CN113488992A - Method for judging large disturbance stability of electric power system - Google Patents

Abstract

The invention belongs to the technical field of power system stability judgment, and relates to a method for judging the large disturbance stability of a power system. The invention considers the large disturbance power angle stability, the large disturbance voltage stability and the large disturbance dynamic stability, firstly, the voltage amplitude values of a rotor angle and a bus of a generator after large disturbance are collected, the generator and the bus which are seriously disturbed are obtained through the calculation of the mean value and the range, the voltage amplitude values of the rotor angle and the bus which are seriously disturbed of the generator which is seriously disturbed are respectively used as the input of a convolution layer, a pooling layer, a batch normalization layer and a full connection layer, then the two types of output are merged and input to the full connection layer to obtain the large disturbance stable output, the structure of a large disturbance discrimination model is formed, the parameter to be solved is solved to obtain the final large disturbance stable discrimination model, and the final large disturbance stable discrimination model is used for large disturbance stability discrimination. The invention comprehensively considers the large disturbance stability of various power systems, and only selects the characteristics of the generator and the bus with serious disturbance as input in the selection of input variables, so that the model has better applicability.

Description

Method for judging large disturbance stability of electric power system

Technical Field

The invention belongs to the technical field of power system stability judgment, and relates to a method for judging the large disturbance stability of a power system.

Background

The stability judgment of the power system after large disturbance is an important problem to be considered in power system mode calculation and safety prevention and control. The large disturbance stability of the power system comprises transient power angle stability, transient voltage stability and dynamic stability. In recent years, with the improvement of data in power systems and the development of information technology, data-driven transient power angle stabilization and transient voltage stabilization methods have attracted much attention.

In recent years, various artificial intelligence methods are applied to the field of large-disturbance stability analysis of power systems. The patent application publication number CN108832619A discloses a method for constructing a transient stability evaluation model based on a deep convolutional neural network, which uses bus characteristics as input, and obtains a stability label by considering the relative power angle difference after fault removal, that is, mainly considering the transient power angle stability. The chinese patent with application publication number CN112069723A proposes a transient stability evaluation model based on a convolutional neural network, which uses the generator characteristics as input and also uses the transient power angle stability after a fault as a learning target. The patent application publication number CN112290539A discloses a transient voltage stability margin evaluation method based on an extreme gradient boost model, which takes branch and bus characteristics as input to evaluate the transient voltage stability margin of a power system. The existing method usually only considers one of large disturbance stability, and different models are respectively trained for different stable subjects. However, various instability phenomena in an actual power grid are often interwoven together, and it is difficult to fully utilize the association relationship of different types of variables on the macro scale by adopting a mode of respectively training different models for the large disturbance stability of different topics, and the management of a plurality of models is relatively complex. In addition, the existing method usually adopts the characteristics of all generators or all buses and branches as input, and when the equipment input condition in the power grid changes, the equipment variables which are out of operation can only be considered as missing values individually.

Disclosure of Invention

The invention aims to provide a method for judging the large disturbance stability of an electric power system, which comprehensively considers the stability of a transient power angle, the stability of a transient voltage and the dynamic stability of the large disturbance, adopts a two-dimensional multi-input deep convolution neural network model for judging the stability of the large disturbance, and only selects relatively important electric quantity in a dynamic curve after the large disturbance as input in the selection of input variables so as to improve the applicability of the method for judging the stability under the condition of different equipment investment.

The method for judging the large disturbance stability of the power system firstly calculates the transient voltage stability, the transient power angle stability and the large disturbance dynamic stability of s operation conditions under f faults according to a time domain simulation calculation method and a safety and stability guide rule of the power system to obtain all the generator rotor angles delta of n sampling points after large disturbance under s multiplied by f operation scenes_i ^k(t) voltage amplitude V of all buses_j ^k(t) and a large disturbance stability tag y^k；

All generator rotor angles delta according to n sampling points after large disturbance_i ^k(t) voltage amplitude V of all buses_j ^k(t), selecting a severely disturbed generator and a severely disturbed bus, and calculating to obtain a characteristic vector X under the kth operation scene^k；

The feature vector X of each operation scene^kThe generator variable and the bus variable are respectively arranged into two groups of two-dimensional data, the two groups of two-dimensional data are respectively input into the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, and output is obtained to obtain the output o of the generator variable_k ¹And output o of the bus variable_k ²Changing the generator variable o_k ¹And output o of the bus variable_k ²Merging and inputting the merged data to a full connection layer to obtain the structure of a large-disturbance stability discrimination model M;

according to the characteristic vector X of s multiplied by f operation scenes^kLarge disturbance stability label y^kAnd a gradient descent algorithm based on self-adaptive moment estimation, and iteratively calculating the parameter W to be solved_l(1) ¹、H_g(1) ¹、W_l(2) ²、H_g(2) ²And H_g(3) ³Obtaining a final large disturbance stability discrimination model M;

and obtaining the rotor angles of all the generators and the voltage amplitudes of all the buses after the large disturbance, and obtaining a large disturbance stability judgment result of the power system by calculating and inputting the rotor angles and the voltage amplitudes into the large disturbance stability judgment model M of the power system.

The method for judging the large disturbance stability of the power system has the characteristics and advantages that:

the method for judging the large disturbance stability of the power system comprehensively considers the large disturbance power angle stability, the large disturbance voltage stability and the large disturbance dynamic stability, firstly, the rotor angle and the bus voltage amplitude of a generator after large disturbance are collected, then a generator and a bus with serious disturbance are obtained through calculation of mean value and range, the rotor angle of the generator with serious disturbance and the voltage amplitude of the bus with serious disturbance are respectively used as the input of a convolution layer, a pooling layer, a batch normalization layer and a full connection layer, then the two types of output are merged and input to the full connection layer to obtain large disturbance stable output, the structure of a large disturbance judgment model is formed, parameters to be solved are solved to obtain a final large disturbance stability judgment model, and the final large disturbance stability judgment model is used for large disturbance stability judgment. In the method, a two-dimensional depth convolution neural network model containing a power angle of a generator and a bus voltage amplitude as input characteristics is constructed, and only the generator with serious disturbance and the bus characteristics are selected as input in the selection of input variables. The method can comprehensively consider various instability phenomena of the power grid after large disturbance, and the model can be suitable for different operation scenes of the power grid due to the fact that the quantity of the input variables is screened.

Drawings

Fig. 1 is a flow chart of a method for determining the stability of large disturbance of an electric power system according to the present invention.

FIG. 2 is a schematic structural diagram of the large disturbance stability discriminant model in step (3) of the method of the present invention.

Detailed Description

The method for judging the large disturbance stability of the power system provided by the invention firstly calculates the transient voltage stability, the transient power angle stability and the large disturbance dynamic stability of s operation conditions under f faults according to a time domain simulation calculation method and a safety and stability guide rule of the power system to obtain the large disturbance stability under s multiplied by f operation scenesAll generator rotor angles delta of n sampling points_i ^k(t) voltage amplitude V of all buses_j ^k(t) and a large disturbance stability tag y^k；

All generator rotor angles delta according to n sampling points after large disturbance_i ^k(t) voltage amplitude V of all buses_j ^k(t), selecting a severely disturbed generator and a severely disturbed bus, and calculating to obtain a characteristic vector X under the kth operation scene^k；

Feature vector X of each scene^kThe generator variable and the bus variable are respectively arranged into two groups of two-dimensional data, the two groups of two-dimensional data are respectively input into the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, and output is obtained to obtain the output o of the generator variable_k ¹And output o of the bus variable_k ²Changing the generator variable o_k ¹And output o of the bus variable_k ²Merging and inputting the merged data to a full connection layer to obtain the structure of a large-disturbance stability discrimination model M;

according to the characteristic vector X of s multiplied by f operation scenes^kLarge disturbance stability label y^kAnd a gradient descent algorithm based on self-adaptive moment estimation, and iteratively calculating the parameter W to be solved_l(1) ¹、H_g(1) ¹、W_l(2) ²、H_g(2) ²And H_g(3) ³Obtaining a final large disturbance stability discrimination model M;

and obtaining the rotor angles of all the generators and the voltage amplitudes of all the buses after the large disturbance, and obtaining a large disturbance stability judgment result of the power system by calculating and inputting the rotor angles and the voltage amplitudes into the large disturbance stability judgment model M of the power system.

The method for judging the large disturbance stability of the power system comprises the following steps as shown in fig. 1:

(1) for an electric power system with N generators and M buses, transient voltage stability, transient power angle stability and large disturbance of s operation conditions under f faults are calculated according to a time domain simulation calculation method and an electric power system safety and stability guide ruleCalculating the dynamic stability to obtain all the generator rotor angles delta of n sampling points after large disturbance in s multiplied by f operation scenes_i ^k(t) voltage amplitude V of all buses_j ^k(t) and a large disturbance stability tag y^k：

Wherein, the superscript k represents the kth operation scenario, k is 1,2, …, sxf, the subscript i represents the ith generator in the power system, i is 1,2, …, N, the subscript j represents the jth bus in the power system, j is 1,2, …, M, t represents the tth sampling point, t is 1,2, …, N, N is the artificially set sampling point number, the time interval of sampling is manually specified, in one embodiment of the invention, the sampling point number N is 30, the sampling interval is 0.01s, the large disturbance stability label y of the power system in the kth operation scenario is y, the subscript i represents the ith generator in the power system, i is 1,2, …, N, the subscript j represents the jth bus in the power system, j is 1,2, …, M, t represents the tth sampling point number, t is 0.01s, and the large disturbance stability label y of the power system in the kth operation scenario is y^k＝(y_k ¹,y_k ²,y_k ³) Wherein, y_k ¹Representing the transient power angle stability of the power system after large disturbance in the kth operation scene by y_k ¹1 represents that the power system can not keep the transient power angle stable, and y represents that_k ¹When the value is 0, the power system can keep the transient power angle stable, and y_k ²Representing the transient voltage stability of the power system after large disturbance in the kth operation scene by y_k ²1 means that the power system cannot keep the transient voltage stable, and y is used_k ²When the value is 0, the power system can keep the transient voltage stable, and y_k ³Representing the dynamic stability of the power system after large disturbance in the kth operation scene by y_k ³1 means that the power system cannot be kept dynamically stable, and y is used_k ³When the value is 0, the power system does not generate divergent oscillation or continuous oscillation after large disturbance, namely, the power system can keep dynamic stability;

(2) all generator rotor angles delta according to n sampling points after large disturbance_i ^k(t) electricity of all busesAmplitude of pressure V_j ^k(t), selecting a severely disturbed generator and a severely disturbed bus, and calculating to obtain a characteristic vector X under the kth operation scene^kThe method comprises the following specific steps:

(2-1) obtaining all the generator rotor angles delta according to the step (1)_i ^k(t), calculating the relative rotor angle of a severely disturbed generators in the power system, and specifically comprising the following steps:

(2-1-1) sequentially calculating the relative rotor angles of all generators at a sampling point t in the kth operation scene of the power system

Wherein the content of the first and second substances,

the relative rotor angle of the ith generator at a sampling point t in the kth operation scene is represented, subscript i represents the ith generator in the power system, i is 1,2, …, N, t represents the tth sampling point, and t is 1,2, …, N;

(2-1-2) sequentially calculating the average value of the absolute values of the relative rotor angles of the ith generator at n sampling points in the kth operation scene

(2-1-3) sequentially calculating the pole difference of the relative rotor angle of the ith generator at n sampling points in the kth operation scene of the power system

(2-1-4) sorting the absolute average values of the relative rotor angles of all the generators in the kth operation scene of the power system obtained in the step (2-1-2) from large to small to obtain a sorting r of the absolute average value of the relative rotor angle of the generator i in the kth operation scene₁ ^k(i)，r₁ ^k(i) 1 represents that the absolute average value of the relative rotor angle of the generator i is arranged in the first position from large to small in the kth operation scene, all the relative rotor angle range differences of the generators in the kth operation scene of the power system obtained in the step (2-1-3) are sequenced from large to small, and the relative rotor angle range difference of the generator i is sequenced in the kth operation scene₂ ^k(i) According to the order r of the generator i in the k-th operating scenario₁ ^k(i) And rank r₂ ^k(i) And calculating the disturbance severity evaluation index D of the generator i in the kth operation scene_{gen_i} ^k：

(2-1-5) evaluating indexes D of disturbance severity of all generators in the kth operation scene obtained in the step (2-1-4)_{gen_i} ^kSorting from small to large, selecting a generators arranged in the front a as severely disturbed generators, and respectively marking the serial numbers as G₁ ^k，G₂ ^k，…，G_a ^kSequentially collecting the disturbed severe generator G in the kth operation scene₁ ^k，G₂ ^k，…，G_a ^kThe relative rotor angle at sampling point t is obtained

Wherein the value of a is set by human and satisfies 2<a<N，t＝1,2,…,n；

(2-2) obtaining the voltage amplitude V of all buses according to the step (1)_j ^k(t), sequentially calculating voltage amplitudes of b buses with the most serious disturbance of the power system in the kth operation scene, and specifically comprising the following steps:

(2-2-1) sequentially calculating the average value mean (V) of the voltage amplitudes of the jth bus at n sampling points in the kth operation scene of the power system_j ^k)：

Wherein, V_j ^k(t) represents the voltage amplitude of the jth bus at the sampling point t in the kth operation scene, wherein t is 1,2, …, n, the subscript j represents the jth bus in the power system, and j is 1,2, …, M;

(2-2-2) sequentially calculating the range (V) of the voltage amplitude of the jth bus at n sampling points in the kth operation scene of the power system_j ^k)：

(2-2-3) sorting the voltage amplitude average values of all buses of the power system obtained in the step (2-1-1) in the kth operation scene from small to large to obtain the sorting r of the voltage amplitude average value of the bus j in the kth operation scene₃ ^k(j) Sorting the range of the voltage amplitudes of all the buses obtained in the step (2-2-2) from large to small to obtain the sequence r of the range of the voltage amplitudes of the bus j in the kth operation scene₄ ^k(j) According to the sequence r of the bus j in the k-th operation scene₃ ^k(j) And rank r₄ ^k(j) And calculating the disturbed severity evaluation index D of the bus j in the kth operation scene_{bus_j} ^k：

(2-2-4) subjecting the product obtained in the step (2-2-3) toEvaluation index D for disturbed severity of all buses in k operating scenes_{bus_j} ^kSorting from small to large, selecting the buses arranged in the front B as seriously disturbed buses, and respectively marking the serial numbers as B₁ ^k，B₂ ^k，…，B_b ^kSequentially collecting the disturbed severe bus B in the kth operation scene₁ ^k，B₂ ^k，…，B_b ^kThe voltage amplitude at the sampling point t is obtained

Wherein the value of b is set by human and satisfies 2<b<M，t＝1,2,…,n；

(2-3) obtaining the relative rotor angle of the a severely disturbed generators according to the step (2-1-5)

And the voltage amplitudes of the b disturbed severe buses obtained in the step (2-2-4)

Feature vectors forming the kth operational scenario

(3) Feature vector X of each scene^kGenerator variable of

And bus bar variables

Respectively arranged into two groups of two-dimensional data, respectively input the two groups of two-dimensional data into the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, and output to obtain the output o of the generator variable_k ¹And output o of the bus variable_k ²Finally, varying the generatoro_k ¹And output o of the bus variable_k ²Combining and inputting the combined input to the full connection layer to obtain a structure of a large-disturbance-stability discrimination model M, wherein a structural schematic diagram of the large-disturbance-stability discrimination model M is shown in FIG. 2, and the method specifically comprises the following steps:

(3-1) subjecting the product obtained in the step (2-3)

Performing maximum and minimum normalization, arranging the data into two-dimensional data according to the generator dimension and the time dimension, wherein the dimension of the two-dimensional data is a multiplied by n, and inputting the two-dimensional data into c₁A convolution layer, p₁Individual pooling layer, z₁The batch normalization layer is then input to q after the output results are tiled₁A full connection layer to obtain an output o_k ¹Wherein, c is₁All the parameters to be obtained of the first (1) convolutional layer in each convolutional layer are recorded as W_l(1) ¹，l(1)＝1,…,c₁Q is prepared by₁All the parameters to be obtained of the g (1) th full connection layer in the full connection layers are recorded as H_g(1) ¹，g(1)＝1,…,q₁(ii) a Wherein the number and position of the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, the number and size of convolution kernels in the convolution layer, the size of the pooling layer and the number of neurons in the full-connection layer are all artificially determined, in one embodiment of the invention, 2 convolution layers, 2 pooling layers, 0 batch normalization layer and 1 full-connection layer are set in total, namely c₁＝2，p₁＝2，z₁＝0，q₁1, wherein the first layer is a convolutional layer, the first convolutional layer has 32 convolutional cores, the size of the convolutional cores is 5 × 5, the second layer is a pooling layer, the size of the pooling layer is 2 × 2, the third layer is a convolutional layer, the second convolutional layer has 64 convolutional cores, the size of the convolutional cores is 3 × 3, the fourth layer is a pooling layer, the size of the pooling layer is 2 × 2, the output of the fourth layer is tiled and then input into the fifth fully-connected layer, and the number of neurons of the fully-connected layer is 60;

(3-2) subjecting the product obtained in the step (2-3)

Arranging two-dimensional data with dimension of b x n according to bus dimension and time dimension, and inputting the two-dimensional data into c₂A convolution layer, p₂Individual pooling layer, z₂The batch normalization layer is then input to q after the output results are tiled₂A full connection layer to obtain an output o_k ²Wherein, c is₂The parameter to be obtained of each convolution layer is recorded as W_l(2) ²，l(2)＝1,…,c₂Q is prepared by₂All the parameters to be obtained of the g (2) th full connection layer in the full connection layers are recorded as H_g(2) ²，g(2)＝1,…,q₂(ii) a Wherein the convolution layer, the pooling layer, the batch normalization layer, the number and the position of the fully-connected layers, the number and the size of convolution kernels in the convolution layer, the size of the pooling layer and the number of neurons in the fully-connected layers are all artificially determined, and in one embodiment of the invention, 2 convolution layers, 2 pooling layers, 0 batch normalization layer and 1 fully-connected layer are set in total, namely c₁＝2，p₁＝2，z₁＝0，q₁1, wherein the first layer is a convolutional layer, the first convolutional layer has 32 convolutional cores, the size of the convolutional cores is 5 × 5, the second layer is a pooling layer, the size of the pooling layer is 2 × 2, the third layer is a convolutional layer, the second convolutional layer has 64 convolutional cores, the size of the convolutional cores is 3 × 3, the fourth layer is a pooling layer, the size of the pooling layer is 2 × 2, the output of the fourth layer is tiled and then input into the fifth fully-connected layer, and the number of neurons of the fully-connected layer is 120;

(3-3) subjecting o obtained in the step (3-1)_k ¹And o obtained in step (2-2)_k ²Are combined into [ o ]_k ¹，o_k ²]Then [ o ] will_k ¹，o_k ²]Inputting the parameters into w full-connection layers, and recording all the parameters to be obtained of the g (3) th full-connection layer in the w full-connection layers as H_g(3) ³G (3) ═ 1, …, w; the number of the full-connection layers w and the number of the neurons of the first w-1 full-connection layers are determined empirically, the output dimensionality of the w-th full-connection layer is 3, the activation function of the w-th full-connection layer is a sigmoid function, and the output is outputIs marked as O_k＝(O_k ¹,O_k ²,O_k ³) Wherein, if O_k ¹More than or equal to 0.5 represents that the transient power angle instability phenomenon exists after the system is greatly disturbed, if O_k ¹Less than 0.5 indicates that the system has no transient power angle instability phenomenon after large disturbance, if O_k ²More than or equal to 0.5 represents that the transient voltage instability phenomenon exists after the system is greatly disturbed, if O_k ²Less than 0.5 indicates that the system has no transient voltage instability phenomenon after large disturbance, if O_k ³More than or equal to 0.5 represents that the system has dynamic instability phenomenon after large disturbance, if O_k ³< 0.5 indicates that the system does not generate divergent oscillation or continuous oscillation after large disturbance, and in one embodiment of the invention, w is 2, wherein the number of the neurons in the first fully-connected layer is 30;

(4) according to the s multiplied by f operation scene characteristic vector X obtained in the step (2)^kAnd (2) obtaining a large-disturbance stability label y in step (1)^kAnd (3) iteratively calculating all parameters to be solved W in the step (3) based on a gradient descent algorithm of the adaptive moment estimation, namely Adam algorithm_l(1) ¹、H_g(1) ¹、W_l(2) ²、H_g(2) ²And H_g(3) ³And obtaining a final large-disturbance stability discrimination model M, wherein the adopted specific calculation formula of the loss function loss is as follows:

wherein N is_trainRepresenting the number of operational scenarios, N, selected as a training set from the s x f operational scenarios_trainIs manually set and satisfies 0.5 × s × f<N_train<s × f, the remainder s × f-N_trainUsing the kind of operation scene as the verification set, in one embodiment of the present patent, randomly selecting from the s × f kinds of operation scenes obtained in step (1)

Operation sceneAs training samples, i.e.

Remainder of

Taking a seed operation scene as a verification set, and taking a model with the highest accuracy of the verification set as a final large-disturbance stability discrimination model M;

(5) obtaining rotor angles of all generators and voltage amplitudes of all buses after large disturbance from simulation data of the power system or measurement data of a wide area measurement system, calculating and inputting the rotor angles and the voltage amplitudes into the power system large disturbance stability discrimination model M obtained in the step (4) to obtain a power system large disturbance stability discrimination result, and specifically comprising the following steps of:

(5-1) obtaining the angle delta of the generator rotor at n sampling points after large disturbance by using simulation data of the power system or directly collecting measurement data of a wide area measurement system_i ^new(t) bus Voltage amplitude V_j ^new(t) obtaining the relative rotor angle of the a severely disturbed generators by adopting the method in the step (2-1)

Obtaining the voltage amplitudes of the b disturbed severe buses by adopting the method in the step (2-2)

And (5-2) carrying out normalization processing on the relative rotor angles of the a disturbed severe generators obtained in the step (5-1), arranging the relative rotor angles into a x n two-dimensional data, arranging the voltage amplitudes of the b disturbed severe buses obtained in the step (5-1) into b x n two-dimensional data, and then inputting the a x n two-dimensional data and the b x n two-dimensional data into the large disturbance stability judgment model M of the power system obtained in the step (4) to obtain a large disturbance stability judgment result of the power system.

Claims (2)

1. A method for judging the stability of large disturbance of an electric power system is characterized in that the transient voltage stability, the transient power angle stability and the dynamic stability of large disturbance of s operation conditions under f faults are calculated according to a time domain simulation calculation method and the safety and stability guide rule of the electric power system, and the angle stability of all the generator rotors of n sampling points after large disturbance under s multiplied by f operation scenes is obtained

Voltage amplitude of all buses

And large disturbance stability label y^k；

According to all the generator rotor angles of n sampling points after large disturbance

Voltage amplitude of all buses

Selecting a severely disturbed generator and a severely disturbed bus, and calculating to obtain a characteristic vector X under the kth operation scene^k；

Feature vector X of each scene^kThe generator variable and the bus variable are respectively arranged into two groups of two-dimensional data, the two groups of two-dimensional data are respectively input into the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, and output is obtained to obtain the output o of the generator variable_k ¹And output o of the bus variable_k ²Changing the generator variable o_k ¹And output o of the bus variable_k ²Merging and inputting the merged data to a full connection layer to obtain the structure of a large-disturbance stability discrimination model M;

according to the characteristic vector X of s multiplied by f operation scenes^kLarge disturbance stability label y^kAnd a gradient descent algorithm based on self-adaptive moment estimation, and iteratively calculating the parameter W to be solved_l(1) ¹、H_g(1) ¹、W_l(2) ²、H_g(2) ²And H_g(3) ³Obtaining a final large disturbance stability discrimination model M;

and obtaining the rotor angles of all the generators and the voltage amplitudes of all the buses after the large disturbance, and obtaining a large disturbance stability judgment result of the power system by calculating and inputting the rotor angles and the voltage amplitudes into the large disturbance stability judgment model M of the power system.

2. The method for determining the stability of the large disturbance of the power system as claimed in claim 1, wherein the method comprises the following steps:

(1) for an electric power system with N generators and M buses, transient voltage stability, transient power angle stability and large disturbance dynamic stability of s operation conditions under f faults are calculated according to a time domain simulation calculation method and an electric power system safety and stability guide rule, and all generator rotor angle stability of N sampling points after large disturbance under s multiplied by f operation scenes are obtained

Voltage amplitude of all buses

And large disturbance stability label y^k：

The subscript k represents the kth operation scene, k is 1,2, …, s × f, the subscript i represents the ith generator in the power system, i is 1,2, …, N, the subscript j represents the jth bus in the power system, j is 1,2, …, M, t represents the tth sampling point, t is 1,2, …, N, N is the number of artificially set sampling points, and the large disturbance stability label y of the power system in the kth operation scene is marked by a large disturbance stability label y^k＝(y_k ¹,y_k ²,y_k ³) Wherein, y_k ¹Representing the transient power angle stability of the power system after large disturbance in the kth operation scene by y_k ¹1 represents that the power system can not keep the transient power angle stable, and y represents that_k ¹When the value is 0, the power system can keep the transient power angle stable, and y_k ²Representing the transient voltage stability of the power system after large disturbance in the kth operation scene by y_k ²1 means that the power system cannot keep the transient voltage stable, and y is used_k ²When the value is 0, the power system can keep the transient voltage stable, and y_k ³Representing the dynamic stability of the power system after large disturbance in the kth operation scene by y_k ³1 means that the power system cannot be kept dynamically stable, and y is used_k ³When the value is 0, the power system does not generate divergent oscillation or continuous oscillation after large disturbance, namely, the power system can keep dynamic stability;

(2) all generator rotor angles delta according to n sampling points after large disturbance_i ^k(t) voltage amplitude of all buses

Selecting a severely disturbed generator and a severely disturbed bus, and calculating to obtain a characteristic vector X under the kth operation scene^kThe method comprises the following specific steps:

(2-1) all the generator rotor angles obtained according to the step (1)

The method comprises the following steps of calculating the relative rotor angle of a severely disturbed generators in the power system, and specifically comprising the following steps:

(2-1-1) sequentially calculating the relative rotor angles of all generators at a sampling point t in the kth operation scene of the power system

Wherein the content of the first and second substances,

the relative rotor angle of the ith generator at a sampling point t in the kth operation scene is represented, subscript i represents the ith generator in the power system, i is 1,2, …, N, t represents the tth sampling point, and t is 1,2, …, N;

(2-1-2) sequentially calculating the average value of the absolute values of the relative rotor angles of the ith generator at n sampling points in the kth operation scene

(2-1-3) sequentially calculating the pole difference of the relative rotor angle of the ith generator at n sampling points in the kth operation scene of the power system

(2-1-4) sorting the absolute average values of the relative rotor angles of all the generators in the kth operation scene of the power system obtained in the step (2-1-2) from large to small to obtain a sorting r of the absolute average value of the relative rotor angle of the generator i in the kth operation scene₁ ^k(i) Sorting all the generators in the k operation scene of the power system obtained in the step (2-1-3) from large to small to obtain the relative rotor angle range of the generator i in the k operation scene₂ ^k(i) According to the order r of the generator i in the k-th operating scenario₁ ^k(i) And rank r₂ ^k(i) And calculating the disturbance severity evaluation index D of the generator i in the kth operation scene_{gen_i} ^k：

(2-1-5) evaluating indexes D of disturbance severity of all generators in the kth operation scene obtained in the step (2-1-4)_{gen_i} ^kSorting from small to large, selecting a generators arranged in the front a as severely disturbed generators, and respectively marking the serial numbers as G₁ ^k，G₂ ^k，…，G_a ^kSequentially collecting the disturbed severe generator G in the kth operation scene₁ ^k，G₂ ^k，…，G_a ^kThe relative rotor angle at sampling point t is obtained

Wherein the value of a is set by human and satisfies 2<a<N，t＝1,2,…,n；

(2-2) obtaining the voltage amplitude of all buses according to the step (1)

B buses with the most serious disturbance of the power system in the kth operation scene are obtained through calculation in sequence, and the method specifically comprises the following steps:

(2-2-1) sequentially calculating the average value mean (V) of the voltage amplitudes of the jth bus at n sampling points in the kth operation scene of the power system_j ^k)：

Wherein, V_j ^k(t) represents the voltage amplitude of the jth bus at the sampling point t in the kth operation scene, wherein t is 1,2, …, n, the subscript j represents the jth bus in the power system,j＝1,2,…,M；

(2-2-2) sequentially calculating the range (V) of the voltage amplitude of the jth bus at n sampling points in the kth operation scene of the power system_j ^k)：

(2-2-3) sorting the voltage amplitude average values of all buses of the power system obtained in the step (2-1-1) in the kth operation scene from small to large to obtain the sorting r of the voltage amplitude average value of the bus j in the kth operation scene₃ ^k(j) Sorting the range of the voltage amplitudes of all the buses obtained in the step (2-2-2) from large to small to obtain the sequence r of the range of the voltage amplitudes of the bus j in the kth operation scene₄ ^k(j) According to the sequence r of the bus j in the k-th operation scene₃ ^k(j) And rank r₄ ^k(j) And calculating the disturbed severity evaluation index D of the bus j in the kth operation scene_{bus_j} ^k：

(2-2-4) evaluating indexes D of disturbance severity of all buses in the kth operation scene obtained in the step (2-2-3)_{bus_j} ^kSorting from small to large, selecting the buses arranged in the front B as seriously disturbed buses, and respectively marking the serial numbers as B₁ ^k，B₂ ^k，…，B_b ^kSequentially collecting the disturbed severe bus B in the kth operation scene₁ ^k，B₂ ^k，…，B_b ^kThe voltage amplitude at the sampling point t is obtained

Wherein the value of b is set by human and satisfies 2<b<M，t＝1,2,…,n；

(2-3) obtaining the relative rotor angle of the a severely disturbed generators according to the step (2-1-5)

And the voltage amplitudes of the b disturbed severe buses obtained in the step (2-2-4)

Feature vectors forming the kth operational scenario

(3) Feature vector X of each scene^kGenerator variable of

And bus bar variables

Respectively arranged into two groups of two-dimensional data, respectively input the two groups of two-dimensional data into the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, and output to obtain the output o of the generator variable_k ¹And output o of the bus variable_k ²Finally, the generator variable o is set_k ¹And output o of the bus variable_k ²Merging and inputting the merged data to a full connection layer to obtain the structure of a large-disturbance stability discrimination model M, and specifically comprising the following steps of:

(3-1) subjecting the product obtained in the step (2-3)

Performing maximum and minimum normalization, arranging the data into two-dimensional data according to the generator dimension and the time dimension, wherein the dimension of the two-dimensional data is a multiplied by n, and inputting the two-dimensional data into c₁A convolution layer, p₁Individual pooling layer, z₁The batch normalization layer is then input to q after the output results are tiled₁A full connection layer to obtain an output o_k ¹Wherein, c is₁All the parameters to be obtained of the first (1) convolutional layer in each convolutional layer are recorded as W_l(1) ¹，l(1)＝1,…,c₁Q is prepared by₁All the parameters to be obtained of the g (1) th full connection layer in the full connection layers are recorded as H_g(1) ¹，g(1)＝1,…,q₁(ii) a The number and the positions of the convolution layer, the pooling layer, the batch normalization layer and the full-connection layer, the number and the size of convolution kernels in the convolution layer, the size of the pooling layer and the number of nerve cells in the full-connection layer are determined artificially;

(3-2) subjecting the product obtained in the step (2-3)

Arranging two-dimensional data with dimension of b x n according to bus dimension and time dimension, and inputting the two-dimensional data into c₂A convolution layer, p₂Individual pooling layer, z₂The batch normalization layer is then input to q after the output results are tiled₂A full connection layer to obtain an output o_k ²Wherein, c is₂The parameter to be obtained of each convolution layer is recorded as W_l(2) ²，l(2)＝1,…,c₂Q is prepared by₂All the parameters to be obtained of the g (2) th full connection layer in the full connection layers are recorded as H_g(2) ²，g(2)＝1,…,q₂(ii) a Wherein the convolution layer, the pooling layer, the batch normalization layer, the number and the position of the full-connection layers, the number and the size of convolution kernels in the convolution layer, the size of the pooling layer and the number of neurons in the full-connection layers are all determined artificially;

(3-3) subjecting o obtained in the step (3-1)_k ¹And o obtained in step (2-2)_k ²Are combined into [ o ]_k ¹，o_k ²]Then [ o ] will_k ¹，o_k ²]Inputting the parameters into w full-connection layers, and recording all the parameters to be obtained of the g (3) th full-connection layer in the w full-connection layers as H_g(3) ³G (3) ═ 1, …, w; it is composed ofThe number of the w full-link layers and the number of neurons of the first w-1 full-link layers are determined empirically, the output dimensionality of the w full-link layer is 3, the activation function of the w full-link layer is a sigmoid function, and the output is recorded as O_k＝(O_k ¹,O_k ²,O_k ³) Wherein, if O_k ¹More than or equal to 0.5 represents that the transient power angle instability phenomenon exists after the system is greatly disturbed, if O_k ¹Less than 0.5 indicates that the system has no transient power angle instability phenomenon after large disturbance, if O_k ²More than or equal to 0.5 represents that the transient voltage instability phenomenon exists after the system is greatly disturbed, if O_k ²Less than 0.5 indicates that the system has no transient voltage instability phenomenon after large disturbance, if O_k ³More than or equal to 0.5 represents that the system has dynamic instability phenomenon after large disturbance, if O_k ³< 0.5 indicates that the system does not exhibit divergent oscillations or sustained oscillations after a large disturbance;

(4) according to the s multiplied by f operation scene characteristic vector X obtained in the step (2)^kAnd (2) obtaining a large-disturbance stability label y in step (1)^kAnd (3) iteratively calculating all parameters to be solved W in the step (3) based on a gradient descent algorithm of the adaptive moment estimation, namely Adam algorithm_l(1) ¹、H_g(1) ¹、W_l(2) ²、H_g(2) ²And H_g(3) ³And obtaining a final large-disturbance stability discrimination model M, wherein the adopted specific calculation formula of the loss function loss is as follows:

wherein N is_trainRepresenting the number of operational scenarios, N, selected as a training set from the s x f operational scenarios_trainIs manually set and satisfies 0.5 × s × f<N_train<s × f, the remainder s × f-N_trainTaking the seed operation scene as a verification set;

(5) obtaining rotor angles of all generators and voltage amplitudes of all buses after large disturbance from simulation data of the power system or measurement data of a wide area measurement system, calculating and inputting the rotor angles and the voltage amplitudes into the power system large disturbance stability discrimination model M obtained in the step (4) to obtain a power system large disturbance stability discrimination result, and specifically comprising the following steps of:

(5-1) obtaining the angle delta of the generator rotor at n sampling points after large disturbance by using simulation data of the power system or directly collecting measurement data of a wide area measurement system_i ^new(t) bus Voltage amplitude V_j ^new(t) obtaining the relative rotor angle of the a severely disturbed generators by adopting the method in the step (2-1)

Obtaining the voltage amplitudes of the b disturbed severe buses by adopting the method in the step (2-2)

t＝1,2,…,n；

And (5-2) carrying out normalization processing on the relative rotor angles of the a disturbed severe generators obtained in the step (5-1), arranging the relative rotor angles into a x n two-dimensional data, arranging the voltage amplitudes of the b disturbed severe buses obtained in the step (5-1) into b x n two-dimensional data, and then inputting the a x n two-dimensional data and the b x n two-dimensional data into the large disturbance stability judgment model M of the power system obtained in the step (4) to obtain a large disturbance stability judgment result of the power system.

Images