Power distribution network state estimation method considering bad data and topological transformation

Abstract

The invention discloses a power distribution network state estimation method considering bad data and topological transformation. The method comprises the following steps: S1, acquiring historical measurement initial data and real-time measurement initial data of a power distribution network to be estimated; the real-time measurement initial data comprises bad data with a preset proportion; preprocessing the historical measurement initial data and the real-time measurement initial data to obtain historical measurement data and real-time measurement data; s2, training the first model according to the historical measurement data to obtain a second model; the first model comprises a measurement bad data processing model based on a gating circulation unit; s3, obtaining a pseudo measurement value at the estimation moment according to the second model, and replacing bad data in the real-time measurement data with the pseudo measurement value to obtain first measurement data; s4, obtaining a power distribution network state estimation result according to the first measurement data, the real-time measurement data, the third model and the fourth model; the third model comprises a robust estimation model based on a neural network; the fourth model comprises a state estimation model based on linear estimation.

Description

Technical Field

[0001] The present invention relates to the field of active distribution network state estimation, and in particular to a distribution network state estimation method considering bad data and topology transformation. Background Art

[0002] As the scale of distributed power access increases, the distribution network is transformed into an active distribution network, which puts higher requirements on monitoring and control. State estimation estimates the state quantities such as system voltage and current by receiving and processing the measurement data of the active distribution network, and uses the obtained state quantities for active distribution network monitoring and control. Therefore, ensuring high-precision state estimation results is crucial for the stability and safe operation of the system.

[0003] The accuracy of state estimation can be improved by improving the accuracy of measurement data or improving the estimation model. In terms of improving measurement accuracy, methods based on linear interpolation and back propagation neural network (BPN) are widely used; in terms of improving estimation models, state estimation methods based on weighted least squares (WLS) and weighted least absolute value (WLAV) are widely used and have high estimation accuracy. However, the power fluctuations of distributed power sources and the complex changes in loads have brought challenges to the measurement equipment to track and accurately record the real-time data of the active distribution network, affecting the measurement accuracy. Bad measurement data caused by measurement equipment failure or communication noise will not only reduce the measurement accuracy, but also increase the difficulty of improving the estimation accuracy. In addition, when the topology of the active distribution network changes, the delay and loss of topology data may lead to incorrect topology parameters, which will lead to increased estimation errors, especially when using traditional estimation methods that rely on accurate topology parameters.

[0004] Therefore, a new technical solution is urgently needed to solve the technical problem of how to ensure the accuracy of active distribution network measurement and state estimation while considering the influence of bad data and topology transformation. Summary of the invention

[0005] The present invention provides a distribution network state estimation method considering bad data and topology transformation, so as to solve the technical problem of how to ensure the accuracy of active distribution network measurement and state estimation under the influence of bad data and topology transformation.

[0006] To achieve the above object, the present invention provides a distribution network state estimation method considering bad data and topology transformation, comprising the following steps:

[0007] S1. Obtain historical measurement initial data and real-time measurement initial data of the distribution network to be estimated; the real-time measurement initial data includes a preset proportion of bad data; pre-process the historical measurement initial data and the real-time measurement initial data to obtain historical measurement data and real-time measurement data.

[0008] S2. Train the first model according to historical measurement data to obtain the second model; the first model includes a measurement bad data processing model based on a gated recurrent unit.

[0009] S3. Obtain a pseudo measurement value at the estimation time according to the second model, and replace bad data in the real-time measurement data with the pseudo measurement value to obtain first measurement data.

[0010] S4. Obtain distribution network state estimation results according to the first measurement data, the real-time measurement data, the third model and the fourth model; the third model includes a robust estimation model based on a neural network; the fourth model includes a state estimation model based on linear estimation.

[0011] Preferably, preprocessing the historical measurement initial data and the real-time measurement initial data includes:

[0012] The bad data in the real-time measurement initial data is subjected to a first preprocessing; the first preprocessing includes adding a first preset percentage of the amplitude value of the real-time measurement initial data.

[0013] The normal data in the real-time measurement initial data and the historical measurement initial data are subjected to a second preprocessing; the second preprocessing includes adding a second preset percentage of the amplitude value of the real-time measurement initial data.

[0014] Get historical measurement data, including

[0015] Historical active power measurement z1 = [z 1，1 ,...z 1，t , ..., z 1，T-1 ]; z 1，t =[z 1,t，1 ,...z 1,t，m , ..., z 1,t,n ] P ;

[0016] Historical reactive power measurement z2 = [z 2，1 ,...z 2，t , ..., z 2，T-1 ]; z 2，t =[z 2,t，1 ,...z 2,t，m , ..., z 2,t，n ] P ;

[0017] Get real-time measurement data, including:

[0018] Real-time active power measurement z3 = [z 3，2 ,...z 3，t , ..., z 3,T ]; z 3，t =[z 3,t，1 ,...z 3,t，m , ..., z 3,t,n ] P ;

[0019] Real-time reactive power measurement z4=[z 4，2 ,...z 4，t , ..., z 4,T ]; z 4，t =[z 4,t，1 ,...z 4,t，m , ..., z 4,t，n ] P ;

[0020] Real-time voltage measurement z5 = [z 5，2 ,...z 5，t , ..., z 5,T ]; z 5，t =[z 5,t，1 ,...z 4,t,q , ..., z 4,t,Q ] P ;

[0021] Real-time current measurement 6，2 ,...z 6，t , ..., z 6,T ]; z 6，t =[z 6,t，1 ,...z 6,t,q , ..., z 6,t,Q ] P ;

[0022] Among them, the real-time active power measurement z3 and the real-time reactive power measurement z4 contain bad data; t = 1, 2, ..., T, T is the sequence number of historical measurement data and real-time measurement data, that is, the number of sections at the estimated time; m = 1, 2, ..., n, n is the number of distribution network nodes; q = 1, 2, ..., Q, Q is the number of installed synchronous phasor measurement devices; [] P Indicates transpose.

[0023] Preferably, S1 further comprises:

[0024] The true value data of the distribution network to be estimated is obtained through distribution network flow calculation. The true value data includes:

[0025] Active power measurement true value V1 = [V 1，2, ... V 1，t , ..., V 1，T ]; V 1，t =[V 1，2，1 , ... V 1，t，m , ..., V 1，T,n ] P .

[0026] Reactive power measurement true value V2 = [V 2，2 , ... V 2，t , ..., V 2，T ]; V 2，t =[V 2，2，1 , ... V 2，t，m , ..., V 2，T，n ] P .

[0027] The true value of the voltage amplitude of each node VV1=[VV 1，2 ,...VV 1，t , ..., V 1，T ]; VV 1，t =[VV 1，2，1 ,...VV 1，t，m , ..., VV 1，T,n ] P .

[0028] The true value of the voltage phase angle of each node VV2 = [VV 2，2 ,...VV 2，t , ..., VV 2，T ]; VV 2，t =[VV 2，2，1 ,...VV 2，t，m , ..., VV 2，T,n ] P .

[0029] Preferably, before S2, it includes:

[0030] Constructing a measurement bad data processing model based on a gated recurrent unit, that is, constructing a first model, including:

[0031] Construct a bad data processing model with an input layer, a GRU layer, a fully connected layer, and an output layer. The input of the bad data processing model is x t,h , the output is y t,h ;x t，h and t，h The relationships between them include:

[0032] x t,h =[x h，t，1 ,...x h,t，m , ..., x h,t，n ]P =z h , t

[0033] z h，t =[z h,t，1 ,...z h，t，m , ..., z h,t，n ] P

[0034] x t，h,m =z h,t,m

[0035] h t =f(x t,h )

[0036] s t =σ(W (s) x t，h +U (s) h t-1 )

[0037] r t =σ(W (r) x t，h +U (r) h t-1 )

[0038]

[0039] a1=σ(w1h t +b1)

[0040] y t,h =a1

[0041] y t,h =[y h,t，1 ,...y h,t，m , ..., y h,t，n ] P

[0042] Where, f() represents the mapping function; W (s) , U (s) , W (r) and U (r) is a trainable parameter; h t and a1 are the outputs of the GRU layer and the fully connected layer respectively; σ is the sigmoid activation function; tanh is the hyperbolic tangent function; represents matrix multiplication; w l is the weight of the fully connected layer; b l is the bias of the fully connected layer; when h = 1, z h，t Indicates historical active power measurement; when h = 2, z h，t Indicates historical reactive power measurement.

[0043] The loss function of the bad data processing model can be expressed as follows:

[0044]

[0045] Among them, y t,h =a1, when h = 1, y h,t，m and V h，t，m They represent active power and true value of active power respectively. When h=2, y h,t，m and V h，t，m Represent reactive power and reactive power true value respectively.

[0046] Preferably, S2 comprises:

[0047] Translate the measured bad data processing model training into the first optimization model:

[0048]

[0049] in, Represents the loss function, and min means minimization.

[0050] The measurement bad data processing model is iteratively trained through the first optimization model and historical measurement data to obtain a second model.

[0051] Preferably, S3 includes:

[0052] Generate a pseudo measurement value y at the estimation time t according to the second model h,t,m .

[0053] Replace the bad data in real-time active power measurement z3 and real-time reactive power measurement z4 with pseudo measurement value y h,t,m , get the real-time power measurement data of the distribution network, that is, get the first measurement data z′ h，t .

[0054] Preferably, before S4, it includes:

[0055] Construct a robust estimation model based on neural network, namely the third model, including:

[0056] Construct a robust estimation model based on a neural network, which consists of one input layer, five fully connected layers and two output layers. The input of the robust estimation model is the measurement data z t，h , the outputs are respectively the voltage amplitude pre-estimated state quantity X v , voltage phase angle pre-estimation state quantity X θ ; The relationship between the input and output of the robust estimation model includes:

[0057] z′ h,t =[z′ 3,t; z′ 4,t ]

[0058] z t,h =[z′ h,t ;z 5,t ;z 6,t ]

[0059] c1=σ(ω1z t，h +bis1)

[0060] c2=σ(ω2c1+bis2)

[0061] c3=σ(ω3c2+bis3)

[0062] c4=σ(ω4C3+bis4)

[0063] c5=tanh(ω5C3+bis5)

[0064] X v =c4

[0065] X θ =c5

[0066]

[0067] X v =[X 1，t，1 , ...X 1，t，m , ..., X 1，t，n ]

[0068] X θ =[X 2，t，1 , ...X 2,t，m , ..., X 2，t,n ]

[0069] X h,t =[X h，t，1 , ...X h，t，m , ..., X h，t，n ]

[0070] Among them, z′ 3，t and z′ 4，t They are respectively the active power measurement and reactive power measurement after bad data processing and replacement; c1, c2, c3, c4 and c5 are the outputs of the first to fifth fully connected layers respectively; ω l is the weight of the lth fully connected layer; bis l is the bias of the lth fully connected layer, l = 1, 2, 3, 4, 5; X h，t Estimate the output for the network.

[0071] The loss function of the robust estimation model can be expressed by the following relationship:

[0072]

[0073] When h = 1, X h,t，m and VV h，t，m represent the estimated value of voltage amplitude and the true value of voltage amplitude respectively; when h=2, X h，t，m and VV h，t，m They represent the estimated voltage phase angle and the true voltage phase angle respectively.

[0074] Preferably, S4 includes:

[0075] S41, transcribe the third model training into the second optimization model:

[0076] min L ω，bis (VV h,t , X h,t )

[0077]

[0078] in, Represents the loss function, min represents minimization;

[0079] The third model is obtained by combining the second optimization model and the first measurement data z′ h，t Perform iterative training to obtain the weights ω of each layer l And the bias bis l , and then the fifth model is obtained.

[0080] S42, at the estimation time t, the first measurement data z′ h，t , Real-time voltage measurement 5，t and real-time current measurement 6，t Input the fifth model, the fifth model outputs the first estimated value X h，t .

[0081] S43, based on the first estimated value X h，t , real-time voltage measurement z5 and real-time current measurement z6 obtain voltage state estimation values ​​through the fourth model, and the voltage state estimation values ​​include estimated values ​​of voltage amplitude and voltage phase angle.

[0082] Preferably, before S43, the following is included:

[0083] Constructing a state estimation model based on linear estimation, i.e., the fourth model, including:

[0084] The input Z of the state estimation model t，h Includes first estimate X h，t , Real-time voltage measurement 5，t and real-time current measurement 6，t; Input Z of the state estimation model t，h The estimated value of the output voltage state The relationships include:

[0085] Z t,h =[X h，t ;z 5,t ;z 6,t ]=H*xx h，t +ε

[0086] xx h,t =[H P R -1 H] -1 H P R -1 Z

[0087] xx h,t,m =[xx h,t，1 ,...xx h,t，m ,...,xx h,t，n ]

[0088] U=[xx 1,t，1 ,...xx 1,t，m ,...,xx 1，t，n ]

[0089] θ=[xx 2，t，1 ,...xx 2,t，m ,...,xx 2,t,n ]

[0090] xx h,t =[xx h,t，1 ,...xx h,t，m ,...,xx h,t，n ]

[0091] Where H is the conversion matrix; ε is the measurement error; R is the measurement error weight matrix; xx h，t is the voltage state estimation value. When h=1, it indicates the estimated value of the voltage amplitude. When h=2, it indicates the estimated value of the voltage phase angle.

[0092] Preferably, it also includes:

[0093] Calculate the first indicator, including:

[0094] Root mean square error RMSE mea，t And the mean absolute error MAE mea，t :

[0095]

[0096] The accuracy of the measurement data is evaluated according to the first indicator.

[0097] Calculate the second index, including:

[0098] Mean absolute error (MAE) t and the average estimated time T s :

[0099] VV h,t =[VV h,t，1 ,...VV h，t，m , ..., VV h，t，n ]

[0100]

[0101] Among them, Num represents the number of estimations, t i Represents the estimated time of the i-th time.

[0102] The state estimation accuracy is evaluated according to the second indicator.

[0103] The present invention has the following beneficial effects:

[0104] The distribution network state estimation method considering bad data and topological transformation of the present invention uses a bad data processing model based on a gated cyclic unit to process bad data. The model uses historical measurement data sequence adaptive learning to generate high-precision pseudo-measurement values ​​for replacing bad data in real-time measurement data. Compared with traditional solutions, this method improves the measurement data accuracy; the robust estimation model based on a neural network used in the present invention does not need to input topological parameter information, overcomes the adverse effects of incorrect topological parameters on the decline in state estimation accuracy, and ensures the robustness of state estimation under bad data and topological changes; the state estimation model based on linear estimation used in the present invention introduces an estimated value, increases measurement redundancy, and further improves the estimation accuracy. The present invention uses a robust state estimation model that considers the influence of bad data and topological transformation, effectively improves the state estimation accuracy and robustness under bad data and topological transformation, and can ensure power grid safety monitoring and operation control.

[0105] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0106] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0107] Figure 1 It is a flow chart of a preferred embodiment of the present invention.

[0108] Figure 24 is a schematic diagram of the pseudo-measurement root mean square error of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0109] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0110] See also Figure 1 In a preferred embodiment of the present invention, a method for estimating the state of a distribution network considering bad data and topology transformation is provided, comprising the following steps:

[0111] S1. Obtain historical measurement initial data and real-time measurement initial data of the distribution network to be estimated; the real-time measurement initial data includes a preset proportion of bad data; pre-process the historical measurement initial data and the real-time measurement initial data to obtain historical measurement data and real-time measurement data.

[0112] In a preferred embodiment of the present invention, the real-time measurement initial data contains 15% bad data.

[0113] In a preferred embodiment of the present invention, preprocessing the historical measurement initial data and the real-time measurement initial data includes:

[0114] The bad data in the real-time measurement initial data is subjected to a first preprocessing; the first preprocessing includes adding a first preset percentage of noise to the amplitude value of the real-time measurement initial data; in a preferred embodiment of the present invention, the first preset percentage can be selected as 15% or 20%.

[0115] The normal data in the real-time measurement initial data and the historical measurement initial data are subjected to a second preprocessing; the second preprocessing includes adding a second preset percentage of noise to the amplitude value of the real-time measurement initial data; in the preferred embodiment of the present invention, the second preset percentage is 1%.

[0116] Get historical measurement data, including

[0117] Historical active power measurement z1 = [z 1，1 , …z 1，t , ..., z 1，T-1 ]; z 1，t =[z 1,t，1 ,...z 1,t，m , ..., z 1,t,n ] P ;

[0118] Historical reactive power measurement z2 = [z 2，1 , …z 2，t , ..., z 2，T-1 ]; z 2，t =[z 2,t，1 ,...z 2,t，m , ..., z2,t,n ] P ;

[0119] Get real-time measurement data, including:

[0120] Real-time active power measurement z3 = [z 3，2 ,...z 3，t , ..., z 3,T ]; z 3，t =[z 3,t，1 ,...z 3,t，m , ..., z 3,t,n ] P ;

[0121] Real-time reactive power measurement z4=[z 4，2 ,...z 4，t , ..., z 4,T ]; z 4，t =[z 4,t，1 ,...z 4,t，m , ..., z 4,t,n ] P ;

[0122] Real-time voltage measurement z5 = [z 5，2 ,...z 5，t , ..., z 5,T ]; z 5，t =[z 5,t，1 ,...z 4,t,q , ..., z 4，t,Q ] P ;

[0123] Real-time current measurement 6，2 ,...z 6，t , ..., z 6,T ]; z 6，t =[z 6,t，1 ,...z 6,t,q , ..., z 6,t,Q ] P ;

[0124] Among them, the real-time active power measurement z3 and the real-time reactive power measurement z4 contain bad data; t = 1, 2, ..., T, T is the sequence number of historical measurement data and real-time measurement data, that is, the number of sections at the estimated time; m = 1, 2, ..., n, n is the number of distribution network nodes; q = 1, 2, ..., Q, Q is the number of installed synchronous phasor measurement devices; [] P Indicates transpose.

[0125] In a preferred embodiment of the present invention, S1 further includes:

[0126] The true value data of the distribution network to be estimated is obtained through distribution network flow calculation. The true value data includes:

[0127] Active power measurement true value V1 = [V 1，2 , ... V 1，t , ..., V 1，T ]; V 1，t =[V 1，2，1 , ... V 1，t，m , ..., V 1，T，n ] P ;

[0128] Reactive power measurement true value V2 = [V 2，2 , ... V 2，t , ..., V 2，T ]; V 2，t =[V 2，2，1 , ... V 2，t，m , ..., V 2，T，n ] P ;

[0129] The true value of the voltage amplitude of each node VV1=[VV 1，2 ,...VV 1，t , ..., V 1，T ]; VV 1，t =[VV 1，2，1 ,...VV 1，t，m , ..., VV 1，T,n ] P ;

[0130] The true value of the voltage phase angle of each node VV2 = [VV 2，2 ,...VV 2，t , ..., VV 2，T ]; VV 2，t =[VV 2，2，1 ,...VV 2，t，m , ..., VV 2，T,n ] P .

[0131] In a preferred embodiment of the present invention, before S2, the method further includes:

[0132] Constructing a measurement bad data processing model based on a gated recurrent unit, that is, constructing a first model, including:

[0133] Construct a bad data processing model with an input layer, a GRU layer, a fully connected layer, and an output layer. The input of the bad data processing model is x t，h , the output is y t,h ;x t，h and t，h The relationships between them include:

[0134] x t,h =[x h,t，1 ,...x h,t，m , ..., x h,t，n ] P =z h,t

[0135] z h，t =[z h,t，1 ,...z h,t，m , ..., z h,t，n ] P

[0136] x t，h,m =z h,t,m

[0137] h t =f(x t,h )

[0138] s t =σ(W (s) x t，h +U (s) h t-1 )

[0139] r t =σ(W (r) x t，h +U (r) h t-1 )

[0140]

[0141]

[0142] a1=σ(w1h t +b1)

[0143] y t,h =a1

[0144] y t,h =[y h,t，1 ,...y h,t，m , ..., y h,t，n ] P

[0145] Where, f() represents the mapping function; W (s) , U (s) , W (r) and U (r) is a trainable parameter; h t and a1 are the outputs of the GRU layer and the fully connected layer respectively; σ is the sigmoid activation function; tanh is the hyperbolic tangent function; represents matrix multiplication; wl is the weight of the fully connected layer; b l is the bias of the fully connected layer; when h = 1, z h，t Indicates historical active power measurement; when h = 2, z h，t Indicates historical reactive power measurement;

[0146] The loss function of the bad data processing model can be expressed as follows:

[0147]

[0148] Among them, y t,h =a1, when h = 1, y h,t，m and V h，t，m They represent active power and true value of active power respectively. When h=2, y h,t，m and V h，t，m Represent reactive power and reactive power true value respectively.

[0149] S2. Train the first model based on historical measurement data to obtain a second model; the first model includes a measurement bad data processing model based on a gated recurrent unit. S2 specifically includes:

[0150] Translate the measured bad data processing model training into the first optimization model:

[0151]

[0152] in, Represents the loss function, min represents minimization;

[0153] The measurement bad data processing model is iteratively trained through the first optimization model and historical measurement data to obtain a second model.

[0154] S3, obtaining a pseudo measurement value at the estimation time according to the second model, replacing bad data in the real-time measurement data with the pseudo measurement value, and obtaining first measurement data. S3 specifically includes:

[0155] Generate a pseudo measurement value y at the estimation time t according to the second model h,t,m ;

[0156] Replace the bad data in real-time active power measurement z3 and real-time reactive power measurement z4 with pseudo measurement value y h,t,m , get the real-time power measurement data of the distribution network, that is, get the first measurement data z′ h，t .

[0157] In a preferred embodiment of the present invention, before S4, the step further includes:

[0158] Construct a robust estimation model based on neural network, namely the third model, including:

[0159] Construct a robust estimation model based on a neural network, which consists of one input layer, five fully connected layers and two output layers. The input of the robust estimation model is the measurement data z t，h , the outputs are respectively the voltage amplitude pre-estimated state quantity X v , voltage phase angle pre-estimation state quantity X θ ; The relationship between the input and output of the robust estimation model includes:

[0160] z′ h，t =[z′ 3,t ; z′ 4,t ]

[0161] z t，h =[z′ h,t ;z 5,t ;z 6,t ]

[0162] c1=σ(ω1z t，h +bis1)

[0163] c2=σ(ω2c1+bis2)

[0164] c3=σ(ω3c2+bis3)

[0165] c4=σ(ω4C3+bis4)

[0166] c5=tanh(ω5C3+bis5)

[0167] X v =c4

[0168] X θ =c5

[0169]

[0170] X v =[X 1，t，1 , ...X 1，t，m , ..., X 1，t，n ]

[0171] X θ =[X 2，t，1 , ...X 2,t，m , ..., X 2，t,n ]

[0172] X h,t =[X h，t，1 , ...X h，t，m , ..., X h，t，n ]

[0173] Among them, z′ 3，t and z′ 4，tThey are respectively the active power measurement and reactive power measurement after bad data processing and replacement; c1, c2, c3, c4 and c5 are the outputs of the first to fifth fully connected layers respectively; ω l is the weight of the lth fully connected layer; bis l is the bias of the lth fully connected layer, l = 1, 2, 3, 4, 5; X h，t Estimate output for the network;

[0174] The loss function of the robust estimation model can be expressed by the following relationship:

[0175]

[0176] When h = 1, X h,t，m and VV h，t，m represent the estimated value of voltage amplitude and the true value of voltage amplitude respectively; when h=2, X h，t，m and VV h，t，m They represent the estimated voltage phase angle and the true voltage phase angle respectively.

[0177] S4, obtaining a distribution network state estimation result according to the first measurement data, the real-time measurement data, the third model and the fourth model; the third model includes a robust estimation model based on a neural network; the fourth model includes a state estimation model based on linear estimation. S4 specifically includes:

[0178] S41, transcribe the third model training into the second optimization model:

[0179] min L ω，bis (VV h，t , X h，t )

[0180]

[0181] in, Represents the loss function, min represents minimization;

[0182] The third model is obtained by combining the second optimization model and the first measurement data z′ h，t Perform iterative training to obtain the weights ω of each layer l And the bias bis l , and then the fifth model is obtained.

[0183] S42, at the estimation time t, the first measurement data z′ h，t , Real-time voltage measurement 5，t and real-time current measurement 6，t Input the fifth model, the fifth model outputs the first estimated value X h，t .

[0184] In a preferred embodiment of the present invention, before S43, the following steps are further included:

[0185] Constructing a state estimation model based on linear estimation, i.e., the fourth model, including:

[0186] The input Z of the state estimation model t，h Includes first estimate X h，t , Real-time voltage measurement 5，t and real-time current measurement 6，t ; Input Z of the state estimation model t，h The estimated value of the output voltage state The relationships include:

[0187] Z t,h =[X h，t ;z 5,t ;z 6,t ]=H*xx h，t +ε

[0188] xx h,t =[H P R -1 H] -1 H P R -1 Z

[0189] xx h,t,m =[xx h,t，1 ,...xx h,t，m ,...,xx h,t，n ]

[0190] U=[xx 1,t，1 ,...xx 1,t，m ,...,xx 1,t,n ]

[0191] θ=[xx 2,t，1 ,...xx 2,t，m ,...,xx 2,t,n ]

[0192] xx h,t =[xx h,t，1 ,...xx h,t，m ,...,xx h,t，n ]

[0193] Where H is the conversion matrix; ε is the measurement error; R is the measurement error weight matrix; xx h，t is the voltage state estimation value. When h=1, it indicates the estimated value of the voltage amplitude. When h=2, it indicates the estimated value of the voltage phase angle.

[0194] S43, based on the first estimated value X h，t, real-time voltage measurement z5 and real-time current measurement z6 obtain voltage state estimation values ​​through the fourth model, and the voltage state estimation values ​​include estimated values ​​of voltage amplitude and voltage phase angle.

[0195] In a preferred embodiment of the present invention, it also includes:

[0196] Calculate the first indicator, including:

[0197] Root mean square error RMSE mea，t And the mean absolute error MAE mea，t :

[0198]

[0199] Evaluate the accuracy of the measurement data according to the first indicator;

[0200] Calculate the second index, including:

[0201] Mean absolute error (MAE) t and the average estimated time T s :

[0202] VV h，t =[VV h,t，1 ,...VV h，t，m , ..., VV h，t，n ]

[0203]

[0204] Among them, Num represents the number of estimations, t i represents the estimated time of the i-th time;

[0205] The state estimation accuracy is evaluated according to the second indicator.

[0206] In a preferred embodiment of the present invention, a power distribution network IEEE 33-node system is set, that is, the value of n in the method of the present invention is 33; the method of the present invention is used to complete the state estimation of the distribution network, and the results are compared with those of the existing scheme, including:

[0207] In order to demonstrate the effectiveness of the method of the present invention in improving measurement accuracy and state estimation accuracy, experimental comparisons are made with the traditional pseudo-measurement linear interpolation method (method 1), BPN (method 2), traditional WLS (method 3), and WLAV (method 4) estimation methods. In terms of topological structure, four common topological structures are selected as the types of distribution network topology transformation. The topological types are shown in Table 1.

[0208] Table 1 Topology type of IEEE33 node example setting

[0209]

[0210] (1) Verification of the effectiveness of the bad data processing model of the present invention in improving measurement and estimation accuracy

[0211] Comparison of measurement accuracy indicators under different pseudo-measurement methods: Measurement root mean square error (RMSE mea,t ), measurement mean absolute error (MAE mea,t ), the smaller the error, the higher the accuracy. The measurement accuracy index calculation used is as follows:

[0212]

[0213] Among them, RMSE mea，t Represents the measurement root mean square error corresponding to the estimated time t, MAE mea，t represents the mean absolute error of the pseudo-measurement corresponding to the estimated time t. The root mean square error obtained under different pseudo-measurement methods is shown in Table 2, and the comparison results of the absolute error of pseudo-measurement are shown in Figure 2 .

[0214] Table 2 Comparison of pseudo-measurement root mean square error

[0215]

[0216] From Table 2 and Figure 2 The results show that the bad data processing method proposed in the present invention has smaller root mean square error and mean absolute error than method 1 and method 2. This shows that the bad data processing method of the present invention is better than method 1 and method 2. This is because the proposed method can reduce the impact of the power fluctuation of distributed power sources and the complex changes of load on the measurement accuracy by adaptively learning the historical data sequence, and improve the quality of measurement data by generating high-precision pseudo-measurement values ​​to replace the bad data in the measurement.

[0217] (2) Verification of the effectiveness of the robust estimation model and state estimation model of the present invention in improving estimation accuracy and estimation efficiency

[0218] Comparison of state estimation accuracy and efficiency indicators under different pseudo-measurement state estimation methods: mean absolute error (MAE t ), average estimated time (T s ), the smaller the error, the higher the accuracy, and the shorter the estimation time, the higher the estimation efficiency. The state estimation accuracy and estimation efficiency indicators used are calculated as follows:

[0219]

[0220] Among them, RMSE t Represents the root mean square error of the state estimation corresponding to the estimation time t, MAE t represents the mean absolute error of the state estimation corresponding to the estimation time t, T srepresents the average estimation time under 500 estimations, t i is the estimation time of the ith time, and Num=500 represents the number of estimations. The average absolute error and estimation time of state estimation of different methods under different topologies are shown in Tables 3 and 4.

[0221] Table 3 Comparison of mean absolute error of state estimation

[0222]

[0223]

[0224] Table 4 Comparison of average estimated time

[0225]

[0226] It can be seen from Table 3 and Table 4 that the average absolute error of the voltage amplitude under different topologies of the method of the present invention is smaller than that of Method 3 and Method 4, and the average absolute error of the voltage phase angle is smaller than that of Method 3, which is better than the estimation results of most topologies under Method 4. In addition, the average estimation time is shorter than that of Method 3 and Method 4. This shows that the robust estimation model and state estimation model of the present invention overcome the adverse effects of bad data and incorrect topology parameters on estimation accuracy through network estimation and linear estimation, improve estimation accuracy and estimation efficiency, and ensure the robustness of estimation under bad data and topology changes.

[0227] In summary, the distribution network state estimation method considering bad data and topological transformation of the present invention uses a bad data processing model based on a gated cyclic unit to process bad data. The model uses historical measurement data sequence adaptive learning to generate high-precision pseudo-measurement values ​​to replace bad data in real-time measurement data. Compared with traditional solutions, this method improves the measurement data accuracy; the robust estimation model based on a neural network used in the present invention does not need to input topological parameter information, overcomes the adverse effects of incorrect topological parameters on the decline in state estimation accuracy, and ensures the robustness of state estimation under bad data and topological changes; the state estimation model based on linear estimation used in the present invention introduces an estimated value, increases measurement redundancy, and further improves the estimation accuracy. The present invention uses a robust state estimation model that considers the influence of bad data and topological transformation, effectively improves the accuracy and robustness of state estimation under bad data and topological transformation, and can ensure power grid safety monitoring and operation control.

[0228] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A distribution network state estimation method considering bad data and topology transformation, characterized in that: The following steps are involved: S1. Obtaining historical measurement initial data and real-time measurement initial data of the distribution network to be estimated; The real-time measurement initial data includes a preset proportion of bad data; the historical measurement initial data and the real-time measurement initial data are pre-processed to obtain historical measurement data and real-time measurement data; S2. Training a first model according to the historical measurement data to obtain a second model; the first model includes a measurement bad data processing model based on a gated recurrent unit; S3. Obtain a pseudo measurement value at the estimation time according to the second model, and replace bad data in the real-time measurement data with the pseudo measurement value to obtain first measurement data; S4. Obtaining a distribution network state estimation result according to the first measurement data, the real-time measurement data, the third model and the fourth model; The third model includes a robust estimation model based on a neural network; and the fourth model includes a state estimation model based on linear estimation.

2. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 1, characterized in that: Preprocessing the historical measurement initial data and the real-time measurement initial data includes: Performing a first preprocessing on bad data in the real-time measurement initial data; the first preprocessing includes adding a first preset percentage of the amplitude value of the real-time measurement initial data; Performing a second preprocessing on the normal data in the real-time measurement initial data and the historical measurement initial data; the second preprocessing includes adding a second preset percentage of the amplitude value of the real-time measurement initial data; The historical measurement data is obtained, including Historical active power measurement z1 = [z 1，1 ,...z 1，t , ..., z 1，T-1 ]; z 1，t =[z 1,t，1 ,...z 1,t，m , ..., z 1,t，n ] P ; Historical reactive power measurement z2 = [z 2，1 ,...z 2，t , ..., z 2，T-1 ]; z 2，t =[z 2,t，1 ,...z 2,t，m , ..., z 2,t,n ] P ; Get real-time measurement data, including: Real-time active power measurement z3 = [z 3，2 ,...z 3，t , ..., z 3,T ]; z 3，t =[z 3,t，1 ,...z 3,t，m , ..., z 3,t,n ] P ; Real-time reactive power measurement z4=[z 4，2 ,...z 4，t , ..., z 4,T ]; z 4，t =[z 4,t，1 ,...z 4,t，m , ..., z 4，t,n ] P ; Real-time voltage measurement z5 = [z 5，2 ,...z 5，t , ..., z 5,T ]; z 5，t =[z 5,t，1 ,...z 4,t,q , ..., z 4，t，Q ] P ; Real-time current measurement 6，2 ,...z 6，t , ..., z 6,T ]; z 6，t =[z 6,t，1 ,...z 6,t,q , ..., z 6,t,Q ] P ; Wherein, the real-time active power measurement z3 and the real-time reactive power measurement z4 contain bad data; t=1,2,...,T, T is the sequence number of the historical measurement data and the real-time measurement data, that is, the number of sections at the estimated time; m=1,2,...,n, n is the number of distribution network nodes; q=1,2,...,Q, Q is the number of installed synchronous phasor measurement devices; [] P Indicates transpose.

3. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 2, characterized in that: The S1 further comprises: The true value data of the distribution network to be estimated is obtained by calculating the distribution network flow, and the true value data includes: Active power measurement true value V1 = [V 1，2 , ... V 1，t , ..., V 1，T ]; V 1，t =[V 1，2，1 , ... V 1，t，m , ..., V 1，T，n ] P ; Reactive power measurement true value V2 = [V 2，2 , ... V 2，t , ..., V 2，T ]; V 2，t =[V 2，2，1 , ... V 2，t，m , ..., V 2，T，n ] P ; The true value of the voltage amplitude of each node VV1=[VV 1，2 ,...VV 1，t , ..., V 1，T ]; VV 1，t =[VV 1，2，1 ,...VV 1，t，m , ..., VV 1，T,n ] P ; The true value of the voltage phase angle of each node VV2 = [VV 2，2 ,...VV 2，t , ..., VV 2，T ]; VV 2，t =[VV 2，2，1 ,...VV 2，t，m , ..., VV 2，T,n ] P .

4. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 3, characterized in that: Before said S2, it includes: Constructing a measurement bad data processing model based on a gated recurrent unit, that is, constructing the first model, including: Construct a measurement bad data processing model including an input layer, a GRU layer, a fully connected layer and an output layer. The input of the measurement bad data processing model is x t,h , the output is y t,h ;x t，h and t，h The relationships between them include: x t,h =[x h,t，1 ,...x h,t，m ,...,x h,t，n ] P =Z h,t With h，t =[of h,t，1 ,...With h,t，m ,...,With h,t，n ] P x t，h,m =z h,t, m h t =f(x t,h ) s t =σ(W (s) x t，h +U (s) h t-1 ) r t =σ(W (r) x t，h +U (r) h t-1 ) a1=σ(w1h t +b1) y t,h =a1 and t,h =[and h,t，1 ,...and h,t，m ,...,and h,t，n ] P Where, f() represents the mapping function; W (s) , U (s) , W (r) and U (r) is a trainable parameter; h t and a1 are the outputs of the GRU layer and the fully connected layer respectively; σ is the sigmoid activation function; tanh is the hyperbolic tangent function; represents matrix multiplication; w l is the weight of the fully connected layer; b l is the bias of the fully connected layer; when h = 1, z h，t Indicates historical active power measurement; when h = 2, z h，t Indicates historical reactive power measurement; The loss function of the measurement bad data processing model can be expressed by the following relationship: Among them, y t,h =a1, when h = 1, y h,t，m and V h，t，m They represent active power and true value of active power respectively. When h=2, y h,t，m and V h，t，m Represent reactive power and reactive power true value respectively.

5. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 4, characterized in that: The S2 includes: The measured bad data processing model training is transcribed into a first optimization model: in, Represents the loss function, min represents minimization; The measurement bad data processing model is iteratively trained through the first optimization model and the historical measurement data to obtain the second model.

6. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 5, characterized in that: The S3 includes: Generate a pseudo measurement value y at the estimation time t according to the second model h,t,m ; Replace the bad data in the real-time active power measurement z3 and the real-time reactive power measurement z4 with the pseudo measurement value y h,t,m , obtain the real-time power measurement data of the distribution network, that is, obtain the first measurement data z′ h，t .

7. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 6, characterized in that: Before the S4, it includes: Constructing the neural network-based robust estimation model, i.e., the third model, includes: Construct a robust estimation model based on a neural network, which includes an input layer, five fully connected layers and two output layers. The input of the robust estimation model is the measurement data z t，h , the outputs are respectively the voltage amplitude pre-estimated state quantity X v , voltage phase angle pre-estimation state quantity X θ ; The relationship between the input and output of the robust estimation model includes: With' h，t =[z′ 3,t ;With' 4,t ] With t,h =[z′ h,t ;With 5,t ;With 6,t ] c1=σ(ω1z t，h +bis1) c2=σ(ω2c1+bis2) c3=σ(ω3c2+bis3) c4=σ(ω4c3+bis4) c5=tanh(ω5C3+bis5) X v =c4 X θ =c5 X v =[X 1，t，1 ,...X 1，t，m ,...,X 1,t，n ] X θ =[X 2，t，1 ,...X 2,t，m ,...,X 2,t,n ] X h，t =[X h，t，1 ,...X h,t，m ,...,X h，t，n ] Among them, z′ 3，t and z′ 4，t They are respectively the active power measurement and reactive power measurement after bad data processing and replacement; c1, c2, c3, c4 and c5 are the outputs of the first to fifth fully connected layers respectively; ω l is the weight of the lth fully connected layer; bis l is the bias of the lth fully connected layer, l = 1, 2, 3, 4, 5; X h，t Estimate output for the network; The loss function of the robust estimation model can be expressed by the following relationship: When h = 1, X h,t，m and VV h，t，m represent the estimated value of voltage amplitude and the true value of voltage amplitude respectively; when h=2, X h，t，m and VV h，t，m They represent the estimated voltage phase angle and the true voltage phase angle respectively.

8. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 7, characterized in that: The S4 includes: S41, transcribe the third model training into a second optimization model: min L ω，bis (VV h,t ,X h，t ) in, Represents the loss function, min represents minimization; The third model is obtained by combining the second optimization model and the first measurement data z′ h，t Perform iterative training to obtain the weights ω of each layer l And the bias bis l , and then the fifth model is obtained; S42, at the estimation time t, the first measurement data z′ h，t , Real-time voltage measurement 5，t and real-time current measurement 6,t Input the fifth model, and the fifth model outputs the first estimated value X h，t ; S43, according to the first estimated value X h，t , the real-time voltage measurement z5 and the real-time current measurement z6 obtain voltage state estimation values ​​through the fourth model, and the voltage state estimation values ​​include estimated values ​​of voltage amplitude and voltage phase angle.

9. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 8, characterized in that: Before the S43, the following is included: Constructing the state estimation model based on linear estimation, i.e., the fourth model, includes: The input Z of the state estimation model is t，h Including the first estimated value X h，t , Real-time voltage measurement 5，t and real-time current measurement 6，t ; The input Z of the state estimation model t，h The estimated value of the output voltage state The relationships include: WITH t,h =[X h，t ;With 5,t ;WITH 6,t ]=H*xx h，t +ε xx h,t =[H P R -1 H] -1 H P R -1 Z xx h,t,m =[xx h,t，1 ,...xx h,t，m ,...,xx h,t，n ] U=[xx 1,t，1 ,...xx 1,t，m ,...,xx 1,t,n ] θ=[xx 2,t，1 ,...xx 2,t，m ,...,xx 2,t,n ] xx h，t =[xx h,t，1 ,...xx h,t，m ,...,xx h,t，n ] Where H is the conversion matrix; ε is the measurement error; R is the measurement error weight matrix; xx h，t is the voltage state estimation value. When h=1, it indicates the estimated value of the voltage amplitude. When h=2, it indicates the estimated value of the voltage phase angle.

10. The method for estimating the state of a distribution network considering bad data and topology transformation according to claim 9, characterized in that: Also includes: Calculate the first indicator, including: Root mean square error RMSE mea，t And the mean absolute error MAE mea，t : Evaluate the accuracy of the measurement data according to the first indicator; Calculate the second index, including: Mean absolute error (MAE) t and the average estimated time T s : VV h，t =[VV h，t，1 ,...VV h，t，m ,...,VV h，t，n ] Among them, Num represents the number of estimations, t i represents the estimated time of the i-th time; The state estimation accuracy is evaluated according to the second indicator.

Images