A training method for a transient voltage stability assessment model considering uncertainties of power sources and loads

Abstract

The present invention relates to the technical field of power system operation control, and discloses a training method for a transient voltage stability preventive control evaluation model considering source-load uncertainty. The method preliminarily trains a stability evaluation model by selecting a relatively small number of training samples with relatively sparse distribution to obtain an initial stability evaluation model; then inputs the initial stability evaluation model into the preventive control optimization solution; and then, according to the changes in the optimal solutions obtained from the preventive control optimization model, adds training samples in the sparse area of the training samples and in the area near the optimal solutions respectively to avoid missing the optimal solutions and improve the accuracy of the optimal solutions, and retrains the evaluation model with the updated training samples; after several rounds of training, the stability evaluation model is gradually improved. Compared with the existing training methods for stability evaluation models, this method effectively reduces the number of training samples, reduces the model training time-consuming under the premise of ensuring accuracy, and also improves the preventive control optimization efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation control, and in particular to a training method for a transient voltage stability assessment model considering source-load uncertainty. Background Art

[0002] With the intensification of the replacement effect of new energy for conventional energy, the intensive commissioning of high-voltage direct current transmission projects, and the increasing proportion of induction motors in load centers, the problem of transient voltage stability in power systems has become increasingly prominent. Preventive control is the first line of defense for power system safety and stability control. It adjusts measures such as generator active power and voltage before a power system fault occurs, so that the system can still operate stably after the fault occurs. Therefore, it is often described as an optimal power flow problem considering transient stability constraints. When formulating preventive control strategies, transient voltage stability assessment needs to be carried out repeatedly. Common stability assessment methods include time-domain simulation methods and data-driven methods. The former is realized by numerical integration calculation, and the results are intuitive and reliable, but it is very time-consuming. The latter learns the mapping relationship between input features and stability indicators through machine learning algorithms, and has extremely high efficiency in online applications. However, the process of generating training samples and training the model is also time-consuming. With the large-scale grid connection of renewable energy power generation such as wind and light and the increasing popularity of the use of electric vehicles, the random factors in power systems have increased sharply. After considering source-load uncertainty, transient voltage stability preventive control faces greater stability assessment pressure, and there is an urgent need to further improve efficiency.

[0003] The prior art discloses a preventive control method for power system transient stability based on a surrogate model. First, a surrogate model for characterizing the transient stability characteristics of a power system is constructed based on a large amount of simulation data to achieve a fast fitting of the mapping relationship between multi-source input information such as system operating state and disturbance type and transient stability margin. Then, a preventive control model for power system transient stability is constructed based on a genetic algorithm, and the surrogate model is embedded into the preventive control model to replace the complex and time-consuming transient stability constraint calculation, so as to efficiently realize the iterative evolution of the preventive control strategy to obtain the optimal solution. Although this method can effectively improve the optimization and solution efficiency of the transient stability preventive control strategy and meet the requirements of rapid decision-making for power system preventive control, a large number of training samples are required for training the convolutional neural network model, which is time-consuming and occupies a large amount of computing resources. Summary of the Invention

[0004] The primary object of the present invention is to overcome the problems existing in the prior art and provide a preventive control method for transient voltage stability considering source-load uncertainty. The present invention can reduce the samples used for training the model while ensuring the model accuracy, thereby reducing the time and computing resources required for training the model.

[0005] To achieve the above object, the present invention provides a transient voltage stability preventive control method considering source-load uncertainty, and the method includes the following steps:

[0006] S1: Obtain a random variable sample set, where the random variables include active power of the load, reactive power of the load, and active power of wind power;

[0007] S2: Obtain the total sample set of decision variables X h which includes the first decision variable sample set X h0 and the second decision variable sample set X hr where the decision variable is the active power of a conventional generator;

[0008] S3: Based on the random variable sample set and the first decision variable sample set X h0 construct an initial sample set X input ;

[0009] S4: Use power system calculation and analysis software to simulate a power system fault, and calculate the transient voltage stability indexes of each node after the power system fault occurs. The transient voltage stability indexes include the first transient voltage stability index and the second transient voltage stability index T VSI ( X input );

[0010] S5: Based on the initial sample set and the second transient voltage stability index, construct a first training sample;

[0011] S6: Construct a stability evaluation model and a preventive control optimization model, and use the first training sample to train the stability evaluation model to obtain an initial stability evaluation model;

[0012] S7: Use the initial stability evaluation model to solve the preventive control optimization model successively to obtain an optimal solution, and determine whether the number of solution times is greater than 1. If it is greater than 1, calculate the distance between adjacent preventive control optimal solutions, otherwise the first data n c takes and the second data n d takes 0. The first data n c is the number of decision variable samples added to improve the coverage rate of the training sample input space, and the second data n dIncrease the number of decision variable samples to improve the accuracy of the optimal solution, and go to S9;

[0013] S8: Determine whether the distance satisfies the distance threshold condition. If not, calculate the first data n c and the second data n d , otherwise stop the solution;

[0014] S9: Select from the second decision variable sample set n c samples that are farthest from the first decision variable sample set, denoted as the first sample, and delete the first sample from the second decision variable sample set; Select from the second decision variable sample set n d samples that are closest to the optimal solution, denoted as the second sample, and delete the second sample from the second decision variable sample set;

[0015] S10: Perform a cross design on the first sample and the second sample with the random variable sample set to obtain an incremental sample set Δ X input , and calculate the second transient voltage stability index corresponding to the incremental sample set Δ X input (Δ T VSI (Δ X input ));

[0016] S11: Add the incremental sample set Δ X input , the second transient voltage stability index T VSI (Δ X input ) to the first training sample to obtain a second training sample, and go to S6 to continue training the initial stability evaluation model until the distance between adjacent two preventive control optimal solutions satisfies the distance threshold condition, and obtain the final stability evaluation model.

[0017] Furthermore, the method for obtaining the active power of the wind power is as follows:

[0018]

[0019] wherein, is the rated power of the wind power; , and are the cut-in wind speed, rated wind speed and cut-out wind speed respectively.

[0020] Furthermore, S3 specifically includes:

[0021] S3.1: Select samples from the total sample set of decision variables using Latin hypercube sampling as the first decision variable sample set, and denote the remaining samples in the total sample set of decision variables as the second decision variable sample set;

[0022] S3.2: Conduct a cross-design on the random variable sample set and the decision variable sample set to obtain an initial sample set X input .

[0023] Furthermore, in S4, calculate the transient voltage stability index of each node in the power system, where the calculation method of the first transient voltage stability index is as follows:

[0024]

[0025] In the formula, V 0 and V min are the initial value of the node voltage and the lowest value of the node voltage after fault clearing; V th and T th are the safety threshold of the voltage and the tolerable time when the voltage is lower than V th after fault clearing; T max,span is the maximum duration when the node voltage is lower than V th after the fault occurs; is the average value of the voltage in the last 1 s of the observation period;

[0026] Calculate the second transient voltage stability index T VSI ( X input ) as follows:

[0027]

[0028] In the formula, x is X input one of the samples in, X input is the initial sample set; T VSI ( x ) is x the corresponding transient voltage stability level; j is the fault line label; T VIj1 and TVIj2 For x the corresponding power flow mode, when faults occur at the head and end of the faulty line j the average value of the three maximum values among the transient voltage stability indexes of all system nodes in; top3( T VIj ) represents T Vij the three elements with the largest numerical values in; average represents the average operation.

[0029] Furthermore, the preventive control optimization model in S6 is specifically a preventive control optimization model established with the goal of minimizing the expected value of the total generation cost, as follows:

[0030]

[0031] where C is the expected value of the total generation cost; E{·} represents the expectation; P Gi is the active power of the i th conventional generator; a i , b i and c i are the cost coefficients of the i th conventional generator;

[0032] Constraint 1 is the power flow equation constraint:

[0033]

[0034] In the formula, n b is the total number of system nodes; P Gi and Q Gi are the active and reactive powers of the conventional generator; P wi and Q wi are the active and reactive powers of the wind power; P Di and Q Di are the active and reactive powers of the load; V i and V j are the voltage amplitudes of system nodes i and j , θ ij is the phase angle difference between nodes i and j ; Gij and B ij are the real and imaginary parts of the elements of the nodal admittance matrix;

[0035] Constraint 2: Static security inequality constraint:

[0036]

[0037] where P{·} represents probability; V imax and V imin are the upper and lower limits of the magnitude of the system nodal voltage V i respectively; Q Gimax and Q Gimin are the upper and lower limits of the reactive power of the conventional generator Q Gi respectively; S limax is the apparent power of the i th branch of the system S li upper limit; n l is the total number of branches in the system; b V , b Q and b S are thresholds, and they are both set to 95%;

[0038] Constraint 3: Transient security inequality constraint:

[0039]

[0040] where m R and s R are the expectation and standard deviation of the transient voltage stability level respectively; m tv and s tv are thresholds, and m tv is set to 2.8, s tv is set to 0.6.

[0041] Furthermore, in S7, the distance between two adjacent optimal solutions of preventive control is calculated as follows:

[0042]

[0043] wheren G is the number of traditional generators in the system except for the balancing machine, that is, the dimension of the decision variable; and are the active powers of the k th and the ( k -1)th traditional generators obtained from the preventive control optimization solution for the i th traditional generator; P Gin is the rated active power of the i th traditional generator.

[0044] Furthermore, the threshold condition described in S8 is d 1 ≤ 10 -4 , d where 1 is the distance between the optimal solutions of two adjacent generations of preventive control.

[0045] Furthermore, the method for calculating the first data n c and the second data n d in S8 is as follows:

[0046]

[0047] .

[0048] Furthermore, S9 specifically includes:

[0049] S9.1: Calculate the distance between each sample in the second decision variable sample set and the first decision variable sample set, sort all the distances from largest to smallest, and mark the first n c samples as the first samples, and delete the first samples from the second decision variable sample set;

[0050] S9.2: Calculate the distance between each sample in the second decision variable sample set and the optimal solution; sort all the distances from smallest to largest, and mark the first n d samples as the second samples, and delete the second samples from the second decision variable sample set.

[0051] Furthermore, it is characterized in that the method for calculating the distance in S9.1 is as follows:

[0052]

[0053] In the formula, x q is X h0 a sample in xhrp is X hr a sample in; denotes x hrp the distance between X h0 and; denotes x hrp the distance between x q and; x hrpi and x qi respectively denote x hrp and x q the active power of the i th traditional generator in;

[0054] S9.2 calculates the distance as follows:

[0055]

[0056] wherein, denotes x hrj the distance between x opt and; x hrji and x opti respectively denote x hrj and x opt the active power of the i th traditional generator in.

[0057] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0058] The present invention initially trains a stability evaluation model by selecting a relatively small number of training samples with a relatively sparse distribution to obtain an initial stability evaluation model; then inputs the initial stability evaluation model into the preventive control optimization solution; and then, according to the change of the optimal solution obtained by the preventive control optimization model, adds training samples in the sparse area of the training samples and the area near the optimal solution respectively to avoid missing the optimal solution and improving the accuracy of the optimal solution, and retrains the evaluation model with the updated training samples; after several rounds of training, the stability evaluation model is gradually improved. Compared with the existing stability evaluation model training method, this method effectively reduces the number of training samples, reduces the model training time-consuming, and improves the preventive control optimization efficiency on the premise of ensuring the accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flowchart of a training method for a transient voltage stability assessment model considering source-load uncertainty in Embodiment 1 of the present invention;

[0060] Figure 2 It is in Embodiment 2 of the present invention d Variation curve of 1;

[0061] Figure 3 It is in Embodiment 2 of the present invention n c and n d Variation curve. Specific embodiments

[0062] The following combines the drawings and embodiments to further describe in detail the specific embodiments of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0063] Embodiment 1

[0064] As Figure 1 shown, a transient voltage stability preventive control method considering source-load uncertainty in a preferred embodiment of an embodiment of the present invention includes the following steps:

[0065] S1: Obtain a random variable sample set, where the random variables include load active power, load reactive power, and active power of wind power;

[0066] In a feasible embodiment, it is assumed that the load power follows a normal distribution with known parameters. First, use the Sobol sequence to obtain samples in the [0, 1] interval, and then obtain the load active power and load reactive power that follow the normal distribution through the cumulative distribution inverse transformation;

[0067] It is assumed that the wind speed follows a Weibull distribution with known parameters, and the correlation between the wind speeds of different wind farms is described by a correlation coefficient matrix with known parameters. Generate wind speed samples for each wind farm considering the correlation, and then obtain the active power of wind power according to Equation (1).

[0068] (1)

[0069] Among them, is the rated power of wind power; , and are the cut-in wind speed, rated wind speed, and cut-out wind speed respectively.

[0070] S2: Obtain the total decision variable sample set X h , the total decision variable sample set contains the first decision variable sample set X h0and the second decision variable sample set X hr , where the decision variable is the active power of the traditional generator;

[0071] S3: Based on the random variable sample set and the first decision variable sample set X h0 construct an initial sample set X input ;

[0072] In a feasible embodiment, assume that the active power of each traditional generator randomly varies between the rated values and obtain the total decision variable sample set X h through Latin hypercube sampling. Then, S3 specifically includes:

[0073] S3.1: Use the Latin hyper-sampling method to select X h samples from the total decision variable sample set as the first decision variable sample set X h0 . In this embodiment, select samples from the total decision variable sample set as the first decision variable sample set X h0 . The remaining samples in the total decision variable sample set are denoted as the second decision variable sample set X hr . In this embodiment, the first decision variable sample set is also called the selected decision variable sample set X h0 ; The second decision variable sample set is also called the remaining decision variable sample set and is denoted as X hr .

[0074] S3.2: Conduct a cross-design on the random variable sample set and the decision variable sample set to obtain the initial sample set X input .

[0075] S4: Use power system calculation and analysis software to simulate the occurrence of a fault in the power system and calculate the transient voltage stability indices of each node after the fault occurs in the power system. The transient voltage stability indices include the first transient voltage stability index and the second transient voltage stability index T VSI ( X input );

[0076] In a feasible embodiment, the fault of the power system is a three-phase short circuit at both ends of all AC lines. Time-domain simulation is performed using the power system calculation and analysis software BPA, and the first transient voltage stability index of each node in the system after the fault occurs is calculated according to Equation (2). :

[0077] (2)

[0078] In the formula, V 0 and V min are the initial value of the node voltage and the lowest value of the node voltage after fault clearing; V th and T th are the safety threshold of the voltage and the tolerable time when the voltage is lower than V th after fault clearing, respectively; T max,span and are the maximum duration when the node voltage is lower than V th after the fault occurs, and the average value of the voltage in the last 1 s of the observation period, respectively;

[0079] According to Equation (2), the value range of the first transient voltage stability index is . means that the voltage remains above the threshold V th after fault clearing; means that the voltage drops to V th or below, but the duration is less than the threshold ; means that the duration when the voltage is lower than exceeds , but it can recover to or above; means that the voltage drops to or below and cannot recover to or above. The higher the value of the first transient voltage stability index , the worse the transient voltage performance of the node.

[0080] Calculate the second transient voltage stability index T VSI ( X input ) in the manner of Equation (3):

[0081] (3)

[0082] In the formula,x is X input a sample in X input the initial sample set; T VSI ( x ) is x the corresponding transient voltage stability level; j the fault line label; T VIj1 and T VIj2 are x under the corresponding power flow mode, when faults occur at the head and end of the fault line j the average of the three maximum values among all the transient voltage stability indexes of the system nodes; top3( ) represents T VIj ) indicates T Vij the three elements with the largest values in average

[0083] It can be seen that T VSI ( x ) has a value range of [0, 4). T VSI The higher it is, the lower the transient voltage stability level of the system under this power flow mode.

[0084] S5: Construct the first training sample based on the initial sample set and the second transient voltage stability index;

[0085] In a feasible embodiment, based on the initial sample set X input and the second transient voltage stability index T VSI ( X input ) construct the first training sample X 0.

[0086] S6: Construct a stability evaluation model and a preventive control optimization model, and use the first training sample to train the stability evaluation model to obtain an initial stability evaluation model;

[0087] In a feasible embodiment, the preventive control optimization model in S6 is specifically a preventive control optimization model established with the goal of minimizing the expected total power generation cost, as follows:

[0088]

[0089] Wherein, Cis the expected value of the total power generation cost; E{·} represents the expectation; P Gi is the active power of the i th conventional generator; a i , b i and c i are the cost coefficients of the i th conventional generator;

[0090] Constraint 1 is the power flow equation constraint:

[0091]

[0092] wherein, n b is the total number of system nodes; P Gi and Q Gi are the active and reactive powers of the conventional generator; P wi and Q wi are the active and reactive powers of the wind power; P Di and Q Di are the active and reactive powers of the load; V i and V j are the i and j voltage amplitudes of system nodes, θ ij is the phase angle difference between node i and j ; G ij and B ij are the real and imaginary parts of the elements of the node admittance matrix;

[0093] Constraint 2 is the static security inequality constraint:

[0094]

[0095] wherein, P{·} represents the probability; V imax and V imin are respectively the upper and lower limits of the system node voltage amplitude V i ; Q Gimax and Q Gimin are respectively the reactive power of the conventional generator QGi The upper and lower limits; S limax is the i apparent power of the S li upper limit of the n l is the total number of system branches; b V , b Q and b S are thresholds, and both are set to 95%;

[0096] Constraint 3 is the transient security inequality constraint:

[0097]

[0098] In the formula, m R and s R are the expectation and standard deviation of the transient voltage stability level respectively; m tv and s tv are thresholds, and set m tv = 2.8, s tv = 0.6.

[0099] S7: Use the initial stability evaluation model to solve the preventive control optimization model successively to obtain the optimal solution, and judge whether the number of solutions is greater than 1. If it is greater than 1, calculate the distance between the adjacent two preventive control optimal solutions, and let n c = 5, n d = 0. The first data n c is the number of decision variable samples added to improve the coverage rate of the training sample input space, and the second data n d is the number of decision variable samples added to improve the accuracy of the optimal solution, and go to S9;

[0100] In a feasible embodiment, the distance between the adjacent two preventive control optimal solutions in S7 is calculated as follows:

[0101]

[0102] In the formula, n G is the number of traditional generators in the system except the swing machine, that is, the dimension of the decision variable; and For the active power of the k th and the k ( i - 1)th traditional generators obtained from the preventive control optimization solution; P Gin For the i th traditional generator's rated active power.

[0103] S8: Determine whether the distance satisfies the distance threshold condition. If not, calculate the first data n c and the second data n d , otherwise stop the solution;

[0104] In a feasible embodiment, the threshold condition in S8 is d 1 ≤ 10 -4 , d where 1 is the distance between the preventive control optimal solutions of two adjacent generations.

[0105] Furthermore, the method for calculating the first data n c and the second data n d in S8 is as follows:

[0106]

[0107] .

[0108] S9: Screen out n c samples in the second decision variable sample set that are farthest from the first decision variable sample set, denoted as the first sample, and delete the first sample from the second decision variable sample set; Screen out n d samples in the second decision variable sample set that are closest to the optimal solution, denoted as the second sample, and delete the second sample from the second decision variable sample set;

[0109] In a feasible embodiment, S9 specifically includes:

[0110] S9.1: Calculate the distance between each sample in the second decision variable sample set and the first decision variable sample set. The specific method for calculating the distance is as follows:

[0111]

[0112] In the formula, x q is Xh0 a sample in x hrp is X hr a sample in denotes x hrp the distance between X h0 ; denotes x hrp the distance between x q ; x hrpi and x qi respectively denote x hrp and x q the active power of the i th traditional generator in

[0113] Furthermore, sort all the distances from largest to smallest, and denote the first n c samples as the first samples, and delete the first samples from the second decision variable sample set; specifically, select from X hr the sample that is farthest from X h0 x far . Add it to X h0 , and at the same time remove X hr from x far , as follows:

[0114]

[0115] where arg max denotes that when achieves the maximum value, the value of the variable .

[0116] S9.2: Calculate the distance between each sample in the second decision variable sample set and the optimal solution, and sort all the distances from smallest to largest. Denote the first n d samples as the second samples, and delete the second samples from the second decision variable sample set; the specific calculation method is as follows:

[0117]

[0118] where​ Indicate x hrj and x opt distance; arg Indicate when reaches the minimum value, the variable value; x hrji and x opti respectively indicate x hrj and x opt in the i active power of the traditional generator.

[0119] S10: Cross - design the first sample and the second sample with the random variable sample set to obtain an incremental sample set Δ X input , and calculate the corresponding second transient voltage stability index X input (Δ T VSI ) of the incremental sample set Δ X input ;

[0120] In a feasible embodiment, the calculation method of the corresponding second transient voltage stability index X input (Δ T VSI ) of the incremental sample set Δ X input is as shown in equation (3) in S4, which will not be elaborated here.

[0121] S11: Add the incremental sample set Δ X input , the second transient voltage stability index T VSI (Δ X input ) to the first training sample to obtain a second training sample, and go to S6 to continue training the initial stability evaluation model until the distance between adjacent two optimal solutions of preventive control meets the distance threshold condition to obtain the final stability evaluation model.

[0122] Embodiment 2

[0123] In this embodiment, to verify the superiority of the training method of a transient voltage stability assessment model considering source-load uncertainty proposed in Embodiment 1, an improved New England 10-machine 39-node system is used for testing. The original system contains 10 generators, 39 nodes, and 34 lines. The synchronous generators connected to nodes 30 and 37 in the original system are replaced by wind turbine generators. The wind power adopts a constant power factor control method, and the rated active power is set to 250 MW and 540 MW; the cut-in wind speed is 3 m / s, the rated wind speed is 10 m / s, and the cut-out wind speed is 18 m / s; the scale parameter is 6.1 and the shape parameter is 2.

[0124] In this embodiment, let the scale of the random variable sample set be 80, and generate the random variable sample set according to step S1 X R ;

[0125] Let the sampling scale of the decision variable be 100. Generate the decision variable sample set, the selected sample set of the decision variable X h and the remaining sample set of the decision variable X h0 according to step S2 X hr .

[0126] Obtain X input according to step S3

[0127] In step S4, the generator adopts a detailed model, and the load adopts a comprehensive model, with constant impedance and induction motor loads each accounting for 50%. Set a three-phase short circuit at both ends of the AC line at 1 s, and both sides of the protection trip simultaneously to remove the fault 0.12 s after the fault. The total simulation duration is 11 s. The voltage safety threshold V th and the tolerable time when the voltage is lower than V th after the fault occurs T th are taken as 0.8 pu and 1 s respectively, and T VSI ( X input ) is calculated.

[0128] Obtain X 0 according to step S5

[0129] Train the stability assessment model according to step S7. Among them, the radial basis function is selected as the kernel function, the regularization parameter C is set to 11.55, and the scale parameter e of the kernel function is set to 3.

[0130] Execute steps 7-11. During the preventive control optimization solution process, d The change curve ofFigure 2 As shown n c and n d The changes of are shown in Figure 3 As shown by Figure 2 It can be seen that in this embodiment, the solution is obtained a total of 7 times. At the initial stage of the solution, the difference between the optimal solutions obtained by the preventive control optimization model in two adjacent calculations is relatively significant ( d 1 is large), and at this time, decision variable samples for improving the coverage rate of the input space are mainly added n c . As the distance d 1 between two adjacent optimal solutions decreases, the coverage rate of the training samples within their value ranges becomes higher and higher. At this time, the number of decision variable samples for improving the accuracy of the optimal solution is mainly added n d . During the entire preventive control optimization solution process, this embodiment generated (20 + 6×5)×80 = 4000 training samples, where 20 is the number of initial decision variable samples, and 6×5 = 30 is the number of decision variable samples added subsequently. The conventional method generates 100×80 = 8000 training samples at one time, where 100 is the total number of decision variable samples and 80 is the number of random variable samples.

[0131] To evaluate the accuracy of the stability evaluation model established for the invention, the mean absolute error (mean absolute error, M AE ) and the mean square error (Mean Square Error, M SE ) are used. M AE The calculation formulas of and M SE are shown in Equation (4). The smaller the two values are, the higher the prediction accuracy.

[0132] (4)

[0133] In the formula, N T is the number of evaluation samples; and are respectively the actual value of the transient voltage stability level obtained by time-domain simulation of the sample t and the predicted value given by the stability evaluation model.

[0134] The evaluation accuracies of the following 3 cases are compared: 1) Training the stability evaluation model with the above 8000 training samples; 2) Selecting 50 from 100 total decision variable samples using Latin hypercube sampling, and X RCross - design to form 4000 training samples for training the stability assessment; 3) The method described in the present invention.

[0135] The prediction accuracy of the stability assessment model near the optimal solution of the preventive control is crucial for the optimal solution of the preventive control optimization. Therefore, the optimal solution obtained by solving the preventive control model is compared with X R Cross - design to obtain 80 power flow samples, and calculate the transient voltage stability levels corresponding to these 80 power flow samples according to step 6. Then combine these 80 power flow samples with the corresponding transient voltage stability levels to obtain 80 test samples. The results comparison is shown in Table 1:

[0136] Table 1 Prediction accuracy of the stability assessment models obtained by two training methods at the optimal solution of the preventive control

[0137]

[0138] As can be seen from Table 1, although the prediction accuracy of the assessment model obtained in this embodiment near the optimal solution of the preventive control model is slightly lower than the conventional method, it is higher than the model obtained by the conventional method using Latin hypercube sampling to select 50 samples. The conventional method uses 8000 training samples, and the latter two methods both use only 4000 training samples. This embodiment uses fewer training samples to generate a stability assessment model with relatively ideal accuracy.

[0139] Table 2 Comparison of preventive control schemes obtained by two methods

[0140]

[0141] Table 2 compares the preventive control results obtained by applying the conventional method (8000 training samples) and the method described in the embodiment of the present invention using the stability assessment model. It can be seen that the two methods obtain basically the same preventive control scheme, but the time consumption of the method described in the present invention is about half of the conventional method. This shows that using the method described in the present invention greatly shortens the time required to establish the stability assessment model and greatly improves the preventive control solution efficiency without reducing the preventive control solution accuracy.

[0142] In summary, the embodiment of the present invention provides a training method for a transient voltage stability evaluation model considering source-load uncertainty. The method preliminarily trains the stability evaluation model by selecting a relatively small number of training samples with sparse distribution to obtain an initial stability evaluation model. Then, the initial stability evaluation model is put into the preventive control optimization solution. Next, according to the changes in the optimal solutions obtained from the preventive control optimization model, training samples are added in the sparse region of the training samples and the region near the optimal solutions respectively to avoid missing the optimal solutions and improving the accuracy of the optimal solutions, and the evaluation model is retrained with the updated training samples. After several rounds of training, the stability evaluation model is gradually improved. Compared with the existing training methods for stability evaluation models, this method effectively reduces the number of training samples, reduces the model training time, and improves the preventive control optimization efficiency on the premise of ensuring accuracy.

[0143] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and replacements can be made, and these improvements and replacements should also be regarded as the protection scope of the present invention.

Claims

1. A training method for a transient voltage stability assessment model considering source-load uncertainty, characterized in that The training method includes the following steps: S1: Obtain a set of random variable samples, where the random variables include active load power, reactive load power, and active power of wind power; S2: Obtain the total sample set of decision variables X h , the total sample set of decision variables includes the first decision variable sample set X h0 and the second decision variable sample set X hr , where the decision variable is the active power of a conventional generator; S3: Based on the random variable sample set and the first decision variable sample set X h0 Construct an initial sample set X input ; S4: Use power system calculation and analysis software to simulate a power system fault, and calculate the transient voltage stability indices of each node after the power system fault occurs. The transient voltage stability indices include a first transient voltage stability index and a second transient voltage stability index T VSI ( X input ); S5: Construct a first training sample based on the initial sample set and the second transient voltage stability index; S6: Construct a stability evaluation model and a preventive control optimization model, and use the first training sample to train the stability evaluation model to obtain an initial stability evaluation model; S7: Successively solve the prevention and control optimization model using the initial stability evaluation model to obtain the optimal solution. Determine whether the number of solution times is greater than 1. If it is greater than 1, calculate the distance between two adjacent optimal solutions of prevention and control. Otherwise, the first data n c Take , the second data n d Take 0, the first data n c is the number of decision variable samples added to increase the coverage rate of the training sample input space. The second data n d is the number of decision variable samples added to improve the accuracy of the optimal solution, and go to S9; S8: Determine whether the distance meets the distance threshold condition. If not, calculate the first data n c and the second data n d , otherwise stop the solution; S9: Select from the second decision variable sample set n c samples that are farthest from the first decision variable sample set, denoted as the first sample, and delete the first sample from the second decision variable sample set; Select from the second decision variable sample set n d samples that are closest to the optimal solution, denoted as the second sample, and delete the second sample from the second decision variable sample set; S10: Cross-design the first sample and the second sample with the random variable sample set to obtain an incremental sample set Δ X input , and calculate the corresponding second transient voltage stability index of the incremental sample set Δ X input (Δ T VSI ); X input ) S11: Incorporate the incremental sample set Δ X input , the second transient voltage stability index T VSI (Δ X input ) into the first training sample to obtain a second training sample, and go to S6 to continue training the initial stability assessment model until the distance between the optimal solutions of two adjacent preventive controls meets the distance threshold condition, and obtain the final stability assessment model.

2. The training method of a transient voltage stability assessment model considering source-load uncertainty according to claim 1, characterized in that The method for obtaining the active power of the wind power is as follows: Among them, is the rated power of the wind power; , and are the cut-in wind speed, rated wind speed and cut-out wind speed respectively.

3. A training method for a transient voltage stability assessment model considering source-load uncertainty according to claim 1, characterized in that S3 specifically includes: S3.1: Select samples of from the total sample set of decision variables using the Latin hypercube sampling method as the first decision variable sample set, and mark the remaining samples in the total sample set of decision variables as the second decision variable sample set; ; Denote the remaining samples in the total sample set of decision variables as the second decision variable sample set. S3.2: Cross-design the random variable sample set and the decision variable sample set to obtain an initial sample set X input .

4. The training method of a transient voltage stability assessment model considering source-load uncertainty according to claim 1, characterized in that, Calculate the transient voltage stability indices of each node in the power system in S4, where the first transient voltage stability index is calculated in the following manner: Wherein, V 0 and V min are the initial value of the node voltage and the lowest value of the node voltage after fault clearing; V th and T th are the safety threshold of the voltage and the tolerable time when the voltage is lower than V th after fault clearing; T max,span is the maximum duration when the node voltage is lower than V th after the fault occurs; is the average value of the voltage in the last 1 s of the observation period; Calculate the second transient voltage stability index T VSI ( X input ) is as follows: Wherein, x is X input one of the samples in, X input is the initial sample set; T VSI ( x ) is x the corresponding transient voltage stability level; j is the fault line label; T VIj1 and T VIj2 are x Under the corresponding power flow mode, when a fault occurs at the head and end of the fault line j The average value of the three maximum values of the transient voltage stability indexes of all nodes in the system; top3( T VIj ) represents T Vij The three elements with the largest values in; average represents the average operation.​ 5. The training method of a transient voltage stability assessment model considering source-load uncertainty according to claim 1, characterized in that The preventive control optimization model in S6 is specifically a preventive control optimization model established with the goal of minimizing the expected total power generation cost, as follows: Among them, C is the expected value of the total power generation cost; E{·} represents the expectation; P Gi is the active power of the i th traditional generator; a i , b i and c i are the cost coefficients of the i th traditional generator; Constraint condition 1 is the power flow equation constraint: Wherein, n b is the total number of system nodes; P Gi and Q Gi are the active and reactive powers of the traditional generator; P wi and Q wi are the active and reactive powers of the wind power; P Di and Q Di are the active and reactive powers of the load; V i and V j are the voltage amplitudes of system nodes i and j ; θ ij is the phase angle difference between nodes i and j ; G ij and B ij are the real and imaginary parts of the elements of the nodal admittance matrix; Constraint condition 2 is the static security inequality constraint: Wherein, P{·} represents probability; V imax and V imin are respectively the upper and lower limits of the system node voltage amplitude V i ; Q Gimax and Q Gimin are respectively the upper and lower limits of the reactive power Q Gi of the conventional generator; S limax is the i th branch apparent power S li of the system; n l is the total number of system branches; b V , b Q and b S are thresholds, and all of them are set to 95%; Constraint condition 3 is the transient security inequality constraint: Wherein, m R and s R are the expectation and standard deviation of the transient voltage stability level respectively; m tv and s tv are thresholds, let m tv = 2.8, s tv = 0.

6.

6. The training method of a transient voltage stability evaluation model considering source-load uncertainty according to claim 1, characterized in that The method for calculating the distance between two adjacent preventive control optimal solutions in S7 is as follows: In the formula, n G is the number of traditional generators in the system except the balancing machine, that is, the dimension of the decision variable; and are the active powers of the k -th and the ( k -1)-th traditional generators obtained by preventive control optimization solution; i is the active power of the P Gin -th traditional generator; i is the rated active power of the i -th traditional generator.

7. A training method for a transient voltage stability assessment model considering source-load uncertainty according to claim 1, characterized in that The threshold condition described in S8 is d 1 ≤ 10 -4 , d where 1 is the distance between the optimal solutions of two adjacent generations of preventive control.

8. A training method for a transient voltage stability assessment model considering source-load uncertainty according to claim 7, characterized in that Calculate the first data in S8 n c and the second data n d in the following way: 。 9. A training method for a transient voltage stability assessment model considering source-load uncertainty according to claim 1, characterized in that S9 specifically includes: S9.1: Calculate the distance between each sample in the second decision variable sample set and the first decision variable sample set, sort all the distances from largest to smallest, and denote the first n c samples as the first samples, and delete the first samples from the second decision variable sample set; S9.2: Calculate the distance between each sample in the second decision variable sample set and the optimal solution; sort all the distances in ascending order, mark the first n d samples as the second samples, and delete the second samples from the second decision variable sample set.​​​ 10. A training method for a transient voltage stability assessment model considering source-load uncertainty according to any one of claims 1 to 9, characterized in that, The method for calculating the distance in S9.1 is as follows: In the formula, x q is X h0 a sample in x hrp is X hr a sample in represents x hrp the distance between X h0 and represents x hrp the distance between x q and x hrpi and x qi respectively represent x hrp and x q the active power of the i th traditional generator in The method for calculating the distance in S9.2 is as follows: wherein, represents x hrj the distance from x opt ; x hrji and x opti respectively represent x hrj and x opt the active power of the i th conventional generator in

Images