A method for identifying transient DIM in power systems

Abstract

The present invention discloses a method for identifying transient DIM in a power system, comprising the following steps: 1) obtaining a number of bus voltage curves and power angle curves of the power system; 2) using the K-means clustering algorithm to perform typical fault curve screening and weight calculation on the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and the power angle instability curve; 3) using the MultiRocket algorithm to extract the multidimensional time series characteristics of the bus voltage instability curve and the power angle instability curve; 4) based on the multidimensional time series characteristics, constructing a basic model, and using the Bagging integrated learning and integrated optimization method to optimize the basic model to obtain a DIM identification model; 5) using the DIM identification model to complete the transient DIM identification of the power system. The present invention combines the error-bifurcation theory and proposes a DIM intelligent identification framework based on integrated optimization to improve the prediction accuracy and credibility.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and in particular to a topology change adaptive power system transient DIM (Dominant Instability Mode, data-driven dominant instability mode) identification method. Background Art

[0002] With the rapid development of my country's power grid, especially the construction of UHV projects and the access of a large number of new energy sources, the problems faced by power grid operation and control are becoming more complex, and higher requirements are put forward for the research on transient safe and stable operation of power grid.

[0003] Power system stability refers to the ability of a power system to maintain stable operation after being disturbed by a fault. There are two types of instability: power angle instability is caused by generator power control problems, which leads to the inability to recover the generator phase angle oscillation and eventually lose synchronization; voltage instability is the inability to maintain the voltage within the appropriate range, which may lead to voltage collapse.

[0004] In actual situations, power angle and voltage instability often occur simultaneously and are coupled with each other, which makes it difficult for power grid dispatchers to provide effective safety information and auxiliary decision-making. At the same time, power grid simulation data contains the inherent laws of power grid operation, but it is difficult to grasp due to the complexity of the power grid. Faced with such large and diverse data and models, manual analysis of power grid simulation data has been limited. In the future, with the increase in electromagnetic transient simulation calculations, the amount of data and model complexity will increase exponentially, and manual analysis of power grid operation laws will become more severe.

[0005] Taking the 40-second oscillation simulation of my country's power grid as an example, the simulation result data of the bus voltage of the main grid of the power grid (about 50,000 variables) can exceed 1GB. Every year, a lot of time, manpower and space are required to conduct fault interference simulation experiments and analysis to obtain DIM identification results, and then arrange the operation mode of the power system, formulate stable control strategies, and plan the power grid structure. Traditional power system transient stability analysis methods mainly include time domain simulation and direct method. Time domain simulation simulates the transient oscillation process of the power system by solving a group of differential-algebraic equations. It has high accuracy but extremely high computational burden. The direct method is based on energy function, which is fast but the results are conservative.

[0006] It is worth noting that the above methods can only determine the stability of the power system, but cannot directly obtain the transient DIM of the power system, making it difficult to effectively assist mode analysts to quickly determine the direction of emergency measures. Therefore, there is an urgent need for more effective data-driven methods to solve the problem of power grid DIM identification. Summary of the invention

[0007] The object of the present invention is to provide a method for identifying transient DIM in a power system, comprising the following steps:

[0008] Obtain bus voltage curves and power angle curves of several power systems;

[0009] Perform typical fault curve screening and weight calculation on the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and power angle instability curve;

[0010] Extract the multi-dimensional time series characteristics of bus voltage instability curve and power angle instability curve;

[0011] Based on the multi-dimensional time series characteristics, a basic model is constructed, and the basic model is optimized using bagging ensemble learning and ensemble optimization methods to obtain a DIM recognition model.

[0012] The DIM identification model is used to complete the transient DIM identification of the power system.

[0013] Further, the steps of obtaining a plurality of power system bus voltage curves and power angle curves include:

[0014] Setting fault conditions and performing simulation to obtain a bus voltage value matrix U and a generator power angle value matrix δ; the fault conditions include a power flow operation mode, a fault point location, a fault duration, and a fault type;

[0015] Based on the bus voltage value matrix and generator power angle value matrix under different fault conditions, the bus voltage curve and power angle curve of the power system are established;

[0016] The bus voltage value matrix U and the generator power angle value matrix δ are shown as follows:

[0017]

[0018] Where U T,B is the element in the bus voltage value matrix U; T, B is the number of rows and columns in the bus voltage value matrix U; δ T,M is the element in the generator power angle value matrix δ; T and M are the number of rows and columns in the generator power angle value matrix δ.

[0019] Further, the steps of screening typical fault curves and calculating weights for the bus voltage curve and power angle curve of the power system include:

[0020] Set the initial cluster center of K-means to the two curves with the largest and smallest change amplitudes in the same type of curves, denoted by C max and C min ; The types include power system bus voltage curve and power angle curve;

[0021]

[0022] Wherein, matrix A = U, δ;

[0023] The curves with a change amplitude greater than the preset threshold are classified into one category, and the curves are classified according to the C max The distances are sorted in ascending order; among them, curve A selected With curve C max The distance D i As shown below:

[0024] D i = distance(A selected ,C max ) (3)

[0025] In the formula, curve A selected =U selected ,δ selected ; U selected is the bus voltage curve of the power system with a variation greater than the preset threshold; selected A power angle curve having a variation greater than a preset threshold;

[0026] The weight distribution is calculated based on the distance, and the closer to the curve with the largest change amplitude C max The curve with higher weight is given, that is:

[0027]

[0028] Where W selected is the weight; D selected To be close to the curve C with the largest change max The curve of D max , D min The distance D i The maximum and minimum distance values ​​in ;

[0029] Fill the selected curves according to the weight ratio, and set the selected instability curve matrix as Where T is the duration of the observation window, N selected For the number of instability curves selected, the desired model input characteristic matrix is Where M is the number of fixed input curves required;

[0030] If N selected <M, then copy the instability curve matrix X to fill M bars, that is:

[0031] X input =[XX replicate ] T×M (5)

[0032] Among them, X replicate represents the copied curve matrix;

[0033] When N selected>M, in order to meet the limit of the number of input features, the instability curve matrix X is deleted according to the distance weight value W between the curves, that is:

[0034] X input =[x selected1 x selected2 ...x selectedM ] T×M (6)

[0035] Among them, x selectedi Represents the selected instability curve, corresponding to the distance weight value W selected The largest M curve.

[0036] Further, the step of extracting the multi-dimensional time series characteristics of the bus voltage instability curve and the power angle instability curve includes:

[0037] The time series X corresponding to the multi-dimensional time series characteristics of bus voltage instability curve and power angle instability curve is l}Converted to first-order difference form, we get:

[0038]

[0039] In the formula, X' is the first-order difference sequence; l is the number of time series elements;

[0040] Use the randomly expanded convolution kernel to transform the time series and obtain the feature vector Z, that is:

[0041]

[0042] Where, X i is the time series X of the i-th sample of the input, or the first-order difference series X'; ω is the weight, d is the expansion factor, and b is the bias of the kernel;

[0043] Calculate the characteristic index of the characteristic vector Z to construct the multidimensional time series feature F, that is:

[0044]

[0045] Where, X PPV , X MPV , X MIPV , X LSPV Indicates the corresponding positive value proportion, positive value average, positive value index average, and longest positive value extension when the time series X of the i-th sample is used as input; X' PPV , X' MPV , X' MIPV , X' LSPVIndicates the corresponding positive value proportion, positive value average, positive value index average, and longest positive value extension when the first-order difference sequence X' is used as input; vector F X =[X PPV X MPV X MIPV X LSPV ]; vector F X' =[X' PPV X' MPV X' MIPV X' LSPV ];

[0046] The randomly expanded convolution kernel uses a convolution kernel of length N, which contains multiple weight combinations with different values, and the weights are randomly selected when used;

[0047] Each convolution kernel dilation factor d = [2 0 ],...,[2 max ]; where the exponent max is as follows:

[0048]

[0049] In the formula, l input is the length of the input time series, l kernel is the convolution kernel length.

[0050] Furthermore, the characteristic indicators of the characteristic vector Z include the positive value proportion PPV(Z), the average value of the positive value MPV(Z), the average value of the positive value index MIPV(Z), and the longest positive value extension LSPV(Z);

[0051] Among them, the positive value PPV (Z) is as follows:

[0052]

[0053] Where Z represents the output of the convolution operation; Z i represents the threshold; n is the number of elements of the feature vector Z;

[0054] The mean value MPV(Z) for positive values ​​is as follows:

[0055]

[0056] In the formula, the positive parameter z + The length is m;

[0057] The average value of the positive index MIPV(Z) is as follows:

[0058]

[0059] In the formula, i +represents the index of the positive value, and m is the number of positive values ​​in Z.

[0060] The longest positive stretch LSPV(Z) is as follows:

[0061]

[0062] In the formula, i and j represent the minimum and maximum index values ​​of positive values, z k is the k-th index array of Z.

[0063] Furthermore, the basic model is as follows:

[0064]

[0065] In the formula, The probability distribution of the output categories of the basic model;

[0066] Among them, the probability p of the kth category k As shown below:

[0067]

[0068] In the formula, p k is the probability of the kth category; e is the base of the natural logarithm; K is the total number of categories;

[0069] Linear combination a k As shown below:

[0070]

[0071] In the formula, a k is the linear combination of the kth category, w ki is the weight connecting the i-th feature and the k-th category, b k is the bias of the kth category; F i As a feature.

[0072] Further, the steps of optimizing the basic model include:

[0073] Construct M basic models, and calculate the discrimination result matrix Q of the M basic models for the m DIM samples to be discriminated;

[0074] Among them, the element Q in the i-th row and j-th column of the discriminant result matrix Q ij As shown below:

[0075]

[0076] Let matrix Y = Q T Q, and normalize the elements in matrix Y to obtain:

[0077]

[0078] In the formula, m is the number of samples that were identified; Y ii , Y ij , Y jj is the element in matrix Y; is the element after normalization;

[0079] Construct a DIM recognition model, namely:

[0080]

[0081]

[0082] x i ={0,1} (22)

[0083] In the formula, x i Indicates whether to select the i-th basic model, and s is the optimized integrated model specification. is the objective function of the DIM identification model; is the matrix obtained by normalizing the matrix Y; vector x = [x1,...,x M ].

[0084] A system based on the power system transient DIM identification method includes:

[0085] Data acquisition unit: acquires several bus voltage curves and power angle curves of power systems;

[0086] Instability curve generation unit: Screening and weighting of typical fault curves for the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and power angle instability curve;

[0087] Feature extraction unit: extracts multi-dimensional time series features of bus voltage instability curve and power angle instability curve;

[0088] Model building unit: Based on the multi-dimensional time series characteristics, a basic model is built, and the basic model is optimized using bagging ensemble learning and ensemble optimization methods to obtain a DIM recognition model;

[0089] Transient DIM identification unit: Use the DIM identification model to complete the transient DIM identification of the power system.

[0090] An electronic device comprises: a memory, a processor and a data bus, wherein the memory stores all machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the data bus, and the machine-readable instructions are executed by the processor to execute the above method.

[0091] A computer-readable storage medium stores a computer program, which executes the above method when executed by a processor.

[0092] The technical effect of the present invention is undoubted, and the beneficial effects of the present invention are as follows:

[0093] 1) This invention proposes a basic model for DIM identification based on K-means clustering and MultiRocket. The model uses K-means clustering to select a fixed number of typical fault curves, which is consistent with the cognitive method of experts to judge the dominant instability mode, and can flexibly adapt to changes in the power system topology, thereby effectively improving the accuracy of DIM identification. At the same time, the MultiRocket multidimensional time series classification (TSC) method is used to extract time series features from the original time series and its first-order difference series. This process helps to capture the slope information of the voltage / power angle curve and determine the instability order of the two curves.

[0094] 2) The present invention proposes a DIM intelligent recognition framework based on ensemble optimization learning. In order to avoid the one-sidedness of a single model output, the Bagging ensemble learning and ensemble optimization framework are used to reduce the model calculation variance by constructing multiple base classifiers and integrating their outputs; in order to achieve the optimal adaptive selection of the number of base classifiers, the present invention combines the error-divergence theory and proposes a DIM intelligent recognition framework based on ensemble optimization to improve the prediction accuracy and credibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] Figure 1 It is a DIM intelligent recognition framework based on ensemble learning;

[0096] Figure 2 This is the single line diagram of the CEPRI-36 node system;

[0097] Figure 3 This is the single line diagram of the CEPRI-37 node system;

[0098] Figure 4 Label distribution of 7800 samples for a 36-node system.

[0099] Figure 5 Label distribution when samples are balanced for a 36-node system;

[0100] Figure 6 The correlation between the number of base models and the ensemble optimization performance. DETAILED DESCRIPTION

[0101] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0102] Embodiment 1:

[0103] See also Figures 1 to 6 , a topology change adaptive power system transient DIM identification method, comprising the following steps:

[0104] 1) Obtain bus voltage curves and power angle curves of several power systems;

[0105] 2) Use K-means clustering algorithm to screen typical fault curves and calculate weights for the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and power angle instability curve;

[0106] 3) Use the MultiRocket algorithm to extract the multidimensional time series characteristics of the bus voltage instability curve and the power angle instability curve;

[0107] 4) Based on the multidimensional time series characteristics, a basic model is constructed, and the basic model is optimized using bagging ensemble learning and ensemble optimization methods to obtain a DIM recognition model;

[0108] 5) Use the DIM identification model to complete transient DIM identification of the power system.

[0109] Embodiment 2:

[0110] See also Figures 1 to 6 , a topology change adaptive power system transient DIM identification method, the technical content is the same as embodiment 1, further, the step of obtaining a plurality of power system bus voltage curves and power angle curves includes:

[0111] 1.1) Set the fault conditions and perform simulation to obtain the bus voltage value matrix U and the generator power angle value matrix δ;

[0112] 1.2) Based on the bus voltage value matrix and generator power angle value matrix under different fault conditions, the bus voltage curve and power angle curve of the power system are established.

[0113] Embodiment 3:

[0114] See also Figures 1 to 6 , a topology change adaptive transient DIM identification method for a power system, the technical content of which is the same as any one of embodiments 1-2, and further, the fault conditions include flow operation mode, fault point location, fault duration and fault type.

[0115] Embodiment 4:

[0116] See also Figures 1 to 6 , a topology change adaptive power system transient DIM identification method, the technical content is the same as any one of embodiments 1-3, further, the bus voltage value matrix U and the generator power angle value matrix δ are as follows:

[0117]

[0118] Where U T,B is the element in the bus voltage value matrix U; T, B is the number of rows and columns in the bus voltage value matrix U; δ T,M is the element in the generator power angle value matrix δ; T and M are the number of rows and columns in the generator power angle value matrix δ.

[0119] Embodiment 5:

[0120] See also Figures 1 to 6 A method for topology change adaptive transient DIM identification of a power system, the technical content of which is the same as any one of embodiments 1-4, further comprising the steps of using a K-means clustering algorithm to perform typical fault curve screening and weight calculation on a bus voltage curve and a power angle curve of the power system:

[0121] 2.1) Set the initial K-means cluster center to the two curves with the largest and smallest change amplitudes in the same type of curves, denoted by C max and C min ; The types include power system bus voltage curve and power angle curve;

[0122]

[0123] Wherein, matrix A = U, δ;

[0124] 2.2) Classify the curves whose change amplitude is greater than the preset threshold into one category, and classify the curves according to the C max The distances are sorted in ascending order; among them, curve A selected With curve C max The distance D i As shown below:

[0125] D i = distance(A selected ,C max ) (3)

[0126] In the formula, curve A selected =U selected ,δ selected ; U selected is the bus voltage curve of the power system with a variation greater than the preset threshold;selected A power angle curve having a variation greater than a preset threshold;

[0127] 2.3) Calculate the weight distribution based on the distance, and the closer to the curve with the largest change amplitude C max The curve with higher weight is given, that is:

[0128]

[0129] Where W selected is the weight; D selected To be close to the curve C with the largest change max The curve of D max , D min The distance D i The maximum and minimum distance values ​​in ;

[0130] 2.4) Fill the selected curves according to the weight ratio, and set the selected instability curve matrix as Where T is the duration of the observation window, N selected For the number of instability curves selected, the desired model input characteristic matrix is Where M is the number of fixed input curves required;

[0131] If N selected <M, then copy the instability curve matrix X to fill M bars, that is:

[0132] X input =[XX replicate ] T×M (5)

[0133] Among them, X replicate represents the copied curve matrix;

[0134] When N selected >M, in order to meet the limit of the number of input features, the instability curve matrix X is deleted according to the distance weight value W between the curves, that is:

[0135] X input =[x selected1 x selected2 ...x selectedM ] T×M (6)

[0136] Among them, x selectedi Represents the selected instability curve, corresponding to the distance weight value W selected The largest M curve.

[0137] Embodiment 6:

[0138] See also Figures 1 to 6A method for identifying transient DIM of a power system with topology change adaptation, the technical content of which is the same as any one of Embodiments 1-5, and further, the step of extracting multi-dimensional time series characteristics of bus voltage instability curve and power angle instability curve by using MultiRocket algorithm comprises:

[0139] 3.1) The time series X corresponding to the multi-dimensional time series characteristics of the bus voltage instability curve and the power angle instability curve is l}Converted to first-order difference form, we get:

[0140]

[0141] In the formula, X' is the first-order difference sequence; l is the number of time series elements;

[0142] 3.2) Use the randomly expanded convolution kernel to transform the time series and obtain the feature vector Z, that is:

[0143]

[0144] Where, X i is the time series X of the i-th sample of the input, or the first-order difference series X'; ω is the weight, d is the expansion factor, and b is the bias of the kernel;

[0145] 3.3) Calculate the characteristic index of the characteristic vector Z to construct the multidimensional time series feature F, that is:

[0146]

[0147] Where, X PPV , X MPV , X MIPV , X LSPV Indicates the corresponding positive value proportion, positive value average, positive value index average, and longest positive value extension when the time series X of the i-th sample is used as input; X' PPV , X' MPV , X' MIPV , X' LSPV Indicates the corresponding positive value proportion, positive value average, positive value index average, and longest positive value extension when the first-order difference sequence X' is used as input. Vector F X =[X PPV X MPV X MIPV X LSPV ]; vector F X' =[X' PPV X' MPV X' MIPV X' LSPV ];

[0148] Embodiment 7:

[0149] See also Figures 1 to 6 , a topology change adaptive transient DIM identification method for a power system, the technical content of which is the same as any one of embodiments 1-6, further, a randomly expanded convolution kernel uses a convolution kernel of length N, which contains a plurality of weight combinations with different values, and the weights are randomly selected when used.

[0150] Each convolution kernel dilation factor d = [2 0 ],...,[2 max ]; where the exponent max is as follows:

[0151]

[0152] In the formula, linput is the length of the input time series, lkernel is the length of the convolution kernel;

[0153] Embodiment 8:

[0154] See also Figures 1 to 6 , a topology change adaptive transient DIM identification method for a power system, the technical content of which is the same as any one of embodiments 1-7, further, the characteristic indicators of the characteristic vector Z include the positive value proportion PPV(Z), the average value MPV(Z) of the positive value, the average value MIPV(Z) of the positive value index, and the longest positive value extension LSPV(Z);

[0155] Among them, the positive value PPV (Z) is as follows:

[0156]

[0157] Where Z represents the output of the convolution operation; Z i represents the threshold; n is the number of elements of the feature vector Z;

[0158] The mean value MPV(Z) for positive values ​​is as follows:

[0159]

[0160] In the formula, z + The length is m and PPV(Z)=|Z + | / n=m / n positive vector;

[0161] The average value of the positive index MIPV(Z) is as follows:

[0162]

[0163] In the formula, i + represents the index of the positive value, and m is the number of positive values ​​in Z.

[0164] The longest positive stretch LSPV(Z) is as follows:

[0165]

[0166] In the formula, i and j represent the minimum and maximum index values ​​of positive values, z k is the k-th index array of Z.

[0167] Embodiment 9:

[0168] See also Figures 1 to 6 , a topology change adaptive power system transient DIM identification method, the technical content is the same as any one of embodiments 1-8, further, the basic model is as follows:

[0169]

[0170] In the formula, The probability distribution of the output categories of the basic model;

[0171] Among them, the probability p of the kth category k As shown below:

[0172]

[0173] In the formula, p k is the probability of the kth category; e is the base of the natural logarithm; K is the total number of categories;

[0174] Linear combination a k As shown below:

[0175]

[0176] In the formula, a k is the linear combination of the kth category, w ki is the weight connecting the i-th feature and the k-th category, b k is the bias of the kth category; F i As a feature.

[0177] Embodiment 10:

[0178] See also Figures 1 to 6 , a topology change adaptive power system transient DIM identification method, the technical content is the same as any one of embodiments 1-9, further, the step of optimizing the basic model by using bagging ensemble learning and ensemble optimization method includes:

[0179] 4.1) Construct M basic models and calculate the discrimination result matrix Q of the M basic models for the m DIM samples to be discriminated;

[0180] Among them, the element Q in the i-th row and j-th column of the discriminant result matrix Q ij As shown below:

[0181]

[0182] 4.2) Let matrix Y = Q T Q, and normalize the elements in matrix Y to obtain:

[0183]

[0184] In the formula, m is the number of samples that were identified; Y ii , Y ij , Y jj is the element in matrix Y; is the element after normalization;

[0185] 4.3) Construct a DIM recognition model, namely:

[0186]

[0187] x i ={0,1} (22)

[0188] In the formula, x i Indicates whether to select the i-th basic model, and s is the optimized integrated model specification. is the objective function of the DIM identification model; is the matrix obtained by normalizing the matrix Y; vector x = [x1,...,x M ].

[0189] Embodiment 11:

[0190] A system based on the method for identifying transient DIM in a power system according to any one of embodiments 1 to 10 includes:

[0191] Data acquisition unit: acquires several bus voltage curves and power angle curves of power systems;

[0192] Instability curve generation unit: Screening and weighting of typical fault curves for the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and power angle instability curve;

[0193] Feature extraction unit: extracts multi-dimensional time series features of bus voltage instability curve and power angle instability curve;

[0194] Model building unit: Based on the multi-dimensional time series characteristics, a basic model is built, and the basic model is optimized using bagging ensemble learning and ensemble optimization methods to obtain a DIM recognition model;

[0195] Transient DIM identification unit: Use the DIM identification model to complete the transient DIM identification of the power system.

[0196] Embodiment 12:

[0197] An electronic device comprises: a memory, a processor and a data bus, wherein the memory stores all machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the data bus, and the machine-readable instructions are executed by the processor to perform the method described in any one of Examples 1-10.

[0198] Embodiment 13:

[0199] A computer-readable storage medium stores a computer program, which executes the method described in any one of embodiments 1-10 when executed by a processor.

[0200] Embodiment 14:

[0201] A transient DIM identification method for power systems that is adaptive to topology changes. This method uses the K-means algorithm to select typical power angles and voltage curves of simulation outputs to imitate the expert DIM identification process, achieve flexible adaptation to power system topology changes, and effectively improve the accuracy of DIM identification. Furthermore, based on the MultiRocket algorithm, time series feature extraction is performed, combined with Bagging ensemble learning and ensemble optimization technology to enhance the stability of the DIM identification model, reduce the model calculation variance, and ensure the accuracy and credibility of the DIM identification results. Finally, the effectiveness of the proposed method is verified by the examples of the 36-node system of China Electric Power Research Institute and its modified system as well as the actual power grid. This technological innovation is conducive to the improvement of transient stability analysis, control strategy formulation and power grid planning of power systems.

[0202] The specific steps are as follows:

[0203] (1) DIM intelligent recognition framework based on ensemble learning

[0204] The DIM intelligent recognition framework of the present invention is as follows Figure 1As shown, first, the fault conditions are set to generate samples. Then, the K-means clustering algorithm is used to screen the typical fault curves and calculate the weights of the voltage and power angle curves output by the simulation to solve the DIM problem of power system topology changes. Next, the MultiRocket algorithm is used to extract multidimensional time series features. Finally, Bagging ensemble learning and ensemble optimization methods are used to reduce model variance and improve the stability and credibility of DIM identification. Bagging generates multiple subsets by randomly resampling the training set. Each subset is used to train a different base classifier, and then their outputs are integrated to reduce the uncertainty of model prediction. For multi-classification problems with n single models, each single model can divide the input samples into m categories, using the representation model h i Predict the probability that sample x belongs to category j, where i represents the index of a single model and j represents the index of the category. The average predicted probability can be expressed as:

[0205]

[0206] Assume that each model h i The predictions between are uncorrelated, and considering the variance formula for multi-classification problems:

[0207]

[0208] Since each model h i It is independent of other models, so h i The variance of is common to all categories j and is denoted by σ 2 . Substitute the result into the variance formula of the integrated model:

[0209]

[0210] The results show that in multi-classification problems, the prediction variance of the bagging ensemble is the variance σ of a single model. 2 As the number of models n increases, Bagging integration can significantly reduce the variance of multi-category classification problems and improve the stability and generalization performance of the model.

[0211] The above methods do not clearly identify the best ensemble model because there is a lack of clear criteria for selecting the base models needed to build the ensemble model. To find the best ensemble model, we first need to have a deep understanding of the accuracy of the base models, the differences between these base models, and how they affect the accuracy of the ensemble model. The error-divergence theory provides a theoretical basis for the relationship between the accuracy of the base models and the base models. Specifically, it describes the use of n base models h1,…,h i ,…,h nTo build an integrated model, and obtain the output result by weighted average, as shown in the formula:

[0212]

[0213] In the formula, w i is the basic model h i The weight value is given by w i ≥0 and Constraints. When a sample x is given, the divergence of the i-th base model is defined as:

[0214] A(h i |x)=(h i (x)-H(x)) 2 (5)

[0215] Where: H(x) represents the output of the integrated model.

[0216] It follows that the divergence of the ensemble model can be defined as the weighted average of the divergence of the base models:

[0217]

[0218] Obviously, divergence can be used to measure the difference of the basic model on the sample. There are many ways to define the error. The present invention chooses to use the mean square error to measure the error of the basic model. Assuming that the true discrimination result of sample x is f(x), the error of the basic model hi(x) and the integrated model H(x) can be expressed as:

[0219] E(h i |x)=(f(x)-h i (x) 2 (7)

[0220] E(H|x)=(f(x)-H(x)) 2 (8)

[0221] The weighted average error of all basic models on the overall sample can be expressed as:

[0222]

[0223] By transforming formula (10), we can get:

[0224]

[0225] Combined Constraints And formula (4)-formula (8) is expanded into formula (10) to obtain:

[0226]

[0227] Arranged:

[0228]

[0229] It can be inferred from formula (12) that the error of the integrated model is determined by the weighted average error of all basic models on the overall sample and the weighted average of the differences between the basic models. Since the second term on the right side of the equation is positive and is subtracted from the first term, it is theoretically guaranteed that the error after integration must be lower than the weighted average error of the basic model. In addition, the higher the accuracy of the basic models, the greater the difference between them, and the higher the accuracy of the integrated model. Therefore, when further optimizing the DIM recognition framework based on the Bagging algorithm, the present invention adopts an integrated model optimization method. Considering M basic models and m DIM samples to be identified, the identification results are recorded in the matrix Q, and the element Q in the i-th row and j-th column of Q is ij As shown in formula (13):

[0230]

[0231] Let Y = Q T Q, then the diagonal element Y of Y ii Represents the basic model h i The number of misjudgments reflects the weighted average error of all basic models in the overall sample, rather than the diagonal element Y ij , then it means the basic model h i With h j The number of simultaneous errors reflects the divergence of the integrated model. Therefore, Y contains the errors and divergences of each basic model, which effectively measures the error of the integrated model. Use formula (14) to normalize the elements in Y. According to the error definition of the basic model in formula (12), when When each element in is the smallest, the ensemble model achieves the best effect.

[0232]

[0233] In the formula, m is the number of samples to be judged.

[0234] The integrated model optimization problem can be transformed into a quadratic integer programming problem:

[0235]

[0236] x i ={0,1} (17)

[0237] In the formula, x i Indicates whether to select the i-th basic model, and s is the optimized integrated model size. Based on the theoretical basis of formula (12), a quadratic integer programming model such as formula (13)-formula (17) is established to find the optimal DIM identification integrated model for a given integrated model size.

[0238] (2) Selection of typical fault curves based on K-means

[0239] The present invention sets different power flow operation modes, fault point locations, fault durations and fault types by using the PSASP software of China Electric Power Research Institute. Since voltage instability is closely related to voltage changes, and power angle instability is closely related to power angle changes, the output is set to the generator power angle curve and voltage curve to generate a large number of valid samples. Assume that the research system includes N buses. The characteristic observation window of the input model lasts for T seconds, and data sampling is performed every 0.01s. For any sample, the matrix composed of bus voltage values ​​and generator power angle values ​​is shown as follows:

[0240]

[0241] The present invention proposes a typical fault curve selection strategy based on the K-means clustering method to simulate the judgment ideas of experts when determining DIM: when the power system is unstable, not all generator power angles and bus voltage curves will be unstable, and often only the bus voltage or generator power angle near the fault point will have obvious instability. Therefore, when experts determine DIM, the focus of their analysis is also on these curves with obvious unstable changes, and the characteristics of these unstable curves will not change significantly due to changes in the power system. The core idea of ​​the typical fault curve method based on K-means proposed in the present invention is to automatically screen out obviously unstable curves from the generator power angle or bus voltage curve cluster, rather than the curves that can eventually restore stability. The method is summarized as follows:

[0242] First, the initial clustering centers of K-means are set as the two curves with the largest and smallest changes in the bus voltage curve (the same applies to the generator power angle curve), represented by C max and C min ,Right now:

[0243]

[0244] Then, the curves with larger changes are grouped into one category, and the curves are sorted according to the max The distances are sorted from small to large to highlight the severity of the disturbance, as shown in formula (20):

[0245] D i = distance(U selected ,C max ) (20)

[0246] Next, the weight distribution is calculated based on the distance, and the closer to the curve with the largest change amplitude C max The curve with higher weight is given as follows:

[0247]

[0248] Finally, fill the selected fault curves according to the weight ratio, and set the selected instability curve matrix as Where T is the duration of the observation window, N selected is the number of instability curves selected. The desired model input characteristic matrix is Where M is the required number of fixed input curves to ensure that the input scale is fixed at M curves, thereby solving the DIM problem under the change of power system topology.

[0249] If N selected <M, then copy the instability curve matrix X to fill M bars, that is:

[0250] X input =[XX replicate ] T×M (twenty two)

[0251] Among them, X replicate Represents the copied curve matrix, whose shape is.

[0252] When N selected >M, in order to meet the limit on the number of input features, the instability curve matrix X is deleted according to the distance weight value W between the curves:

[0253] X input =[x selected1 x selected2 …x selectedM ] T×M (twenty three)

[0254] Among them, x selectedi Represents the selected instability curve, corresponding to the distance weight value W selected The largest M curve.

[0255] (3) DIM discrimination model based on MultiRocket

[0256] DIM identification of power systems is essentially a TSC problem, which aims to identify patterns and signals related to different categories in data. Time series data is defined as a combination of data points containing detailed information about different moments, usually sampled at equal time intervals. The present invention aims to solve the DIM identification problem of power systems by using the TSC method to achieve the classification of the corresponding time series to reveal its inherent laws and dynamically changing parameters.

[0257] For a given time series X = {x1, x2, ..., x l}, it can be converted into the first-order difference by applying formula (24):

[0258]

[0259] MultiRocket uses a large number of randomly expanded convolution kernels to transform time series. It does not use hidden layers or any nonlinearity, and the features generated are independent of each other. Each convolution kernel is applied to each input time series and its first-order difference sequence to generate a feature vector Z, as shown in formula (25):

[0260]

[0261] Among them, X i It can be the time series X of the i-th sample of the input, or its first-order difference sequence X', ω is the weight, d is the expansion factor, and b is the bias of the kernel.

[0262] The random dilation convolution used here is different from the convolutional layer used in typical convolutional neural networks in the following ways:

[0263] 1) Length and weight ω: A convolution kernel with a length of 9 is used, which contains six combinations with a weight value ω of -1 and three combinations with a weight value of 2. There are 84 combinations in total, which are randomly selected when used.

[0264] 2) Dilation: Each convolution kernel uses the same or fixed dilation set, with dilation value d = [2 0 ],...,[2 max ], with indices evenly distributed between 0 and max, where max is set to:

[0265]

[0266] Among them, linput is the length of the input time series, and lkernel is the length of the convolution kernel.

[0267] 3) Bias: The bias of the kernel / dilation combination is extracted from the convolution output. The convolution output of a randomly selected training example is calculated and its quantile is taken as the bias.

[0268] 4) Padding: Padding is alternated between kernel / dilation combinations, half with standard zero padding and half without padding.

[0269] After the dilated convolution operation, MultiRocket calculates four features of each convolution output Z to summarize the values ​​in Z. These features are also called pooling operators. They are as follows:

[0270] 1) Proportion of Positive Values ​​(PPV): PPV is used to calculate the proportion of positive values ​​in the convolution output Z, which is directly related to the bias. The bias can be regarded as the "threshold" of PPV, see formula (27) for details. A positive bias means that PPV captures the proportion of the time series that reflects a weak match between the input and the given pattern, while a negative bias only captures the proportion of strong matches.

[0271]

[0272] Here, Z represents the output of the convolution operation.

[0273] In the PPV calculation, the bias Z is involved i As its "threshold", this is similar to the criterion for judging DIM: in the medium and long-term transient process, the voltage curve has not recovered to the threshold of 0.9pu, and the power angle curve is greater than the threshold of 200°.

[0274] 2) Mean of Positive Values ​​(MPV): MPV is used to measure the strength of positive values ​​in the convolution output, where Z is of length n, as shown in formula (28). In addition, MPV is related to the bias term and can evaluate the degree of match between the input time series and a given pattern at a low computational cost.

[0275]

[0276] Among them, z + The length is m and PPV(Z)=|Z + | / n=m / n positive vector.

[0277] The calculation of MPV pays attention to the degree of feature matching of the instability modes, which can reflect the degree of the two instability modes to a certain extent.

[0278] 3) Mean of Indices of Positive Values ​​(MIPV): MIPV can capture the relative position information of positive values ​​in the convolution output at a low computational cost. It treats the convolution output Z as an array and calculates the average of the indices of all positive values ​​in the array. The calculation method is shown in formula (29):

[0279]

[0280] Among them, i + represents the index of the positive value, and m is the number of positive values ​​in Z.

[0281] The calculation of MIPV can effectively extract the position information of the power angle or voltage curve, which helps to identify the order of instability of the two in the time series to assist in determining DIM.

[0282] 4) Longest Stretch of Positive Values ​​(LSPV): MIPV summarizes all positive values, but cannot distinguish between sequences of many consecutive small positive values ​​and sequences of a small number of long consecutive positive values. Therefore, LSPV is used to return the maximum length of the positive subsequence in the convolution output, calculated using formula (30):

[0283]

[0284] LSPV can effectively emphasize the duration of instability or stability in DIM judgment.

[0285] By default, MultiRocket generates approximately 49728 (using 6216 kernels × 2 time series inputs × 4 pooling operators) feature values, as shown in formula (31):

[0286]

[0287] The converted feature F is used to train a linear classifier. The present invention uses a Logistic classifier. For a given input feature F, the model calculates a linear combination a for each category k. k , as shown below:

[0288]

[0289] Among them, a k is the linear combination of the kth category, w ki is the weight connecting the i-th feature and the k-th category, b k is the bias for the kth class.

[0290] Next, the linear combination a k Converted to probability distribution of categories, this is done through the softmax function, as shown below:

[0291]

[0292] Among them, p k is the probability of the kth class, e is the base of the natural logarithm, and K is the total number of classes.

[0293] This step converts the linear combination of each category a k The probability of conversion to the corresponding category p k , so that the model outputs the probability distribution of the category. The final classification result is usually the category with the highest probability, that is:

[0294]

[0295] Embodiment 15:

[0296] The verification experiment of a topology change adaptive power system transient DIM identification method is as follows:

[0297] 1. Case study

[0298] In order to verify the effectiveness and accuracy of the proposed method, samples were generated on the 36-node system of China Electric Power Research Institute and its modified system and the actual power grid for example analysis. MultiRocket was built by Scikit-learn, and the example test platform was: AMD Ryzen 75800H with Radeon Graphics@3.20GHz CPU, 16GB RAM.

[0299] The CEPRI-36 node system mainly includes 36 bus nodes, 8 generator sets, 26 lines and 10 load nodes. Its single line diagram is as follows: Figure 2 If the load model is a constant impedance model, the voltage instability sample cannot be generated. Therefore, when generating samples, the load model components are set to constant impedance and induction motor, and the ratio of the two can be modified.

[0300] In order to verify that the proposed method is applicable to the change of power system topology, a bus, a generator set and a node are added to the CEPRI-36 node, named CEPRI-37 node system. The single-line diagram is as follows: Figure 3 .

[0301] 2. Sample generation

[0302] For the CEPRI-36 node system, the fault type is set to a three-phase permanent fault, the fault occurrence time is 1s, the duration is 0.1s, 0.2s, 0.25s, 0.3s, the occurrence location is 2%, 20%, 50%, 80%, 98% of all lines excluding the connected transformer, and the power flow level is 90%, 100%, 110%. The specific setting parameters are shown in Table 1.

[0303] Table 1 CEPRI-36 node system sample generation fault settings

[0304]

[0305] At the same time, in order to simulate the actual near-end and far-end tangent behavior of the power grid, according to the provisions of the relevant technical specifications of the State Grid on the fault removal time, the protection action time, intermediate relay time and circuit breaker full opening time are set. The present invention sets the fault removal action time on the near-end and far-end fault point side according to the above technical specification documents. For example, the near-end and far-end removal time of 200kV are both 1.12s, the near-end removal time of 500kV is 1.09s, and the far-end removal time is 1.1s. In order to simulate different working conditions in the actual system, on the basis of this specification, the fault removal times of 0.2s, 0.25s, and 0.3s are added. Finally, 7,800 labeled samples are obtained. In the generated samples, the distribution of the number of labels for stability, voltage instability, and power angle instability is as shown in the figure. Figure 4 To ensure sample balance, 2352 samples out of 7800 samples are used as training and test sets. The distribution of the three labels is shown in Figure 5 .

[0306] For the modified system of CEPRI-36-node system, a bus, a generator, and a node are added to the original system to form a 37-node system. Its fault type, fault occurrence time, occurrence location, and power flow level setting range are the same as those of the 36-node system. The specific settings are the same as Table 1, but the load motor component ratio is set to 30%, 40%, and 50%.

[0307] 3. Verification of the effectiveness of the proposed method

[0308] 1) 36-node system verification

[0309] The samples generated by 36 nodes are used to train and test the proposed method. The label distribution of the training set and the test set in the samples is shown in the figure. Figure 5 shown.

[0310] First, this study evaluates the accuracy of different ensemble model sizes s under test ensemble optimization. In each case, 5 independent tests were performed and the average test set accuracy was calculated. Figure 6 It can be clearly observed that the accuracy of integrated optimization reaches its peak when the number of basic models is 5. Therefore, this study selected 5 basic models as the experimental benchmark.

[0311] The example verification method of the present invention sets M0 to M3, and the setting comparison purpose is shown in Table 3:

[0312] M0: typical curve + 2D-CNN not selected;

[0313] M1: K-Means selects typical fault curves + 2D-CNN;

[0314] M2: K-Means selects typical fault curves + MultiRocket;

[0315] M3: The method proposed in the present invention takes M2 as the basic model and adds an integrated optimization framework.

[0316] Table 2 Design objectives of four examples to verify the methods

[0317]

[0318] The present invention adopts the following four index evaluation methods to verify the effect. The larger the evaluation index value, the better the effect. The test indicators of the four models at 36 nodes are shown in Table 3.

[0319] 1) Accuracy ACC:

[0320]

[0321] Among them, TP represents the number of true categories, TN represents the number of true negative categories, FP represents the number of false positive categories, and FN represents the number of false negative categories.

[0322] 2) Precision PRE:

[0323]

[0324] 3) Recall rate REC:

[0325]

[0326] 4) The harmonic mean F1 of precision and recall:

[0327]

[0328] Table 3 Comparison of indicators of four models

[0329]

[0330] As can be seen from Table 3, the comparison between the results of M0 and M3 verifies that the method proposed in the present invention has improved all indicators compared with the previous data-driven methods; the comparison between the results of M1 and M2 verifies that MultiRocket's TSC method is more suitable for DIM classification problems than 2D-CNN; the various indicators of M3 are better than M2, which verifies the stability advantage of the integrated model compared with a single basic model. At the same time, the 2D-CNN in M1 is replaced with the common data-driven methods 1D-CNN, SVM, KNN, DT, and RF for comparative testing. The indicators are shown in Table 4. The results show that among the data-driven methods for selecting typical fault curves based on K-means, MultiRocket has the highest accuracy.

[0331] Table 4 Comparison of indicators of different data-driven methods in M2

[0332]

[0333] 2) Verification of the adaptability of the proposed method under increasing load levels

[0334] In actual systems, power systems do not operate in a fixed manner. To verify that the proposed method can be applied to load level growth, on the CEPRI-36 node system, as shown in Table 1, the load motor composition table ratio is set to 10%, 20%, 30%, 40%, 50%, and the flow level is set to 115% and 120% to generate samples as test sets. The model test results are shown below, and the test accuracy is 95.4%, which proves that the proposed method has good robustness under the condition of load level growth.

[0335] Table 5 Comparison of indicators of different data-driven methods in M2

[0336]

[0337] 3) Verification of the adaptability of the proposed method under the growth of system nodes and lines

[0338] In order to verify that the method proposed in the present invention is applicable to the change of power system topology, the samples generated by the 37-node system are used to test the model generated by the 36-node training samples, and the various indicators of the method proposed in the present invention are shown in Table 6. It can be seen that the method proposed in the present invention can directly apply the model trained by the 36-node samples to the test samples generated by the 37-node system while ensuring high test accuracy, and the results verify the effectiveness of the method proposed in the present invention.

[0339] Table 6 Comparison of comprehensive indicators of different data-driven methods based on a 37-node system as a test set

[0340]

[0341] In summary, the example verification shows that the method proposed in the present invention is applicable to the DIM identification problem of power systems with different topology changes.

[0342] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0343] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0344] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0345] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0346] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0347] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for identifying transient DIM in a power system, characterized in that: The following steps are involved: Obtain bus voltage curves and power angle curves of several power systems; Perform typical fault curve screening and weight calculation on the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and power angle instability curve; Extract the multi-dimensional time series characteristics of bus voltage instability curve and power angle instability curve; Based on the multi-dimensional time series characteristics, a basic model is constructed, and the basic model is optimized using bagging ensemble learning and ensemble optimization methods to obtain a DIM recognition model. Use the DIM identification model to complete the transient DIM identification of the power system; The steps of screening typical fault curves and calculating weights for the bus voltage curve and power angle curve of the power system include: Set the initial cluster center of K-means to the two curves with the largest and smallest change amplitudes in the same type of curves, denoted by C max and C min ; The types include power system bus voltage curve and power angle curve; Wherein, matrix A = U, δ; The curves with a change amplitude greater than the preset threshold are classified into one category, and the curves are classified according to the C max The distances are sorted in ascending order; among them, curve A selected With curve C max The distance D i As shown below: D i =distance(A selected ,C max ) (2) In the formula, curve A selected =U selected ,δ selected ; U selected is the bus voltage curve of the power system with a variation greater than the preset threshold; selected A power angle curve having a variation greater than a preset threshold; The weight distribution is calculated based on the distance, and the closer to the curve with the largest change amplitude C max The curve with higher weight is given, that is: Where W selected is the weight; D selected To be close to the curve C with the largest change max The curve of D max , D min The distance D i The maximum and minimum distance values ​​in ; Fill the selected curves according to the weight ratio, and set the selected instability curve matrix as Where T is the duration of the observation window, N selected For the number of instability curves selected, the desired model input characteristic matrix is Where M is the number of fixed input curves required; If N selected <M, then copy the instability curve matrix X to fill M bars, that is: X input =[X X replicate ] T×M (4) Among them, X replicate represents the copied curve matrix; When N selected >M, in order to meet the limit of the number of input features, the instability curve matrix X is deleted according to the distance weight value W between the curves, that is: X input =[x selected1 x selected2 … x selectedM ] T×M (5) Among them, x selectedi Represents the selected instability curve, corresponding to the distance weight value W selected The largest M curves; The steps of extracting the multi-dimensional time series characteristics of the bus voltage instability curve and the power angle instability curve include: The time series X corresponding to the multi-dimensional time series characteristics of bus voltage instability curve and power angle instability curve is l }Converted to first-order difference form, we get: In the formula, X' is the first-order difference sequence; l is the number of time series elements; Use the randomly expanded convolution kernel to transform the time series and obtain the feature vector Z, that is: Where, X i is the time series X of the i-th sample of the input, or the first-order difference series X'; ω is the weight, d is the expansion factor, and b is the bias of the kernel; Calculate the characteristic index of the characteristic vector Z to construct the multidimensional time series feature F, that is: Where, X PPV , X MPV , X MIPV , X LSPV Indicates the corresponding positive value proportion, positive value average, positive value index average, and longest positive value extension when the time series X of the i-th sample is used as input; X' PPV , X' MPV , X' MIPV , X' LSPV Indicates the corresponding positive value proportion, positive value average, positive value index average, and longest positive value extension when the first-order difference sequence X' is used as input; vector F X =[X PPV X MPV X MIPV X LSPV ]; vector F X' =[X' PPV X' MPV X' MIPV X' LSPV ]; The randomly expanded convolution kernel uses a convolution kernel of length N, which contains multiple weight combinations with different values, and the weights are randomly selected when used; Each convolution kernel dilation factor d = [2 0 ],...,[2 max ]; where the exponent max is as follows: In the formula, l input is the length of the input time series, l kernel is the convolution kernel length; The characteristic indicators of the characteristic vector Z include the positive value proportion PPV(Z), the average value of the positive value MPV(Z), the average value of the positive value index MIPV(Z), and the longest positive value extension LSPV(Z); Among them, the positive value PPV (Z) is as follows: Where Z represents the output of the convolution operation; Z i represents the threshold; n is the number of elements of the feature vector Z; The mean value MPV(Z) for positive values ​​is as follows: In the formula, the positive parameter z + The length is m; The average value of the positive index MIPV(Z) is as follows: In the formula, i + represents the index of positive values, m is the number of positive values ​​in Z; The longest positive stretch LSPV(Z) is as follows: In the formula, i and j represent the minimum and maximum index values ​​of positive values, z k is the k-th index array of Z.

2. A method for identifying transient DIM in a power system according to claim 1, characterized in that: The steps of obtaining several power system bus voltage curves and power angle curves include: Setting fault conditions and performing simulation to obtain a bus voltage value matrix U and a generator power angle value matrix δ; the fault conditions include a power flow operation mode, a fault point location, a fault duration, and a fault type; Based on the bus voltage value matrix and generator power angle value matrix under different fault conditions, the bus voltage curve and power angle curve of the power system are established; The bus voltage value matrix U and the generator power angle value matrix δ are shown as follows: Where U T,B is the element in the bus voltage value matrix U; T, B is the number of rows and columns in the bus voltage value matrix U; δ T,M is the element in the generator power angle value matrix δ; T and M are the number of rows and columns in the generator power angle value matrix δ.

3. A method for identifying transient DIM in a power system according to claim 1, characterized in that: The basic model is as follows: In the formula, The probability distribution of the output categories of the basic model; Among them, the probability p of the kth category k As shown below: In the formula, p k is the probability of the kth category; e is the base of the natural logarithm; K is the total number of categories; Linear combination a k As shown below: In the formula, a k is the linear combination of the kth category, w ki is the weight connecting the i-th feature and the k-th category, b k is the bias of the kth category; F i As a feature.

4. A method for identifying transient DIM in a power system according to claim 1, characterized in that: The steps to optimize the base model include: Construct M basic models, and calculate the discrimination result matrix Q of the M basic models for the m DIM samples to be discriminated; Among them, the element Q in the i-th row and j-th column of the discriminant result matrix Q ij As shown below: Let matrix Y = Q T Q, and normalize the elements in matrix Y to obtain: In the formula, m is the number of samples that were identified; Y ii , Y ij , Y jj is the element in matrix Y; is the element after normalization; Construct a DIM recognition model, namely: x i ＝{0,1} (22) In the formula, x i Indicates whether to select the i-th basic model, and s is the optimized integrated model specification; is the objective function of the DIM identification model; is the matrix obtained by normalizing the matrix Y; vector x = [x1,...,x M ].

5. A power system transient DIM identification system based on the method according to any one of claims 1 to 4, characterized in that: include: Data acquisition unit: acquires several bus voltage curves and power angle curves of power systems; Instability curve generation unit: Screening and weighting of typical fault curves for the bus voltage curve and power angle curve of the power system to obtain the bus voltage instability curve and power angle instability curve; Feature extraction unit: extracts multi-dimensional time series features of bus voltage instability curve and power angle instability curve; Model building unit: Based on the multi-dimensional time series characteristics, a basic model is built, and the basic model is optimized using bagging ensemble learning and ensemble optimization methods to obtain a DIM recognition model; Transient DIM identification unit: Use the DIM identification model to complete the transient DIM identification of the power system.

6. An electronic device, characterized in that: include: A memory, a processor and a data bus, wherein the memory stores all machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the data bus, and the machine-readable instructions are executed by the processor to execute the method described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 4 is executed.

Images