N is the number of nodes of the power distribution network.
5. The distributed power supply in-place voltage control method based on the graph convolution neural network as claimed in claim 1, wherein the preprocessing of the training object in step 2) is to preprocess each element in a feature information matrix X and a label matrix Z of the training object in a training set before model training; wherein
Characteristic information matrix X for n-th training objectnCarrying out pretreatment, wherein the pretreatment formula is as follows:
Figure FDA00031744618200000210
in the formula (I), the compound is shown in the specification,
Figure FDA00031744618200000211
characteristic information matrix X of n training objects before preprocessingnThe kth input feature of the ith node,
Figure FDA00031744618200000212
the characteristic information matrix X of the n-th training object after preprocessingnThe kth input feature of the ith node,
Figure FDA00031744618200000213
is the mean value of all elements of the kth input feature in the feature information matrix X of the training object,
Figure FDA00031744618200000214
the variance of all elements of the kth input feature in a feature information matrix X of a training object;
label matrix Z for nth training objectnPerforming a pretreatmentThe preprocessing formula is as follows:
Figure FDA0003174461820000031
in the formula (I), the compound is shown in the specification,
Figure FDA0003174461820000032
label matrix Z for n training object before preprocessingnThe ith node in the tree outputs the tag value of the feature,
Figure FDA0003174461820000033
label matrix Z for n training object after pretreatmentnThe tag value of the output characteristic of the ith node, muZIs the mean value, delta, of all elements of the output features in the label matrix Z of the training objectZAnd outputting the variance of all elements of the features in the label matrix Z of the training object.
6. The distributed power supply local voltage control method based on the convolutional neural network of claim 1, wherein the proxy model of the power distribution network based on the convolutional neural network in the step 2) is expressed as:
Figure FDA0003174461820000034
Figure FDA0003174461820000035
in the formula, H(l+1)Outputting a (l + 1) th hidden layer of the power distribution network agent model; h(l)Is the output of the l-th hidden layer; a is a topological information matrix,
Figure FDA0003174461820000036
IMis an M-order identity matrix;
Figure FDA0003174461820000037
is a diagonal matrix and is characterized in that,
Figure FDA0003174461820000038
W(l)is the weight matrix of the l layer; y is an output matrix of the graph convolution neural network; h(k-1)The output of the last hidden layer; w(k)Is the weight of the output layer; σ (-) is a nonlinear activation function, expressed as:
σ(x)＝LeakyReLU(x，α)＝max(0，x)+α×min(0，x) (9)
wherein α is a negative slope; f (-) is the output layer function, F (-) 1 (-);
the power distribution network agent model based on the graph convolution neural network takes the minimum loss function as a target function and a batch gradient descent algorithm as a training mode, and the training process is expressed as follows:
Figure FDA0003174461820000039
Figure FDA00031744618200000310
in the formula (I), the compound is shown in the specification,
Figure FDA00031744618200000311
the method is a loss function of a power distribution network proxy model based on a graph convolution neural network when a weight coefficient is omega;
Figure FDA00031744618200000312
output matrix Y for the nth training objectnOutputting a fitting value of the characteristic by the ith node;
Figure FDA00031744618200000313
label matrix Z for n training object after pretreatmentnOutput bit of the ith nodeA signed tag value; n is a radical ofTThe number of the training objects is N, and the number of the nodes of the power distribution network is N; omega is the current weight coefficient of the distribution network proxy model based on the graph convolution neural network, and omegalThe first weight coefficient of the distribution network agent model based on the graph convolution neural network is gamma, and the gamma is the learning rate.
7. The distributed power supply local voltage control method based on the graph convolution neural network as claimed in claim 1, wherein the local voltage control curve of the distributed power supply in the step 4) is obtained by using the following formula:
Figure FDA0003174461820000041
in the formula, Vt,iThe voltage amplitude of node i, g (V), for a period of tt,i) For the expression of the distributed power supply in-situ voltage reactive power control strategy, g (V)t,i) With dead zone of regulation [ V ]i q,min,Vi q,max]，Vi q,minAnd Vi q,maxThe reactive power control method is characterized in that the reactive power control method is a local voltage control curve parameter of the distributed power supply, and the reactive power generated by the distributed power supply in the regulation dead zone is 0.
8. The distributed power supply local voltage control method based on the graph convolutional neural network as claimed in claim 1, wherein the step 4) takes the minimum node voltage deviation in the area covered by each edge computing device as an objective function, and is represented as:
Figure FDA0003174461820000042
wherein f is the node voltage deviation, NtIs the number of time slices, NaCalculating the total number of nodes in the area covered by the device for the a-th edge; vt，iIs at t timeThe voltage amplitude of segment node i;
Figure FDA0003174461820000043
in order to be the maximum voltage threshold value,
Figure FDA0003174461820000044
at minimum voltage threshold, when Vt，iOut of desired voltage interval
Figure FDA0003174461820000045
The objective function is used to reduce the voltage deviation.
9. The distributed power supply local voltage control method based on the graph convolution neural network as claimed in claim 1, wherein the step 4) of setting the local voltage control curve parameters of the distributed power supply in each region comprises:
(4.1) selecting upper and lower limits V of voltage regulation dead zone of distributed power supplyi q,minAnd Vi q,maxIs [0.9,1.1 ]]And the constraint conditions of the upper limit and the lower limit of the dead zone are expressed as follows:
Figure FDA0003174461820000046
(4.2) based on the upper and lower limits V of the dead zonei q,minAnd Vi q,maxAnd power distribution network measurement information in the area, and updating the characteristic information matrix to obtain an updated characteristic information matrix X';
(4.3) inputting the updated characteristic information matrix X' into a power distribution network subagent model to obtain system voltage distribution in an area corresponding to the upper limit and the lower limit of the dead zone, and calculating the node voltage deviation, wherein the formula is as follows:
Figure FDA0003174461820000047
in the formula, NtIs the number of time slices, NaCalculating the total number of nodes in the area covered by the device for the a-th edge; vt,iThe voltage amplitude of the node i is t time period;
Figure FDA0003174461820000048
in order to be the maximum voltage threshold value,
Figure FDA0003174461820000049
at minimum voltage threshold, when Vt,iOut of desired voltage interval
Figure FDA0003174461820000051
The objective function is used to reduce the voltage deviation.
And (4.4) scanning each dead zone upper and lower limit combination, and outputting the dead zone upper and lower limit combination with the minimum node voltage deviation as an optimal solution, namely the local voltage control curve parameters of the distributed power supply in the region.
CN202110839908.1A 2021-07-22 2021-07-22 Distributed power supply local voltage control method based on graph convolution neural network Active CN113422371B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110839908.1A CN113422371B (en) 2021-07-22 2021-07-22 Distributed power supply local voltage control method based on graph convolution neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110839908.1A CN113422371B (en) 2021-07-22 2021-07-22 Distributed power supply local voltage control method based on graph convolution neural network

Publications (2)

Publication Number Publication Date
CN113422371A true CN113422371A (en) 2021-09-21
CN113422371B CN113422371B (en) 2022-05-20

Family

ID=77718199

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110839908.1A Active CN113422371B (en) 2021-07-22 2021-07-22 Distributed power supply local voltage control method based on graph convolution neural network

Country Status (1)

Country Link
CN (1) CN113422371B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114597970A (en) * 2022-03-18 2022-06-07 合肥工业大学 Active power distribution network partitioning method based on graph convolution network
CN116667369A (en) * 2023-08-01 2023-08-29 国网山西省电力公司晋城供电公司 Distributed photovoltaic voltage control method based on graph convolution neural network

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106972539A (en) * 2017-05-13 2017-07-21 天津大学 A kind of distributed power source voltage control strategy setting method on the spot based on cone planning
CN111541685A (en) * 2020-04-20 2020-08-14 南京大学 Edge cloud anomaly detection method based on network structure learning
CN112651436A (en) * 2020-12-23 2021-04-13 深圳市安软科技股份有限公司 Optimization method and device based on uncertain weight graph convolution neural network

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106972539A (en) * 2017-05-13 2017-07-21 天津大学 A kind of distributed power source voltage control strategy setting method on the spot based on cone planning
CN111541685A (en) * 2020-04-20 2020-08-14 南京大学 Edge cloud anomaly detection method based on network structure learning
CN112651436A (en) * 2020-12-23 2021-04-13 深圳市安软科技股份有限公司 Optimization method and device based on uncertain weight graph convolution neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
AIDONG XU: "Graph-Based Time Series Edge Anomaly Detection in Smart Grid", 《IEEE》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114597970A (en) * 2022-03-18 2022-06-07 合肥工业大学 Active power distribution network partitioning method based on graph convolution network
CN114597970B (en) * 2022-03-18 2024-03-08 合肥工业大学 Active power distribution network partitioning method based on graph convolution network
CN116667369A (en) * 2023-08-01 2023-08-29 国网山西省电力公司晋城供电公司 Distributed photovoltaic voltage control method based on graph convolution neural network
CN116667369B (en) * 2023-08-01 2023-10-31 国网山西省电力公司晋城供电公司 Distributed photovoltaic voltage control method based on graph convolution neural network

Also Published As

Publication number Publication date
CN113422371B (en) 2022-05-20

Similar Documents

Publication Publication Date Title
CN113471982B (en) Cloud edge cooperation and power grid privacy protection distributed power supply in-situ voltage control method
CN110535146B (en) Electric power system reactive power optimization method based on depth determination strategy gradient reinforcement learning
CN108846517B (en) Integration method for predicating quantile probabilistic short-term power load
CN112564098B (en) High-proportion photovoltaic power distribution network voltage prediction method based on time convolution neural network
CN113422371B (en) Distributed power supply local voltage control method based on graph convolution neural network
CN113363998B (en) Power distribution network voltage control method based on multi-agent deep reinforcement learning
CN108599172B (en) Transmission and distribution network global load flow calculation method based on artificial neural network
CN110889535B (en) Multi-point wind speed prediction method in wind power plant based on convolution cyclic neural network
CN110070228B (en) BP neural network wind speed prediction method for neuron branch evolution
CN111384726A (en) High-permeability photovoltaic power distribution network partition voltage regulation method
CN111832838B (en) Method for predicting short-term wind power generation output power
CN111799808B (en) Voltage distributed control method and system based on multi-agent deep reinforcement learning
CN109242136A (en) A kind of micro-capacitance sensor wind power Chaos-Genetic-BP neural network prediction technique
CN111525587A (en) Reactive load situation-based power grid reactive voltage control method and system
CN115313403A (en) Real-time voltage regulation and control method based on deep reinforcement learning algorithm
CN113872213B (en) Autonomous optimization control method and device for power distribution network voltage
CN115149542A (en) Distributed power supply self-adaptive voltage control method based on multi-agent reinforcement learning
CN112052985B (en) Middle-short-term low-voltage prediction method based on lightgbm
CN113887794A (en) Method and device for reactive power optimization of power distribution network
CN117172097A (en) Power distribution network dispatching operation method based on cloud edge cooperation and multi-agent deep learning
CN108767867B (en) Distributed power supply in-situ voltage reactive power control strategy solving method based on meta-model
CN114039366B (en) Power grid secondary frequency modulation control method and device based on peacock optimization algorithm
CN115793456A (en) Lightweight sensitivity-based power distribution network edge side multi-mode self-adaptive control method
CN116054179A (en) Event-triggering-based reactive power preference control system and method for power system
Li et al. Model-informed generative adversarial network (mi-gan) for learning optimal power flow

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant