CN116565856A - Power distribution network state estimation method considering unknown topology change - Google Patents

Power distribution network state estimation method considering unknown topology change Download PDF

Info

Publication number
CN116565856A
CN116565856A CN202310627163.1A CN202310627163A CN116565856A CN 116565856 A CN116565856 A CN 116565856A CN 202310627163 A CN202310627163 A CN 202310627163A CN 116565856 A CN116565856 A CN 116565856A
Authority
CN
China
Prior art keywords
distribution network
power distribution
node
encoder
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.)
Pending
Application number
CN202310627163.1A
Other languages
Chinese (zh)
Inventor
曹迪
胡维昊
孙新武
黄越辉
胡家祥
廖启术
李思辰
井实
礼晓飞
韩培东
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.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
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 University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202310627163.1A priority Critical patent/CN116565856A/en
Publication of CN116565856A publication Critical patent/CN116565856A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • 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
    • 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
    • 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

Abstract

The invention discloses a power distribution network state estimation method considering unknown topological changes, which comprises the steps of firstly, utilizing a neural network of a self-encoder structure to perform feature encoding on observed data in a power distribution network, and establishing the encoding features of the observed data and state quantity in the power distribution network as a multi-task Gaussian process model; training the multi-task probability Gaussian process through coding features of a large amount of power distribution network observation data under a known topological structure and a small amount of power distribution network observation data under unknown topological structure change, and correcting Gaussian process parameters by comparing the relation between the state quantity of each node in the power distribution network under the unknown topological structure and the state quantity predicted by the Gaussian process; and finally, estimating the state of the power distribution network in real time by using the trained multi-task Gaussian process model.

Description

Power distribution network state estimation method considering unknown topology change
Technical Field
The invention belongs to the technical field of power distribution network state estimation, and particularly relates to a power distribution network state estimation method considering unknown topology change.
Background
In a traditional power distribution network, current generally flows from a power supply end to a load end, and the unidirectional radial power supply structure is weak and has low operation reliability. With the access of distributed energy sources such as photovoltaic, wind power and the like, the problems that node voltage in a power distribution network is out of limit, short-circuit current is increased, power supply reliability is lowered and the like are obvious. In order to solve the problems, the state quantity of each node of the power distribution network needs to be monitored in real time and processed in real time. The topology structure in the distribution network is changeable, the branches are numerous, and it is difficult to set enough real-time measuring devices to obtain the observability of the whole network. Therefore, reasonable modeling must be performed on numerous types of loads and power output of power sources in the power distribution network, so that state quantities of all nodes of the whole network are accurately estimated.
Accurate power distribution network state estimation has important guiding significance for helping workers to control a power grid, such as control of voltage nodes and reactive power nodes in the power grid, and recovery and reconstruction of the power grid. Conventional power distribution network state estimation mainly has two modes: optimization-based methods and learning-based methods. The topology information of the power grid needs to be accurately described based on an optimization mode, and the requirement on the measured value is high. Meanwhile, the measured value with larger error has larger influence on the state precision of the power distribution network. With the development of advanced measurement devices, information communication technologies and big data analysis technologies, historical observation data of the power distribution network are used for analyzing the distribution characteristics of data of all nodes in the power distribution network, and state variables of the power distribution network can be accurately estimated in real time through the description of the operation rule of the whole power distribution network. This is the core of the learning-based approach. The method based on learning is adopted in the filed patent of the invention, namely a state estimation method considering the topological structure change of the power distribution network.
The existing state estimation modes based on learning are all state estimation based on models, and the optimization of the models depends on a fixed power grid topological structure and a large amount of measurement information and state information. In an actual power distribution network system, the topology structure of the power distribution network is continuously changed, and the rapidly-changed topology structure enables the mapping relation between the state quantity of the node and the measurement information to be continuously changed. The trained model has difficulty producing good results for estimation of state quantities under new topologies. At the same time, the rapid change in topology and the limitation of the sampling frequency of the metrology device result in less metrology device data being available under the same topology. Less training data can lead to poor generalization of the model, and cannot be used for accurate state estimation of other states with the same topological structure. In the learning-based state estimation method, model-free state estimation is more suitable for real-time estimation of the state of the power distribution network than model-based state estimation.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a power distribution network state estimation method considering unknown topological changes, which models the state variable of a power distribution network by using a Gaussian process and then estimates the state of the power distribution network in real time according to the model.
In order to achieve the above object, the present invention provides a power distribution network state estimation method considering unknown topology changes, which is characterized by comprising the following steps:
(1) Randomly accessing the photovoltaic inverter into a power distribution network;
acquiring a topological structure of a power distribution network to be accessed, and randomly and dispersedly accessing photovoltaic inverters on nodes of the topological structure, wherein the total number of the accessed photovoltaic inverters is r;
(2) Obtaining observed quantity and state quantity of the power distribution network;
(2.1) obtaining observed quantity of the power distribution network;
(2.1.1) acquiring real-time measurement data;
randomly selecting n scattered branches on a topological structure, and recording the active power and the reactive power flowing at the moment t of each branch as real-time measurement data, wherein the active power flowing at the moment t of the kth branch is recorded as P k (t) reactive power is denoted as Q k (T), k=1, 2, …, n, t=1, 2, …, T is the number of sampling instants;
(2.1.2) obtaining pseudo-measurement data;
randomly selecting m non-repeated nodes on a topological structure, and recording the active power and reactive power of each node at the time t as pseudo-measurement data, wherein if the node i is connected with the photovoltaic inverter, the active power of the node i at the time t is recordedAnd reactive power->The method comprises the following steps of:
if the node i is not connected into the photovoltaic inverter, the active power of the node i at the time tAnd reactive power->The method comprises the following steps of:
wherein i=1, 2, …, m, P i,PV (t) represents the active power injected into node i by the photovoltaic inverter at time t, P i,load (t) represents the load active power of node i at time t, Q i,PV (t) represents the reactive power injected into node i by the photovoltaic inverter at time t, Q i,load (t) represents the load reactive power of node i at time t;
(2.1.3) combining the t-moment measurement data and the pseudo measurement data into an observed quantity X of the power distribution network 1 (t)；
(2.2) acquiring state quantity of the power distribution network;
traversing each node of the topological structure, and obtaining the voltage amplitude V of each node at the time t j (t) and a voltage phase angle θ j (t), j=1, 2, …, N representing the number of nodes in the distribution network topology;
the voltage amplitude and the voltage phase angle of each node at the moment t are combined into a state quantity Y of the power distribution network 1 (t)；
Y 1 (t)＝{V 1 (t),…,V j (t),…,V N (t)；θ 1 (t),…,θ j (t),…,θ N (t)}
(3) Constructing a training data set;
changing the topological structure of the power distribution network, repeating the step (2), obtaining the observed quantity and the state quantity of the power distribution network under the topological structure, and recording the observed quantity as X 2 (t) the state quantity is Y 2 (t)；
Combining observed quantity and state quantity of the power distribution network at different moments acquired under two topological structures into a training data set (X, Y);
X＝{X 1 (1),…,X 1 (t),…,X 1 (T),X 2 (1),…,X 2 (t),…,X 2 (T)}
Y＝{Y 1 (1),…,Y 1 (t),…,Y 1 (T),Y 2 (1),…,Y 2 (t),…,Y 2 (T)}
(4) Constructing a self-encoder network and training network weights;
(4.1) the self-encoder network comprises an encoder and a decoder, wherein the encoder and the decoder are respectively composed of three layers of neural network modules, each neural network module further comprises a full-connection layer and an activation function layer, and the activation function layer adopts a ReLU activation function;
the input size and the output size of the three-layer neural network module of the encoder are (M, 64, (64, 32), (32,16) respectively, and the input size and the output size of the three-layer neural network module of the decoder are (16, 32), (32, 64), (64, M) respectively, wherein M is the length of an input sample;
(4.2) setting initial values of weight parameters theta in the self-encoder network, wherein the initial values theta conform to Gaussian distribution with a mean value of 0 and a variance of 1; setting the maximum iteration number L, and initializing the current iteration number L=1; setting a learning rate lr;
(4.3) in the first iteration, extracting M samples from the observed quantity set X as training data of a single batch to be input into an encoder to obtain coding features Z, and inputting the coding features Z into a decoder to obtain reconstructed samples
Calculating the loss value after the iteration of the roundThen judging the current iteration times l=L or the loss value loss converges, if yes, stopping iterative training to obtain a trained self-encoder network; otherwise, the weight parameters are updated by using a gradient back propagation mechanism, howeverThen training the next round;
where θ represents the weight parameter before update,representing the updated weight parameters;
(5) The trained self-encoder is utilized to reduce the dimension of the observed data X;
the observed quantity X in the training data set (X, Y) is sequentially input to a trained self-encoder network, and the encoding characteristic Z= { Z is output through an encoder 1 (1),…,Z 1 (t),…,Z 1 (T),Z 2 (1),…,Z 2 (t),…,Z 2 (T) and as input to a multitasking gaussian process regression model;
(6) Constructing a multi-task Gaussian process regression model;
Y 1 (t)＝f(Z 1 (t))+ε 1 ,ε 1 ～N(0,σ 1 2 I)
Y 2 (t)＝f(Z 2 (t))+ε 2 ,ε 2 ～N(0,σ 2 2 I)
f(Z)＝GP(0,K)
wherein ε 1 And epsilon 2 Represents observed quantity noise of two topological structures of power distribution network, and obeys 0 and standard as average valueThe differences are sigma respectively 1 Sum sigma 2 Is a gaussian distribution of (c); f () is a zero-mean gaussian process; GP () represents a multitasking gaussian process with mean value 0; k is covariance matrix; k (a, b) is a gaussian kernel function, a, b being a variable; b is a semi-positive definite matrix,represents Kronecker product; sigma (sigma) f Super parameters which are Gaussian kernel functions; the superscript T denotes a transpose;
(7) Constructing an objective function of the super parameter in the multi-task Gaussian regression process model;
setting the hyper-parameters in the multi-task Gaussian regression process model to be Θ= { sigma f ,σ 1 ,σ 2 B, constructing optimal super parametersIs the objective function of (2)
(8) Substituting the code feature Z into a multi-task Gaussian process regression model, and then searching for optimal parameters by maximizing edge log likelihoodThe state quantity predicted by the multi-task Gaussian process regression model is consistent with the state quantity given in the training data set (X, Y), and finally the optimal parameter +.>Obtaining an optimal multi-task Gaussian regression process model;
(9) Estimating the state of the power distribution network in real time;
acquiring observed quantity X (T+1) of the power distribution network at the T+1 time according to the step (2.1), inputting the observed quantity X (T+1) into a trained self-encoder network, outputting coding characteristics Z (T+1) through an encoder, and inputting the coding characteristics Z (T+1) into an optimal multi-task Gaussian regression process model, so that state quantity Y (T+1) at the T+1 time is predicted.
The invention aims at realizing the following steps:
the invention relates to a power distribution network state estimation method considering unknown topological changes, which comprises the steps of firstly, utilizing a neural network of a self-encoder structure to perform feature encoding on observed data in a power distribution network, and establishing the encoding feature of the observed data and state quantity in the power distribution network as a multi-task Gaussian process model; training the multi-task probability Gaussian process through coding features of a large amount of power distribution network observation data under a known topological structure and a small amount of power distribution network observation data under unknown topological structure change, and correcting Gaussian process parameters by comparing the relation between the state quantity of each node in the power distribution network under the unknown topological structure and the state quantity predicted by the Gaussian process; and finally, estimating the state of the power distribution network in real time by using the trained multi-task Gaussian process model.
Meanwhile, the power distribution network state estimation method considering unknown topological changes has the following beneficial effects:
(1) The invention utilizes the machine learning algorithm to construct the mapping relation between the measurement data and the state quantity of the power distribution network, realizes the state estimation without a physical model, and has no requirement on the topological structure information of the power distribution network.
(2) The invention can adjust model parameters by using a small amount of training data, thereby accurately estimating the state quantity of the power distribution network after the topology structure transformation by using a large amount of measurement data under the original topology structure and a small amount of measurement data under the new topology structure.
(3) On the basis of the traditional Gaussian process, the correlation matrix is built to adaptively analyze the correlation between a large amount of distribution network measurement data under a known topological structure and a small amount of distribution network measurement data with unknown topological structure change, the state estimation precision of the distribution network is ensured, the result of state estimation is corrected by using a small amount of distribution network measurement data with unknown topological structure change, independent model establishment is not needed for the distribution network under each topological state, and when the distribution network is faced with a frequently-changed topological structure, the method has more obvious advantages.
Drawings
FIG. 1 is a flow chart of a method for estimating the state of a power distribution network taking account of unknown topology changes;
FIG. 2 is a schematic diagram of an IEEE33 node topology;
FIG. 3 is a graph of the result of different algorithms on the voltage amplitude state estimation of the nodes of the power distribution network;
FIG. 4 is a graph of the results of state estimation of node voltage and phase angle for various nodes of a power distribution network by various algorithms;
Detailed Description
The following description of the embodiments of the invention is presented in conjunction with the accompanying drawings to provide a better understanding of the invention to those skilled in the art. It is to be expressly noted that in the description below, detailed descriptions of known functions and designs are omitted here as perhaps obscuring the present invention.
Examples
Fig. 1 is a flowchart of a method for estimating a state of a power distribution network in consideration of unknown topology changes.
In this embodiment, as shown in fig. 2, an experiment is performed with IEEE33 nodes, and the power distribution network under the topology α connects 25 nodes and 29 nodes, disconnects 28 nodes and 29 nodes. The power distribution network under the topological structure beta is connected with 18 nodes and 33 nodes on the basis of the topological structure alpha, and 17 nodes and 18 nodes are disconnected. A total of 3 photovoltaic inverters (PV) are connected to the nodes 6,13,31. The photovoltaic power generation data adopts real and effective historical record data. The nodes 1-4,6-10 are nodes for acquiring real-time measurement data, and the nodes 2-33 are nodes for acquiring pseudo measurement data.
The following describes a power distribution network state estimation method considering unknown topology changes in detail with reference to fig. 1, which includes the following steps:
s1, randomly accessing a photovoltaic inverter into a power distribution network;
acquiring a topological structure of a power distribution network to be accessed, and randomly and dispersedly accessing photovoltaic inverters on nodes of the topological structure, wherein the total number of the accessed photovoltaic inverters is r=3;
s2, obtaining observed quantity and state quantity of the power distribution network;
s2.1, obtaining observed quantity of the power distribution network;
s2.1.1, acquiring real-time measurement data;
randomly selecting scattered n=7 branches on a topological structure, and recording active power and reactive power flowing at t time of each branch as real-time measurement data, wherein the active power flowing at t time of the kth branch is recorded as P k (t) reactive power is denoted as Q k (t)，k＝1,2,…,7，t＝1,2,…,500；
S2.1.2, obtaining pseudo measurement data;
randomly selecting non-repeated m=32 nodes on a topological structure, and recording the active power and reactive power of each node at the time t as pseudo-measurement data, wherein if the node i is connected with the photovoltaic inverter, the active power of the node i at the time t is recordedAnd reactive power->The method comprises the following steps of:
if the node i is not connected into the photovoltaic inverter, the active power of the node i at the time tAnd reactive power->The method comprises the following steps of:
wherein i=1, 2, …, m, P i,PV (t) represents the active power injected into node i by the photovoltaic inverter at time t, P i,load (t) represents the load active power of node i at time t, Q i,PV (t) represents the reactive power injected into node i by the photovoltaic inverter at time t, Q i,load (t) represents the load reactive power of node i at time t;
s2.1.3 observed quantity X of power distribution network formed by t moment measurement data and pseudo measurement data 1 (t)；
S2.2, acquiring state quantity of the power distribution network;
traversing each node of the topological structure, and obtaining the voltage amplitude V of each node at the time t j (t) and a voltage phase angle θ j (t)，j＝1,2,…,33；
The voltage amplitude and the voltage phase angle of each node at the moment t are combined into a state quantity Y of the power distribution network 1 (t)；
Y 1 (t)＝{V 1 (t),…,V j (t),…,V 33 (t)；θ 1 (t),…,θ j (t),…,θ 33 (t)}
S3, constructing a training data set;
changing the topological structure of the power distribution network, repeating the step S2, obtaining the observed quantity and the state quantity of the power distribution network under the topological structure, and recording the observed quantity as X 2 (t) the state quantity is Y 2 (t), wherein t=1, 2, …,50;
combining observed quantities and state quantities of different moments acquired from the power distribution network under two topological structures into a training data set (X, Y);
X＝{X 1 (1),…,X 1 (t),…,X 1 (500),X 2 (1),…,X 2 (t'),…,X 2 (50)}
Y＝{Y 1 (1),…,Y 1 (t),…,Y 1 (500),Y 2 (1),…,Y 2 (t'),…,Y 2 (50)}
s4, constructing a self-encoder network and training network weights;
s4.1, the self-encoder network comprises an encoder and a decoder, wherein the encoder and the decoder are respectively composed of three layers of neural network modules, each neural network module further comprises a full-connection layer and an activation function layer, and the activation function layer adopts a ReLU activation function;
the input size and the output size of the three-layer neural network module of the encoder are (78,6,4 (64, 32), (32,16) respectively, and the input size and the output size of the three-layer neural network module of the decoder are (16, 32), (32, 64), (64, 78) respectively, wherein 78 is the length of a single sample in the observed quantity;
s4.2, setting initial values of weight parameters theta in a self-encoder network, wherein the initial values theta conform to Gaussian distribution with mean value of 0 and variance of 1; setting the maximum iteration number as 200, and initializing the current iteration number l=1; setting a learning rate lr;
s4.3, in the first iteration, extracting 64 samples from X in the observed quantity set as training data of a single batch to be input into an encoder to obtain coding features Z, and inputting the coding features Z into a decoder to obtain reconstructed samples
Calculating the loss value after the iteration of the roundThen judging the current iteration times l=L or the loss value loss converges, if yes, stopping iterative training to obtain a trained self-encoder network; otherwise, updating the weight parameters by using a gradient back propagation mechanism, and then training the next round;
where θ represents the weight parameter before update,representing the updated weight parameters;
s5, reducing the dimension of the observed data X by using the trained self-encoder;
the observed quantity X in the training data set (X, Y) is sequentially input into a trained self-encoder network, and the encoding characteristic Z= { Z is output through an encoder 1 (1),…,Z 1 (t),…,Z 1 (500),Z 2 (1),…,Z 2 (t),…,Z 2 (50) -and as input to a regression model of the multitasking gaussian process;
s6, constructing a multi-task Gaussian process regression model;
Y 1 (t)＝f(Z 1 (t))+ε 1 ,ε 1 ～N(0,σ 1 2 I)
Y 2 (t)＝f(Z 2 (t))+ε 2 ,ε 2 ～N(0,σ 2 2 I)
f(Z)＝GP(0,K)
wherein ε 1 And epsilon 2 Represents observed quantity noise under two topological structures of the power distribution network, and obeys the average value to be 0 and the standard deviation to be sigma respectively 1 Sum sigma 2 Is a gaussian distribution of (c); zero average of f ()A value Gaussian process; GP () represents a multitasking gaussian process with mean value 0; k is covariance matrix; k (a, b) is a gaussian kernel function, a, b being a variable; b is a semi-positive definite matrix,represents Kronecker product; sigma (sigma) f Super parameters which are Gaussian kernel functions; the superscript T denotes a transpose;
s7, constructing an objective function of the super parameter in the multi-task Gaussian regression process model;
setting the hyper-parameters in the multi-task Gaussian regression process model to be Θ= { sigma f ,σ 1 ,σ 2 B, constructing optimal super parametersIs the objective function of (2)
S8, training a multi-task Gaussian process regression model;
substituting the code feature Z into a multi-task Gaussian process regression model and then finding the optimal parameters by maximizing the edge log likelihoodThe state quantity predicted by the multi-task Gaussian process regression model is consistent with the state quantity given in the training data set (X, Y), and finally the optimal parameter +.>Obtaining an optimal multi-task Gaussian regression process model;
s9, estimating the state of the power distribution network in real time;
according to the step S2.1, acquiring the observed quantity X (T+1) of the power distribution network at the T+1 time under the second topological structure, inputting the observed quantity X (T+1) into a trained self-encoder network, outputting the coding characteristic Z (T+1) through an encoder, and inputting the coding characteristic Z (T+1) into an optimal multi-task Gaussian regression process model, so that the state quantity Y (T+1) at the T+1 time is predicted.
FIG. 3 is a graph of the result of different algorithms on the voltage amplitude state estimation of the nodes of the power distribution network;
in this embodiment, the node voltage amplitude of each node of the power distribution network at a certain moment is selected to perform state estimation, and the result of the state estimation is shown in fig. 3. The figure enumerates various methods, including the methods herein, WLS stands for least squares, BPN stands for back propagation neural network, SGP stands for regression from sparse gaussian process, MTGP stands for multi-tasking gaussian process regression. As can be seen in fig. 3, the three methods of WLS, SGP and BPN perform state estimation on the nodes of the power distribution network, and the effect of state estimation is poor in the place where the topology structure is significantly changed. At nodes 29,30,31,32, the voltage magnitude estimated by these three methods differs significantly from the exact value. For the amplitude prediction of the voltage, only the method can better predict the amplitude of the voltage after the topological structure is changed.
Fig. 4 is a graph of the state estimation results of different algorithms for the voltage intervals of the nodes of the power distribution network.
In this embodiment, as shown in FIG. 4, SGP-F represents a method of training a Gaussian model using a large number of historical data of the original topology. According to the invention, the prediction of the node voltage interval by SGP and SGP-F can be seen, the prediction range of the node voltage by the method basically comprises the real voltage condition of the node, the state estimation results of the SGP and SGP-F are far different from the real value at the position where the topological structure is changed, and the node voltage interval predicted by SGP does not comprise the real value at the node 29,30,31,32,33. Whereas at nodes 31,32,33 the SGP-F node voltage interval does not contain a true value. Therefore, the method for estimating the state of the node by using the historical data of the two topological structures to carry out Gaussian regression model modeling can reasonably estimate the state of the node after the topological structure is changed, but the difference of the measured data of the network caused by different topological structures is not emphasized, and the estimated state of the node is far from the true state. In contrast, the invention uses the multi-task Gaussian regression model, considers the relation between the measured data under different topological structures, and can well estimate the node voltage and the node phase angle at the position where the topological structure has great change under the condition that the network measured data acquisition under the new topological structure is less. The invention introduces the multi-task Gaussian kernel, not only considers the mapping relation between the measurement data and the node states under various topological structures, but also considers the relativity between various topologies, and the model is promoted to use a small amount of training data to accurately estimate the states of the nodes under other topological structures.
While the foregoing describes illustrative embodiments of the present invention to facilitate an understanding of the present invention by those skilled in the art, it should be understood that the present invention is not limited to the scope of the embodiments, but is to be construed as protected by the accompanying claims insofar as various changes are within the spirit and scope of the present invention as defined and defined by the appended claims.

Claims (1)

1. The power distribution network state estimation method considering unknown topological change is characterized by comprising the following steps of:
(1) Randomly accessing the photovoltaic inverter into a power distribution network;
acquiring a topological structure of a power distribution network to be accessed, and randomly and dispersedly accessing photovoltaic inverters on nodes of the topological structure, wherein the total number of the accessed photovoltaic inverters is r;
(2) Obtaining observed quantity and state quantity of the power distribution network;
(2.1) obtaining observed quantity of the power distribution network;
(2.1.1) acquiring real-time measurement data;
randomly selecting n scattered branches on a topological structure, and recording the active power and the reactive power flowing at the moment t of each branch as real-time measurement data, wherein the active power flowing at the moment t of the kth branch is recorded as P k (t) reactive power is denoted as Q k (T), k=1, 2, …, n, t=1, 2, …, T is the number of sampling instants;
(2.1.2) obtaining pseudo-measurement data;
randomly selecting m non-repeated nodes on a topological structure, and recording the active power and reactive power of each node at the time t as pseudo-measurement data, wherein if the node i is connected with the photovoltaic inverter, the active power of the node i at the time t is recordedAnd reactive power->The method comprises the following steps of:
if the node i is not connected into the photovoltaic inverter, the active power of the node i at the time tAnd reactive power->The method comprises the following steps of:
wherein i=1, 2, …, m, P i,PV (t) represents the active power injected into node i by the photovoltaic inverter at time t, P i,load (t) represents the load active power of node i at time tPower, Q i,PV (t) represents the reactive power injected into node i by the photovoltaic inverter at time t, Q i,load (t) represents the load reactive power of node i at time t;
(2.1.3) combining the t-moment measurement data and the pseudo measurement data into an observed quantity X of the power distribution network 1 (t)；
(2.2) acquiring state quantity of the power distribution network;
traversing each node of the topological structure, and obtaining the voltage amplitude V of each node at the time t j (t) and a voltage phase angle θ j (t), j=1, 2, …, N representing the number of nodes in the distribution network topology;
the voltage amplitude and the voltage phase angle of each node at the moment t are combined into a state quantity Y of the power distribution network 1 (t)；
Y 1 (t)＝{V 1 (t),…,V j (t),…,V N (t)；θ 1 (t),…,θ j (t),…,θ N (t)}
(3) Constructing a training data set;
changing the topological structure of the power distribution network, repeating the step (2), obtaining the observed quantity and the state quantity of the power distribution network under the topological structure, and recording the observed quantity as X 2 (t) the state quantity is Y 2 (t)；
Combining observed quantity and state quantity of the power distribution network at different moments acquired under two topological structures into a training data set (X, Y);
X＝{X 1 (1),…,X 1 (t),…,X 1 (T),X 2 (1),…,X 2 (t),…,X 2 (T)}
Y＝{Y 1 (1),…,Y 1 (t),…,Y 1 (T),Y 2 (1),…,Y 2 (t),…,Y 2 (T)}
(4) Constructing a self-encoder network and training network weights;
(4.1) the self-encoder network comprises an encoder and a decoder, wherein the encoder and the decoder are respectively composed of three layers of neural network modules, each neural network module further comprises a full-connection layer and an activation function layer, and the activation function layer adopts a ReLU activation function;
the input size and the output size of the three-layer neural network module of the encoder are (M, 64, (64, 32), (32,16) respectively, and the input size and the output size of the three-layer neural network module of the decoder are (16, 32), (32, 64), (64, M) respectively, wherein M is the length of an input sample;
(4.2) setting initial values of weight parameters theta in the self-encoder network, wherein the initial values theta conform to Gaussian distribution with a mean value of 0 and a variance of 1; setting the maximum iteration number L, and initializing the current iteration number L=1; setting a learning rate lr;
(4.3) in the first iteration, extracting M samples from the observed quantity set X as training data of a single batch to be input into an encoder to obtain coding features Z, and inputting the coding features Z into a decoder to obtain reconstructed samples
Calculating the loss value after the iteration of the roundThen judging the current iteration times l=L or the loss value loss converges, if yes, stopping iterative training to obtain a trained self-encoder network; otherwise, updating the weight parameters by using a gradient back propagation mechanism, and then training the next round;
where θ represents the weight parameter before update,representing the updated weight parameters;
(5) The trained self-encoder is utilized to reduce the dimension of the observed data X;
the observed quantity X in the training data set (X, Y) is sequentially input to a trained self-encoder network, and the encoding characteristic Z= { Z is output through an encoder 1 (1),…,Z 1 (t),…,Z 1 (T),Z 2 (1),…,Z 2 (t),…,Z 2 (T) and as input to a multitasking gaussian process regression model;
(6) Constructing a multi-task Gaussian process regression model;
Y 1 (t)＝f(Z 1 (t))+ε 1 ,ε 1 ～N(0,σ 1 2 I)
Y 2 (t)＝f(Z 2 (t))+ε 2 ,ε 2 ～N(0,σ 2 2 I)
f(Z)＝GP(0,K)
wherein ε 1 And epsilon 2 Represents observed quantity noise under two topological structures of the power distribution network, and obeys the average value to be 0 and the standard deviation to be sigma respectively 1 Sum sigma 2 Is a gaussian distribution of (c); f () is a zero-mean gaussian process; GP () represents a multitasking gaussian process with mean value 0; k is covariance matrix; k (a, b) is a gaussian kernel function, a, b being a variable; b is a semi-positive definite matrix,represents Kronecker product; sigma (sigma) f Super parameters which are Gaussian kernel functions; the superscript T denotes a transpose;
(7) Constructing an objective function of the super parameter in the multi-task Gaussian regression process model;
setting the hyper-parameters in the multi-task Gaussian regression process model to be Θ= { sigma f ,σ 1 ,σ 2 B, constructing optimal super parametersIs the objective function of (2)
(8) Substituting the code feature Z into a multi-task Gaussian process regression model, and then searching for optimal parameters by maximizing edge log likelihoodThe state quantity predicted by the multi-task Gaussian process regression model is consistent with the state quantity given in the training data set (X, Y), and finally the optimal parameter +.>Obtaining an optimal multi-task Gaussian regression process model;
(9) Estimating the state of the power distribution network in real time;
acquiring observed quantity X (T+1) of the power distribution network at the T+1 time according to the step (2.1), inputting the observed quantity X (T+1) into a trained self-encoder network, outputting coding characteristics Z (T+1) through an encoder, and inputting the coding characteristics Z (T+1) into an optimal multi-task Gaussian regression process model, so that state quantity Y (T+1) at the T+1 time is predicted.
CN202310627163.1A 2023-05-30 2023-05-30 Power distribution network state estimation method considering unknown topology change Pending CN116565856A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310627163.1A CN116565856A (en) 2023-05-30 2023-05-30 Power distribution network state estimation method considering unknown topology change

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310627163.1A CN116565856A (en) 2023-05-30 2023-05-30 Power distribution network state estimation method considering unknown topology change

Publications (1)

Publication Number Publication Date
CN116565856A true CN116565856A (en) 2023-08-08

Family

ID=87501754

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310627163.1A Pending CN116565856A (en) 2023-05-30 2023-05-30 Power distribution network state estimation method considering unknown topology change

Country Status (1)

Country Link
CN (1) CN116565856A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117081082A (en) * 2023-10-17 2023-11-17 国网上海市电力公司 Active power distribution network operation situation sensing method and system based on Gaussian process regression

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117081082A (en) * 2023-10-17 2023-11-17 国网上海市电力公司 Active power distribution network operation situation sensing method and system based on Gaussian process regression
CN117081082B (en) * 2023-10-17 2024-01-23 国网上海市电力公司 Active power distribution network operation situation sensing method and system based on Gaussian process regression

Similar Documents

Publication Publication Date Title
CN109214575A (en) A kind of super short-period wind power prediction technique based on small wavelength short-term memory network
CN109272156B (en) Ultra-short-term wind power probability prediction method
CN111860982A (en) Wind power plant short-term wind power prediction method based on VMD-FCM-GRU
CN111353652B (en) Wind power output short-term interval prediction method
CN104573876A (en) Wind power plant short-period wind speed prediction method based on time sequence long memory model
Ma et al. Short term load forecasting based on iForest-LSTM
WO2022021726A1 (en) Pmu-based power system state estimation performance evaluation method
CN109088407B (en) Power distribution network state estimation method based on deep belief network pseudo-measurement modeling
CN110378504B (en) Photovoltaic power generation climbing event probability prediction method based on high-dimensional Copula technology
CN113762625A (en) Power distribution network state evaluation method and system based on graph convolution network
CN116565856A (en) Power distribution network state estimation method considering unknown topology change
CN108520310A (en) Wind speed forecasting method based on G-L mixed noise characteristic v- support vector regressions
CN110674984A (en) Tri-Training-Lasso-BP network-based static voltage stability margin prediction method
CN113612237A (en) Method for positioning resonance-induced subsynchronous oscillation source in offshore wind farm
CN106372440B (en) A kind of adaptive robust state estimation method of the power distribution network of parallel computation and device
CN107204616B (en) Power system random state estimation method based on self-adaptive sparse pseudo-spectral method
Xu et al. Short-term wind speed prediction based on GRU
Chen et al. A hybrid DNN-KF model for real-time SOC estimation of lithium-ion batteries under different ambient temperatures
CN110555566B (en) B-spline quantile regression-based photoelectric probability density prediction method
CN107918920B (en) Output correlation analysis method for multiple photovoltaic power stations
CN116167465A (en) Solar irradiance prediction method based on multivariate time series ensemble learning
Yang et al. Investigating the predictability of photovoltaic power using approximate entropy
CN116298670A (en) Intelligent fault positioning method and system suitable for multi-branch distribution line
Shi et al. A fault location method for distribution system based on one-dimensional convolutional neural network
CN115146538A (en) Power system state estimation method based on message passing graph neural network

Legal Events

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