CN111756049B - Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network - Google Patents

Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network Download PDF

Info

Publication number
CN111756049B
CN111756049B CN202010560697.3A CN202010560697A CN111756049B CN 111756049 B CN111756049 B CN 111756049B CN 202010560697 A CN202010560697 A CN 202010560697A CN 111756049 B CN111756049 B CN 111756049B
Authority
CN
China
Prior art keywords
data
reactive power
distribution network
neural network
historical
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.)
Expired - Fee Related
Application number
CN202010560697.3A
Other languages
Chinese (zh)
Other versions
CN111756049A (en
Inventor
窦晓波
李鹏
田恩东
刘之涵
孙可
葛晓慧
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.)
Southeast University
Electric Power Research Institute of State Grid Zhejiang Electric Power Co Ltd
Original Assignee
Southeast University
Electric Power Research Institute of State Grid Zhejiang Electric Power 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 Southeast University, Electric Power Research Institute of State Grid Zhejiang Electric Power Co Ltd filed Critical Southeast University
Priority to CN202010560697.3A priority Critical patent/CN111756049B/en
Publication of CN111756049A publication Critical patent/CN111756049A/en
Application granted granted Critical
Publication of CN111756049B publication Critical patent/CN111756049B/en
Expired - Fee Related 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
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for AC mains or AC distribution networks
    • H02J3/18—Arrangements for adjusting, eliminating or compensating reactive power in networks
    • 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
    • H02J3/46—Controlling of the sharing of output between the generators, converters, or transformers
    • H02J3/466—Scheduling the operation of the generators, e.g. connecting or disconnecting generators to meet a given 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
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/46—Controlling of the sharing of output between the generators, converters, or transformers
    • H02J3/50—Controlling the sharing of the out-of-phase component
    • 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
    • H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
    • 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
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/30—Reactive power compensation

Landscapes

  • Engineering & Computer Science (AREA)
  • Power Engineering (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

本发明公开了一种计及配电网实时量测信息缺失的数据驱动无功优化方法，属于交流配电网运行控制技术领域。该方法首先直接利用配电网运行过程中产生的历史负荷数据，通过对历史数据预处理并使用优化算法生成对应历史时刻的无功优化策略。进而分别以可实时采集负荷节点的历史负荷数据和对应时刻的无功优化策略作为神经网络的输入训练集和输出训练集训练神经网络。在实时量测信息缺失的配电网进行日内无功优化时，以可实时采集的负荷数据作为神经网络输入，神经网络输出便为当前时刻无功优化策略。本发明技术方案实现了配电网实时负荷数据缺失情况下的无功优化，为配电网优化运行提供了一条新途径。

Figure 202010560697

The invention discloses a data-driven reactive power optimization method that takes into account the lack of real-time measurement information of a distribution network, and belongs to the technical field of AC distribution network operation control. In this method, the historical load data generated during the operation of the distribution network is directly used, and the reactive power optimization strategy corresponding to the historical moment is generated by preprocessing the historical data and using the optimization algorithm. Then, the neural network is trained by using the real-time collection of historical load data of load nodes and the reactive power optimization strategy at the corresponding moment as the input training set and output training set of the neural network. When the intraday reactive power optimization of the distribution network with missing real-time measurement information is performed, the load data that can be collected in real time is used as the input of the neural network, and the output of the neural network is the reactive power optimization strategy at the current moment. The technical scheme of the invention realizes the reactive power optimization under the condition that the real-time load data of the distribution network is missing, and provides a new way for the optimal operation of the distribution network.

Figure 202010560697

Description

Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network
Technical Field
The invention relates to the technical field of operation control of an alternating-current power distribution network, in particular to a data-driven reactive power optimization method for considering the loss of real-time measurement information of the power distribution network.
Background
Reactive power optimization refers to the purposes of improving reactive power flow distribution, and achieving minimum total network loss and minimum voltage deviation in a power distribution network by adjusting the positions of on-load tap changing transformers in the power distribution network, switching parallel capacitors and the like. More and more distributed power supplies are connected into the power distribution network, and due to the randomness and uncertainty of the distributed power supplies, the reactive power optimization difficulty of the power distribution network is increased, and meanwhile the reactive power optimization flexibility is also increased. The reactive power optimization problem is a multi-objective and multi-constraint nonlinear optimization problem in nature, so that the traditional optimization method or heuristic algorithms such as particle swarm optimization and genetic algorithm are mainly adopted at present.
However, the methods can be used only on the premise that the real-time measurement information of the power distribution network is not lost, otherwise, the load flow calculation cannot be carried out. However, in the prior art, most of the real-time measurement information of the power distribution network is lost, only part of nodes of the power distribution network are provided with load data real-time measurement devices, and the loads of part of nodes cannot be acquired in real time. Therefore, the reactive power optimization of the power distribution network with the missing real-time measurement information is difficult, and relatively deep research is not available at present. A method for controlling reactive voltage by obtaining a voltage of a non-real-time observation node by using a deep convolution network and then utilizing a gradient descent method to solve a model is provided in a document of a low-perceptibility power distribution network voltage control method containing high-proportion distributed power supply access published by Zhangjing and the like, and the distribution of node voltage is improved.
The invention provides a data-driven reactive power optimization method for considering the loss of real-time measurement information of a power distribution network on the basis of the existing research, the method realizes the reactive power optimization of the power distribution network when the real-time measurement information is lost, can meet the requirements of actual engineering, fully excavates the potential of historical data in the power distribution network, and provides a new way for the reactive power optimization of the power distribution network with the loss of the real-time measurement information at the present stage.
Disclosure of Invention
In order to solve the problem that the power distribution network with the real-time measurement information loss cannot be subjected to reactive power optimization by using the traditional optimization method, the invention provides a data-driven reactive power optimization method considering the real-time measurement information loss of the power distribution network, so that the reactive power optimization of the power distribution network under the condition of the real-time measurement information loss is realized. To achieve this object: the invention provides a data-driven reactive power optimization method considering the loss of real-time measurement information of a power distribution network, which comprises two stages of neural network training set construction before optimization, neural network training and application of a neural network during optimization, and comprises the following steps:
1) before actual optimization, collecting historical load data of a distribution network marketing system;
2) then, carrying out data preprocessing by using the data acquired in the step 1) to obtain a historical load database;
3) then, generating a corresponding time reactive power optimization strategy library by using an optimization algorithm aiming at historical load data;
4) then, taking historical load data of nodes capable of being measured in real time as a neural network input training set, and taking a reactive power optimization strategy at a corresponding moment as a neural network output training set to train the neural network;
5) when an optimization cycle begins, load information of nodes capable of being measured in real time is used as input of the neural network trained in the step 4), and output of the neural network is a reactive power optimization strategy at the current moment.
As a further improvement of the invention, in the step 1), the number of nodes of the power distribution network which can be measured in real time is NvThe number of nodes which cannot be measured in real time is NivAnd the total number of nodes of the power distribution network is N. The available distribution network marketing system can measure the historical active load data of the nodes in real time to be Pv＝[P1,P2,…PNv]Historical reactive load data is Qv＝[Q1,Q2,…QNv]. Historical active load data of nodes which cannot be measured in real time by the distribution network marketing system is Piv＝[P1,P2,…PNiv]Historical reactive load data is Qiv＝[Q1,Q2,…QNiv]. Then the total historical active load data PN＝Pv∪PivHistorical reactive load data QN＝Qv∪Qiv。
As a further improvement of the invention, the marketing system historical load data [ PN, QN ] acquired in the step 1) is utilized in the step 2) for preprocessing, so that the data quality is improved.
As a further improvement of the invention, the data preprocessing in the step 2) adopts the following method:
if PNTwo data P inml,nAnd Pmr,nData P with k deletions in betweenml+i,n(i ═ 1,2, 3 … k), the following equation is used:
Pml+i,n＝Pml,n+(Pmr,n-Pml,n)/(k+1)×i(i＝1，2，3...k)
if QNTwo data Q in (1)ml,nAnd Qmr,nData Q with k misses in betweenml+i,n(i ═ 1,2, 3 … k), the following equation is used:
Qml+i,n＝Qml,n+(Qmr,n-Qml,n)/(k+1)×i(i＝1，2，3...k)。
as a further improvement of the invention, in the step 3), by using the historical load data obtained in the step 2) and the topology of the power distribution network, a reactive power optimization strategy corresponding to the historical moment is obtained through a particle swarm algorithm;
the fitness function of the particle swarm is:
Figure BDA0002546179830000021
the constraint conditions are as follows:
Figure BDA0002546179830000022
Figure BDA0002546179830000023
Figure BDA0002546179830000024
Figure BDA0002546179830000025
the particle group velocity and position update formula is:
Vid＝ωVid+C1random(0,1)(Pid-Xid)+C2random(0,1)(Pgd-Xid)
Xid＝Xid+Vid
wherein VidDenotes the particle velocity, XidIndicating the position of the particle, PgdRepresenting the global optimum fitness, PidThe fitness of the particle is the fitness of the particle.
As a further improvement of the invention, the reactive power optimization strategy comprises but not only comprises the on-load tap changer gear TtapThe number of capacitor groups NcbPhotovoltaic reactive power output value QdgThese reactive power optimization strategies together form a strategy set S.
As a further improvement of the invention, the input part of the neural network training set in the step 4) is [ P ] obtained in the step 2)v,Qv]The output part is the strategy set S obtained in the step 3), and a reactive power optimization model for measuring the power distribution network with information missing in real time is obtained finally through training;
and (3) training by using a BP neural network, wherein the training process of the BP neural network comprises the following steps:
4.1) setting a learning rate and initializing each neuron weight w and a threshold, wherein the activation function f is a sigmoid function. Wherein, the weight between the ith neuron of the input layer and the h-th neuron of the hidden layer is vihThe weight between the h-th neuron of the hidden layer and the j-th neuron of the output layer is whj. Output layer jth neuron has a threshold value ofjThe h-th neuron of the hidden layer has a threshold value ofh. The h-th neuron in the hidden layer receives the input with the sum of
Figure BDA0002546179830000031
The j-th neuron of the output layer receives the input sum of
Figure BDA0002546179830000032
bhThe output of the h neuron of the hidden layer;
4.2) training samples (x)k,yk) Computing the output of the neural network at that time
Figure BDA0002546179830000033
And the mean square error, and the specific calculation formula is as follows:
Figure BDA0002546179830000034
Figure BDA0002546179830000035
4.3) updating neural network related parameters v (including w)hj、θj、vih、γh) The formula used is: v ═ v + Δ v 4.4) the specific calculation formula is:
Δwhj＝ηgjbh
Δθj＝-ηgj
Δvih＝ηehxi
Δγh＝-ηeh
Figure BDA0002546179830000036
Figure BDA0002546179830000041
4.5) judging whether a convergence condition is reached, if not, returning to the step 3), and if so, indicating that the training is finished.
As a further improvement of the present invention, in the step 5), the collectable real-time measurement load information is used as an input of the neural network reactive power optimization model obtained by training in the step 4), and an output is a reactive power optimization strategy at the current optimization time.
Compared with the prior art, the invention has the beneficial effects that:
(1) information in the historical load data of the power distribution network is fully mined and utilized, and a big data and artificial intelligence method is used, so that a reactive power optimization strategy is more accurate and effective;
(2) the reactive power characteristic of photovoltaic is actively utilized, so that the flexibility of the reactive power optimization method of the power distribution network is improved;
(3) the constructed neural network has short operation time when being used after being trained, and meets the requirement of online application.
Drawings
FIG. 1 is a flow chart of a reactive power optimization process;
FIG. 2 is a schematic diagram of a power distribution network including on-load tap changing transformers, capacitors, photovoltaic real-time measurement information loss;
fig. 3 is a schematic control diagram of the distribution network and the reactive power optimization module in temporal coordination;
FIG. 4 is a graph comparing the loss of the present method with the loss of the conventional method;
FIG. 5 is a graph comparing node voltages in the present method and the conventional method.
Detailed Description
The invention is described in further detail below with reference to the following detailed description and accompanying drawings:
the data-driven reactive power optimization method for considering the loss of the real-time measurement information of the power distribution network realizes reactive power optimization of the power distribution network when the real-time measurement information is lost, can meet the requirements of actual engineering, fully excavates the potential of historical data in the power distribution network, and provides a new approach for reactive power optimization of the power distribution network with the loss of the real-time measurement information at the present stage.
Example 1:
the following describes the embodiments of the present invention with reference to the accompanying drawings.
The embodiment provides a data-driven reactive power optimization method considering loss of real-time measurement information of a power distribution network, which can perform reactive power optimization on the power distribution network with load information nodes which cannot be acquired in real time at different moments, so as to achieve the purpose of minimizing network loss of the power distribution network.
As shown in fig. 1, the method comprises the following steps:
s1: an off-line training stage;
s2: and (5) an online optimization stage.
Wherein S1 comprises the steps of:
s1.1: collecting historical load data of a distribution network marketing system;
s1.2: carrying out data preprocessing by using the data acquired in the step S1.1 to obtain a historical load database;
s1.3: generating a corresponding time reactive power optimization strategy library by using an optimization algorithm aiming at historical load data;
s1.4: taking historical load data of nodes capable of being measured in real time as a neural network input training set, and taking a reactive power optimization strategy at a corresponding moment as a neural network output training set to be trained to obtain a neural network reactive power optimization module;
specifically, in step S1.1, the active historical load data P of each node may be obtainedNAnd reactive historical load data QN。
When the data in step S1.1 is processed in step S1.2, the data loss due to the sampling problem of the load data is mainly considered. If PNTwo data P inml,nAnd Pmr,nData P with k deletions in betweenml+i,n(i ═ 1,2, 3 … k), the data P is complemented as followsml+i,n(i＝1，2，3…k)：
Pml+i,n＝Pml,n+(Pmr,n-Pml,n)/(k+1)×i(i＝1，2，3...k)
For QNThe method described above may also be used for missing data.
The generation of the reactive power optimization strategy library in the step S1.3 mainly comprises the following steps:
s1.3.1, constructing a power distribution network structure model, wherein the specific method comprises the following steps:
acquiring the number N of nodes of the power distribution network, a node admittance matrix G and historical load data P of a node ii+jQi。
S1.3.2, establishing a reactive power optimization objective function and constraint conditions, specifically:
the objective function is:
Figure BDA0002546179830000051
in the formula PijRepresenting power flowing from node i to node j
The constraint conditions are as follows:
Figure BDA0002546179830000052
Figure BDA0002546179830000053
Figure BDA0002546179830000054
Figure BDA0002546179830000055
Figure BDA0002546179830000056
andU irespectively representing the upper limit and the lower limit allowed by the voltage of the node i;
Figure BDA0002546179830000058
andQ DGirespectively representing the upper limit and the lower limit of the DG reactive power output connected to the node i;
Figure BDA00025461798300000510
andQ SVCirepresenting the upper and lower output limits of the static var compensator at the node i;
Figure BDA00025461798300000512
andC irepresenting the upper and lower limits of the number of capacitor banks put into node i, respectively.
S1.3.3, solving a reactive power optimization strategy S under the conditions of S1.3.1 and S1.3.2 by using a particle swarm optimization:
s1.3.3.1 setting iteration times maxgen, the number sizepop of particles in the population, the particle velocity updating parameter c1c2 and the inertia weight wmax，wmin(ii) a Upper and lower bounds of constraint conditions, and upper and lower limits of particle velocity.
S1.3.3.2 the position and speed of the population particles are initialized and generated by random numbers.
S1.3.3.3 the fitness and particle position of the best particle in the population at that time, the particle velocity, and the fitness and position of the best particle in all cycles are obtained.
The fitness function of the particle swarm is the network loss:
Figure BDA0002546179830000061
s1.3.3.4, judging whether the difference between the fitness of the particle and the fitness of the optimal particle in all the cycles is less than the specified convergence value.
S1.3.3.5 if the convergence condition is satisfied then the loop is ended and if not, it is continued.
S1.3.3.6 updating the particle velocity and calculating the new position and fitness of each particle in the population, the particle velocity and position are updated using the following formula:
Vid＝ωVid+C1random(0,1)(Pid-Xid)+C2random(0,1)(Pgd-Xid)
Xid＝Xid+Vid
wherein VidDenotes the particle velocity, XidIndicating the position of the particle, PgdRepresenting the global optimum fitness, PidThe fitness of the particle is the fitness of the particle.
Go back to S1.3.3.3
The reactive power optimization strategy corresponding to each time point can be calculated through the method.
Step S1.4: the training to obtain the neural network reactive power optimization module mainly comprises the following steps:
preferably, the training is performed using a BP neural network.
Preferably, the reactive power optimization time is selected as each hour time.
P from step S1.2N、QNExtracting a real-time measurable load value P at an integral point moment iiv＝[Pi1,Pi2,…PiNv]And Qiv＝[Qi1,Qi2,…QiNv]Then x is ═ Pv,Qv]As an input part of a neural network training set, where PiNvAnd QiNvAnd the data of the active load and the reactive load of each node of the power distribution network at the integral point time i in history are shown.
Extracting from S1.3iNvAnd QiNvCorresponding y ═ SiAs part of the output of the neural network training set. Thus, a training set D { (x) is formed1,y1),(x2,y2),…(xm,ym)}xi∈R2Nv,yi∈RlWhile x is ═ x1,x2…，xm]y＝[y1,y2…,ym]
In this embodiment, Nv is set to 20, m is set to 1000, l is set to 8, that is, there are 1000 samples, the number of neural network input neurons is 40, and the number of neural network output neurons is 8, which respectively represent the OLTC position, the reactive power output of 4 photovoltaics, the input number of 2 capacitor banks, and the reactive power output of 1 SVC. And the neural network hidden layer is set to layer 1.
And then training the neural network by using an error inverse propagation algorithm until the training reaches a convergence condition to obtain a neural network reactive power optimization module which can be used for carrying out reactive power optimization on the i moment, wherein the specific training process is as follows:
s1.4.1: setting learning rate and initializing each neuron weight w and a threshold theta, wherein an activation function f is a sigmoid function. Wherein, the weight between the ith neuron of the input layer and the h-th neuron of the hidden layer is vihThe weight between the h-th neuron of the hidden layer and the j-th neuron of the output layer is whj. The threshold value of the jth neuron of the output layer is thetajThe threshold of the h-th neuron of the hidden layer is gammah. Implicit toThe h-th neuron of the layer receives the input sum of
Figure BDA0002546179830000071
The j-th neuron of the output layer receives the input sum of
Figure BDA0002546179830000072
bhThe output of the h-th neuron of the hidden layer.
S1.4.2 training sample (x)k,yk) Computing the output of the neural network at that time
Figure BDA0002546179830000073
And the mean square error, and the specific calculation formula is as follows:
Figure BDA0002546179830000074
Figure BDA0002546179830000075
s1.4.3 updating neural network related parameters v (including w)hj、j、vih、h) The formula used is: v + Δ v
The specific calculation formula is as follows:
Δwhj＝ηgjbh
Δθj＝-ηgj
Δvih＝ηehxi
Δγh＝-ηeh
Figure BDA0002546179830000076
Figure BDA0002546179830000077
s1.4.4, judging whether the convergence condition is reached, if not, returning to step S1.4.3, if so, indicating that the training is finished, and obtaining the reactive power optimization module suitable for the integral point time i.
S1.4.5 the reactive power optimization module suitable for the integral time i epsilon [1,2, … 24] is obtained by the method described above.
S1.2: when the actual reactive power optimization is carried out, the active and reactive loads [ P ] of the measurable nodes are collectedv,Qv]As an input of the neural network, an output is a corresponding reactive power optimization strategy S, and a method for matching a specific reactive power optimization module with the power distribution network is shown in fig. 3.
In order to verify the effectiveness of the method, a classic IEEE33 node is modified, photovoltaic power generation and capacitors and SVC are added, and a new power distribution network is formed, as shown in figure 2. Wherein, the nodes which can collect the load data in real time are represented by solid lines, and the nodes which can not collect the load data in real time are represented by dotted lines. The photovoltaic nodes which can be used for participating in reactive power optimization are connected by solid lines, and the photovoltaic nodes which do not participate in reactive power optimization are connected by dotted lines, so that 20 real-time-acquirable load data nodes and 13 non-real-time-acquirable load data nodes are known. The capacity of 8 photovoltaic cells is 100kW, the SVC capacity is 50kVar, the capacity of each capacitor is 25kVar, the gear of a transformer splitting head is +/-8 gears, and the voltage adjustable range is 0.9-1.1. The equipment available for reactive power optimization in the system includes four observable photovoltaics, capacitor banks and an SVC. The inverter is used for controlling the photovoltaic to emit or absorb reactive power, so that the photovoltaic power factor is in the range of-0.9 to 0.9. Where the load and photovoltaic data are derived from some actual data.
After the training of the neural network is finished, 200 groups of data are taken as a test set of the neural network for testing, and errors between the output results of the neural network and the test samples shown in tables 1 and 2 can be obtained.
TABLE 1 relative error of neural network outputs
Figure BDA0002546179830000081
TABLE 2 absolute error of neural network output
Figure BDA0002546179830000082
Therefore, the reactive power optimization strategy output by the neural network is not greatly different from the reactive power optimization strategy obtained by particle swarm calculation, and the effect of the reactive power optimization strategy output by the neural network can be effectively ensured.
As can be seen from fig. 4 and 5, compared with the conventional method, the optimization effect of the method is very small and is obviously better than that of the method without optimization. However, it should be noted that the premise of the traditional particle swarm method is that the load of each node of the power distribution network can be measured in real time, so the method is an effective method under the background of the loss of real-time measurement information of the power distribution network.
Has the advantages that: the method provided by the invention can realize the reactive power optimization of the power distribution network under the condition that the load data of partial nodes of the power distribution network cannot be measured in real time. The method is convenient to implement, high in accuracy and high in operation speed, and can be effectively applied to actual engineering.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, but any modifications or equivalent variations made according to the technical spirit of the present invention are within the scope of the present invention as claimed.

Claims (4)

1. The data-driven reactive power optimization method considering the loss of the real-time measurement information of the power distribution network is characterized by comprising the following steps of: the method comprises two stages of pre-optimization neural network training set construction, neural network training and application of the neural network during optimization, and comprises the following steps:
1) before actual optimization, collecting historical load data of a distribution network marketing system;
in the step 1), the number of nodes of the power distribution network which can be measured in real time is NvThe number of nodes which cannot be measured in real time is NivAnd the total number of the nodes of the power distribution network is N, so that the nodes can be measured in real time by the available power distribution network marketing systemThe historical active load data is Pv＝[P1,P2,…PNv]Historical reactive load data is Qv＝[Q1,Q2,…QNv]The historical active load data of the nodes which cannot be measured in real time by the distribution network marketing system is Piv＝[P1,P2,…PNiv]Historical reactive load data is Qiv＝[Q1,Q2,…QNiv]Then the total historical active load data PN＝Pv∪PivHistorical reactive load data QN＝Qv∪Qiv；
2) Then, carrying out data preprocessing by using the data acquired in the step 1) to obtain a historical load database;
the marketing system historical load data [ P ] acquired in the step 1) is utilized in the step 2)N,QN]Preprocessing is carried out, and the data quality is improved;
the data preprocessing in the step 2) adopts the following method:
if PNTwo data P inml,nAnd Pmr,nData P with k deletions in betweenml+i,nAnd i is 1,2, 3 … k, the following formula is adopted:
Pml+i,n＝Pml,n+(Pmr,n-Pml,n)/(k+1)×i，i＝1，2，3...k
if QNTwo data Q in (1)ml,nAnd Qmr,nData Q with k misses in betweenml+i,nAnd i is 1,2, 3 … k, the following formula is adopted:
Qml+i,n＝Qml,n+(Qmr,n-Qml,n)/(k+1)×i，i＝1，2，3...k；
3) then, generating a corresponding time reactive power optimization strategy library by using an optimization algorithm aiming at historical load data;
4) then, taking historical load data of nodes capable of being measured in real time as a neural network input training set, and taking a reactive power optimization strategy at a corresponding moment as a neural network output training set to train the neural network;
5) when an optimization cycle begins, load information of nodes capable of being measured in real time is used as input of the neural network trained in the step 4), and output of the neural network is a reactive power optimization strategy at the current moment.
2. The data-driven reactive power optimization method considering the lack of the real-time measurement information of the power distribution network according to claim 1, wherein the method comprises the following steps: in the step 3), by using the historical load data obtained in the step 2) and the topology of the power distribution network, obtaining a reactive power optimization strategy corresponding to the historical moment through a particle swarm algorithm;
the fitness function of the particle swarm is:
Figure FDA0003198622800000011
the constraint conditions are as follows:
Figure FDA0003198622800000021
Figure FDA0003198622800000022
Figure FDA0003198622800000023
Figure FDA0003198622800000024
wherein P isijRepresents the power flowing from node i to node j;
Figure FDA00031986228000000213
andU irespectively representing the upper limit and the lower limit allowed by the voltage of the node i;
Figure FDA00031986228000000212
andQ DGirespectively representing the upper limit and the lower limit of the DG reactive power output connected to the node i;
Figure FDA00031986228000000210
andQ SVCirepresenting the upper and lower output limits of the static var compensator at the node i;
Figure FDA00031986228000000211
andC irespectively representing the upper limit and the lower limit of the number of the capacitor groups input at the node i; the particle group velocity and position update formula is:
Vid＝ωVid+C1random(0,1)(Pid-Xid)+C2random(0,1)(Pgd-Xid)
wherein C1 and C2 are speed updating parameters of the particles;
Xid＝Xid+Vid
wherein VidDenotes the particle velocity, XidIndicating the position of the particle, PgdRepresenting the global optimum fitness, PidThe fitness of the particle is the fitness of the particle.
3. The data-driven reactive power optimization method considering the lack of the real-time measurement information of the power distribution network as recited in claim 2, wherein: the reactive power optimization strategy comprises a gear T of the on-load tap changing transformertapThe number of capacitor groups NcbPhotovoltaic reactive power output value QdgThese reactive power optimization strategies together form a strategy set S.
4. The data-driven reactive power optimization method considering the lack of the real-time measurement information of the power distribution network according to claim 3, wherein the method comprises the following steps: the input part of the neural network training set in the step 4) is the P obtained in the step 2)v,Qv]The output part is the strategy set S obtained in the step 3), and a reactive power optimization model for measuring the power distribution network with information missing in real time is obtained finally through training;
and (3) training by using a BP neural network, wherein the training process of the BP neural network comprises the following steps:
4.1) setting learning rate and initializing weight w and threshold theta of each neuron, wherein the activation function f is a sigmoid function, and the weight between the ith neuron of the input layer and the h neuron of the hidden layer is vihThe weight between the h-th neuron of the hidden layer and the j-th neuron of the output layer is whjThe threshold value of the jth neuron of the output layer is thetajThe threshold of the h-th neuron of the hidden layer is gammahThe h-th neuron in the hidden layer receives the input sum of
Figure FDA0003198622800000026
The j-th neuron of the output layer receives the input sum of
Figure FDA0003198622800000027
bhThe output of the h neuron of the hidden layer;
4.2) when the training sample is (x)k,yk) Computing the output of the neural network at that time
Figure FDA0003198622800000028
And the mean square error, and the specific calculation formula is as follows:
Figure FDA0003198622800000029
Figure FDA0003198622800000031
4.3) updating neural network-related parameters v, including whj、θj、vih、γhThe formula used is: v + Δ v
4.4) the specific calculation formula is as follows:
Δwhj＝ηgjbh
Δθj＝-ηgj
Δvih＝ηehxi
Δγh＝-ηeh
Figure FDA0003198622800000032
Figure FDA0003198622800000033
4.5) judging whether a convergence condition is reached, if not, returning to the step 3), and if so, indicating that the training is finished.
CN202010560697.3A 2020-06-18 2020-06-18 Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network Expired - Fee Related CN111756049B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010560697.3A CN111756049B (en) 2020-06-18 2020-06-18 Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010560697.3A CN111756049B (en) 2020-06-18 2020-06-18 Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network

Publications (2)

Publication Number Publication Date
CN111756049A CN111756049A (en) 2020-10-09
CN111756049B true CN111756049B (en) 2021-09-21

Family

ID=72675462

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010560697.3A Expired - Fee Related CN111756049B (en) 2020-06-18 2020-06-18 Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network

Country Status (1)

Country Link
CN (1) CN111756049B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113722303B (en) * 2021-08-16 2024-12-03 盛隆电气集团有限公司 A visualization-based electric energy equipment and information operation and maintenance system
CN113972667A (en) * 2021-11-29 2022-01-25 昆明理工大学 A Reactive Power Optimization Method for Distribution Network Based on One-dimensional Convolutional Neural Network
CN117526443B (en) * 2023-11-07 2024-04-26 北京清电科技有限公司 Power system-based power distribution network optimization regulation and control method and system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105322519A (en) * 2015-11-02 2016-02-10 湖南大学 Big data fusion analysis and running state monitoring method for intelligent power distribution network
CN106058914A (en) * 2016-05-27 2016-10-26 国电南瑞科技股份有限公司 Voltage optimization method of distribution network generation predication technology based on Elman algorithm
CN110535146A (en) * 2019-08-27 2019-12-03 哈尔滨工业大学 The Method for Reactive Power Optimization in Power of Policy-Gradient Reinforcement Learning is determined based on depth
CN111130122A (en) * 2020-01-14 2020-05-08 浙江大学 Online monitoring method for reactive power control capability of wind power plant

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108448585B (en) * 2018-03-29 2019-09-27 清华大学 A kind of electric network swim equation solution method of linearization based on data-driven
CN109144987A (en) * 2018-08-03 2019-01-04 天津相和电气科技有限公司 Electric system based on deep learning measures missing values method for reconstructing and its application
US20200119556A1 (en) * 2018-10-11 2020-04-16 Di Shi Autonomous Voltage Control for Power System Using Deep Reinforcement Learning Considering N-1 Contingency
CN110222883A (en) * 2019-05-22 2019-09-10 镇江市高等专科学校 Load Prediction In Power Systems method based on wind Drive Optimization BP neural network
CN111177278A (en) * 2019-12-19 2020-05-19 绍兴大明电力设计院有限公司 Grid user short-term load prediction real-time processing tool

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105322519A (en) * 2015-11-02 2016-02-10 湖南大学 Big data fusion analysis and running state monitoring method for intelligent power distribution network
CN106058914A (en) * 2016-05-27 2016-10-26 国电南瑞科技股份有限公司 Voltage optimization method of distribution network generation predication technology based on Elman algorithm
CN110535146A (en) * 2019-08-27 2019-12-03 哈尔滨工业大学 The Method for Reactive Power Optimization in Power of Policy-Gradient Reinforcement Learning is determined based on depth
CN111130122A (en) * 2020-01-14 2020-05-08 浙江大学 Online monitoring method for reactive power control capability of wind power plant

Also Published As

Publication number Publication date
CN111756049A (en) 2020-10-09

Similar Documents

Publication Publication Date Title
CN110535146B (en) Electric power system reactive power optimization method based on depth determination strategy gradient reinforcement learning
Li et al. Probabilistic charging power forecast of EVCS: Reinforcement learning assisted deep learning approach
CN111756049B (en) Data-driven reactive power optimization method considering loss of real-time measurement information of power distribution network
CN113872213B (en) Autonomous optimization control method and device for power distribution network voltage
CN105048479B (en) A kind of idle packet adjusting method of photovoltaic plant
CN113489015A (en) Power distribution network multi-time scale reactive voltage control method based on reinforcement learning
CN113158573A (en) Small hydropower station regional distribution network optimal power flow obtaining method based on deep learning
CN115588998A (en) A Method for Optimizing Voltage and Reactive Power of Distribution Network Based on Graph Reinforcement Learning
CN105938578A (en) Large-scale photovoltaic power station equivalent modeling method based on clustering analysis
CN104182816A (en) Method for evaluating power quality comprehensively based on the Vague sets and the improved technique for order preference by similarity to ideal solution and application thereof
CN113300379B (en) A method and system for reactive power and voltage control of power system based on deep learning
CN114430165A (en) Micro-grid group intelligent coordination control method and device based on depth model prediction
CN111079972A (en) Method, device and medium for planning reliability of active power distribution network
CN110212551A (en) Microgrid reactive power autocontrol method based on convolutional neural networks
CN113328435B (en) A joint control method of active and reactive power in active distribution network based on reinforcement learning
CN109950935A (en) Probabilistic power flow method for AC-DC hybrid microgrid operating in island
CN107834593A (en) A kind of droop control type isolated island micro-capacitance sensor static voltage stability probability evaluation method of failure
CN117937599A (en) Multi-agent reinforcement learning distribution network optimization method for distributed photovoltaic consumption
CN111461297B (en) Optimal Prediction Algorithm of Solar Irradiance Based on MPC and ELM Neural Network
Ni et al. A review of line loss analysis of the low-voltage distribution system
CN112418496A (en) Power distribution station energy storage configuration method based on deep learning
He Solar maximum power point tracking algorithm based on wavelet neural network
CN119051151A (en) Load and power supply cooperative control method for multi-agent reinforcement learning
CN113837654A (en) Multi-target-oriented intelligent power grid layered scheduling method
CN110675276B (en) Method and system for inversion droop control of direct current power transmission system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20210921