Fast kernel learning method and system for evaluating the lowest point of power system frequency response

Abstract

The present invention discloses a fast kernel learning method and system for evaluating the lowest point of the frequency response of a power system. The present invention combines a simplified model with artificial intelligence, inputs the second-order motion equation of the generator, and the distribution position information of the power system stabilizer PSS, and outputs the lowest point of the system frequency after the load power disturbance according to the topological structure and parameters of the power system, so as to judge whether the power system meets the frequency safety constraint, thereby avoiding the convergence problem in the process of deep learning parameter optimization.

Description

Technical Field

[0001] The invention belongs to the field of power system frequency response characteristic evaluation, in particular to a fast kernel learning method and system for evaluating the lowest point of power system frequency response. Background Art

[0002] Low-inertia new energy power generation equipment replaces synchronous machines as the main power source of the power system, causing the frequency response characteristics of the new power system to change in a direction that is unfavorable to the system after power disturbances, affecting the role of frequency protection actions, and in severe cases, causing cascading failures that threaten the safe and stable operation of the power grid. In view of the above situation, accurately evaluating the characteristics of the frequency response trajectory of the new power system after power disturbances is of great importance to the safety and stability of the system.

[0003] After the power system is subjected to power disturbance, the maximum deviation of the dynamic frequency from the rated power frequency and the rate of change of the points on the dynamic frequency trajectory are the focus of attention of the system dispatching and control departments, because the above two frequency characteristics have the greatest impact on the frequency protection action. In addition, the "Guidelines for Safety and Stability of Power Systems" also make corresponding provisions for this, that is, to pay attention to the balance of active power during the power outage recovery process to prevent large fluctuations in system frequency.

[0004] The above backgrounds require in-depth analysis and quantitative research on frequency dynamic characteristics. Summary of the invention

[0005] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a fast kernel learning method for evaluating the lowest point of the frequency response of the power system, which combines a simplified model with artificial intelligence and directly obtains training results from limited but independent input data to avoid convergence problems in the deep learning parameter optimization process.

[0006] A technical solution adopted by the present invention is: a fast kernel learning method for evaluating the lowest point of the frequency response of the power system, which comprises the steps of:

[0007] 1) The frequency response dynamics of the generator is approximated by the second-order motion equation, the common-mode component of the frequency response trajectory of the multi-machine power system is used as the source of the lowest point of the frequency response, and an analytical expression for estimating the lowest point of the frequency response by the second-order motion equation of the multi-machine is established;

[0008] 2) taking the total number, mean, kurtosis and skewness of the power system stabilizer PSS as input, and using the trained fast kernel learning algorithm, outputting the frequency response minimum point estimation error caused by not considering the power system stabilizer PSS in step 1);

[0009] 3) The analytical expression of the lowest frequency response point estimated in step 1) is superimposed with the estimated error of the lowest frequency response point obtained in step 2) to obtain the actual lowest frequency response point of the power system, and compare it with the safety threshold. If it is greater than the safety threshold, the output is 1, otherwise the output is 0, thereby realizing the frequency safety assessment of the power system.

[0010] The present invention combines a simplified model with artificial intelligence, inputs the second-order motion equation of the generator and the distribution position information of the power system stabilizer PSS, and can output the lowest point of the system frequency after the load power disturbance according to the topological structure and parameters of the power system, so as to judge whether the power system meets the frequency safety constraints.

[0011] Furthermore, in step 1), the common mode component in the frequency response is used to simulate the frequency trajectory of the power system. When a step disturbance occurs in the load, the common mode frequency response of the power system is:

[0012]

[0013] Where Δω(s) represents the common mode frequency of the power system; s is the Laplace operator; ΔP L (s) represents the frequency domain function of the total load disturbance of the power system. The single-input and single-output transfer functions of the frequency-active power of the power generation equipment in the power system are represented by G i (s),i=1,…,n description, G i (s) Second order model representing only inertia, damping and governor effects:

[0014] G i (s) = J i s+D i +s / K i ,

[0015] Among them, the parameter J i , D i and K i G i (s) Inertia, damping and governor action parameters corresponding to the second-order model;

[0016] The common mode frequency response of the power system is expressed as a unified structural model:

[0017]

[0018] Among them, the parameter J u , D u and K u They are the sum of the corresponding parameters of the second-order model of the power generation equipment in the system.

[0019] Furthermore, in the step 1), the frequency minimum point analytical expression estimated by the multi-machine second-order motion equation is represented by the drop depth coefficient α:

[0020]

[0021] Furthermore, in step 2), the fast kernel learning algorithm is divided into two parts. The first part is to determine the learning model after the polynomial degree s is given. The specific steps are as follows:

[0022] a. Input test set data and the polynomial degree s∈N, N represents a set of non-negative integers, and m is the number of samples;

[0023] b. Order Represents the number of center points selected from the test set and a set of independent input data represents the center point of the input data space, C is the number of combinations, and d represents the vector x i The dimension of

[0024] c. Based on the center point Calculate the semi-positive definite kernel matrix K represents the kernel matrix, and the polynomial kernel function is expressed as:

[0025]

[0026] κ represents the kernel function. The function in the above formula can be used as a kernel function only when it satisfies the following semi-positive definite condition of the kernel matrix K:

[0027]

[0028] d. Let the output vector of the test set be y = (y1,…,y m ) T The undetermined parameters in the training model form a vector c = (c1,…,c n ) T , then c=pinv(A m,n )y, where pinv represents the generalized inverse operation;

[0029] e. Output training results The algorithm terminates, where x represents a column vector consisting of input data;

[0030] The specific steps of the second part are as follows:

[0031] a. Test set data Divide into two, and obtain sub-dataset z1 with cardinality m1 and sub-dataset z2 with cardinality m2;

[0032] b. z1 and Substitute them into the first part of the fast kernel learning algorithm for calculation to obtain a set of training model sequences

[0033] c. Use the results in z2 and b and the following formula to get the final polynomial degree s′ and the corresponding training model:

[0034]

[0035] Among them, Ξ represents no more than The set of non-negative integers, represents the floor function, sgn represents the sign function, Represents the input x in the training set i and output y i ∈[-M,M], where M represents the real number boundary.

[0036] Furthermore, in step 3), when the drop depth coefficient α and the disturbance total power time domain function ΔP L When the product of (t) is greater than 0.9 Hz, it is considered that the frequency safety constraint is not met.

[0037] Another technical solution adopted by the present invention is: a fast kernel learning system for evaluating the lowest point of the frequency response of the power system, which comprises:

[0038] Frequency response lowest point analytical formula establishment unit: approximate the frequency response dynamics of the generator through the second-order motion equation, take the common mode component of the frequency response trajectory of the multi-machine power system as the source of the frequency response lowest point, and establish an analytical formula for estimating the frequency response lowest point by the multi-machine second-order motion equation;

[0039] Fast kernel learning unit: takes the total number, mean, kurtosis and skewness of the power system stabilizer PSS as input, uses the trained fast kernel learning algorithm, and outputs the frequency response minimum point estimation error caused by not considering the power system stabilizer PSS;

[0040] Power system frequency security assessment unit: superimpose the frequency response minimum point analytical expression obtained by the frequency response minimum point analytical expression establishment unit and the frequency response minimum point estimated error obtained by the estimated error output unit to obtain the actual power system frequency response minimum point, and compare it with the safety threshold. If it is greater than the safety threshold, the output is 1, otherwise the output is 0, thereby realizing the power system frequency security assessment.

[0041] The present invention has the following beneficial effects:

[0042] The present invention combines a simplified model with artificial intelligence to obtain training results directly from limited but independent input data, avoiding the convergence problem in the deep learning parameter optimization process. The present invention does not rely on the amount of training data, is suitable for rapid learning of actual operation data of large power grids, and can further improve the effectiveness of frequency characteristic evaluation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the process of the present invention;

[0044] Figure 2 It is a structural block diagram of the system of the present invention;

[0045] Figure 3 The figure is a schematic diagram of the process of the present invention when the method is applied. DETAILED DESCRIPTION

[0046] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0047] Example 1

[0048] This embodiment is a fast kernel learning method for evaluating the lowest point of the power system frequency response. Figure 1 As shown, it includes the steps of:

[0049] 1) The frequency response dynamics of the generator is approximated by the second-order motion equation, the common-mode component of the frequency response trajectory of the multi-machine power system is used as the source of the lowest point of the frequency response, and an analytical expression for estimating the lowest point of the frequency response by the second-order motion equation of the multi-machine is established;

[0050] 2) taking the total number, mean, kurtosis and skewness of the power system stabilizer PSS as input, and using the trained fast kernel learning algorithm, outputting the frequency response minimum point estimation error caused by not considering the power system stabilizer PSS in step 1);

[0051] 3) The analytical expression of the lowest frequency response point estimated in step 1) is superimposed with the estimated error of the lowest frequency response point obtained in step 2) to obtain the actual lowest frequency response point of the power system, and compare it with the safety threshold. If it is greater than the safety threshold, the output is 1, otherwise the output is 0, thereby realizing the frequency safety assessment of the power system.

[0052] Specifically, in step 1), the common mode component in the frequency response is used to simulate the frequency trajectory of the power system. When a step disturbance occurs in the load, the common mode frequency response of the power system is:

[0053]

[0054] Where Δω(s) represents the common mode frequency of the power system; s is the Laplace operator; ΔPL (s) represents the frequency domain function of the total load disturbance of the power system. The single-input and single-output transfer functions of the frequency-active power of the power generation equipment in the power system are represented by G i (s),i=1,…,n description, G i (s) Second order model representing only inertia, damping and governor effects:

[0055] G i (s) = J i s+D i +s / K i ,

[0056] Among them, the parameter J i , D i and K i G i (s) Inertia, damping and governor action parameters corresponding to the second-order model;

[0057] The common mode frequency response of the power system is expressed as a unified structural model:

[0058]

[0059] Among them, the parameter J u , D u and K u They are the sum of the corresponding parameters of the second-order model of the power generation equipment in the system.

[0060] Specifically, in the step 1), the frequency minimum point analytical expression estimated by the multi-machine second-order motion equation is represented by the drop depth coefficient α:

[0061]

[0062] Specifically, in step 2), the fast kernel learning algorithm is divided into two parts. The first part is to determine the learning model after the polynomial degree s is given. The specific steps are as follows:

[0063] a. Input test set data and the polynomial degree s∈N, N represents a set of non-negative integers, and m is the number of samples;

[0064] b. Order Represents the number of center points selected from the test set and a set of independent input data represents the center point of the input data space, C is the number of combinations, and d represents the vector x i The dimension of

[0065] c. Based on the center point Calculate the semi-positive definite kernel matrix K represents the kernel matrix, and the polynomial kernel function is expressed as:

[0066]

[0067] κ represents the kernel function. The function in the above formula can be used as a kernel function only when it satisfies the following semi-positive definite condition of the kernel matrix K:

[0068]

[0069] d. Let the output vector of the test set be y = (y1,…,y m ) T The undetermined parameters in the training model form a vector c = (c1,…,c n ) T , then c=pinv(A m,n )y, where pinv represents the generalized inverse operation;

[0070] e. Output training results The algorithm terminates, where x represents a column vector consisting of input data;

[0071] The specific steps of the second part are as follows:

[0072] a. Test set data Divide into two, and obtain sub-dataset z1 with cardinality m1 and sub-dataset z2 with cardinality m2;

[0073] b. z1 and Substitute them into the first part of the fast kernel learning algorithm for calculation to obtain a set of training model sequences

[0074] c. Use the results in z2 and b and the following formula to get the final polynomial degree s′ and the corresponding training model:

[0075]

[0076] Among them, Ξ represents no more than The set of non-negative integers, represents the floor function, sgn represents the sign function, Represents the input x in the training set i and output y i ∈[-M,M], where M represents the real number boundary.

[0077] Specifically, in step 3), when the drop depth coefficient α and the disturbance total power time domain function ΔP L When the product of (t) is greater than 0.9 Hz, it is considered that the frequency safety constraint is not met.

[0078] Example 2

[0079] This embodiment is a fast kernel learning system for evaluating the lowest point of the power system frequency response, such as Figure 2 As shown, it consists of a frequency response minimum point analytical formula establishment unit, a fast kernel learning unit and a power system frequency security assessment unit.

[0080] Frequency response minimum point analytical formula establishment unit: approximates the frequency response dynamics of the generator through the second-order motion equation, takes the common mode component of the frequency response trajectory of the multi-machine power system as the source of the frequency response minimum point, and establishes an analytical formula for estimating the frequency response minimum point by the multi-machine second-order motion equation.

[0081] Fast kernel learning unit: takes the total number, mean, kurtosis and skewness of the power system stabilizer PSS as input, uses the trained fast kernel learning algorithm, and outputs the frequency response minimum point estimation error caused by not considering the power system stabilizer PSS.

[0082] Power system frequency security assessment unit: superimpose the frequency response minimum point analytical expression obtained by the frequency response minimum point analytical expression establishment unit and the frequency response minimum point estimated error obtained by the estimated error output unit to obtain the actual power system frequency response minimum point, and compare it with the safety threshold. If it is greater than the safety threshold, the output is 1, otherwise the output is 0, thereby realizing the power system frequency security assessment.

[0083] In the frequency response lowest point analytical formula establishment unit, the common mode component in the frequency response is used to simulate the frequency trajectory of the power system. When a step disturbance occurs in the load, the common mode frequency response of the power system is:

[0084]

[0085] Where Δω(s) represents the common mode frequency of the power system; s is the Laplace operator; ΔP L (s) represents the frequency domain function of the total load disturbance of the power system. The single-input and single-output transfer functions of the frequency-active power of the power generation equipment in the power system are represented by G i (s),i=1,…,n description, G i (s) Second order model representing only inertia, damping and governor effects:

[0086] G i (s) = J i s+D i +s / K i ,

[0087] Among them, the parameter J i , D i and K i G i(s) Inertia, damping and governor action parameters corresponding to the second-order model;

[0088] The common mode frequency response of the power system is expressed as a unified structural model:

[0089]

[0090] Among them, the parameter J u , D u and K u They are the sum of the corresponding parameters of the second-order model of the power generation equipment in the system.

[0091] In the frequency response lowest point analytical formula establishing unit, the frequency lowest point analytical formula estimated by the multi-machine second-order motion equation is characterized by the drop depth coefficient α:

[0092]

[0093] In the fast kernel learning unit, the fast kernel learning algorithm is divided into two parts. The first part is to determine the learning model after the polynomial degree s is given. The specific steps are as follows:

[0094] a. Input test set data and the polynomial degree s∈N, N represents a set of non-negative integers, and m is the number of samples;

[0095] b. Order Represents the number of center points selected from the test set and a set of independent input data represents the center point of the input data space, C is the number of combinations, and d represents the vector x i The dimension of

[0096] c. Based on the center point Calculate the semi-positive definite kernel matrix K represents the kernel matrix, and the polynomial kernel function is expressed as:

[0097]

[0098] κ represents the kernel function. The function in the above formula can be used as a kernel function only when it satisfies the following semi-positive definite condition of the kernel matrix K:

[0099]

[0100] d. Let the output vector of the test set be y = (y1,…,y m ) T The undetermined parameters in the training model form a vector c = (c1,…,c n ) T , then c=pinv(A m,n )y, where pinv represents the generalized inverse operation;

[0101] e. Output training results The algorithm terminates, where x represents a column vector consisting of input data;

[0102] The specific steps of the second part are as follows:

[0103] a. Test set data Divide into two, and obtain sub-dataset z1 with cardinality m1 and sub-dataset z2 with cardinality m2;

[0104] b. z1 and Substitute them into the first part of the fast kernel learning algorithm for calculation to obtain a set of training model sequences

[0105] c. Use the results in z2 and b and the following formula to get the final polynomial degree s′ and the corresponding training model:

[0106]

[0107] Among them, Ξ represents no more than The set of non-negative integers, represents the floor function, sgn represents the sign function, Represents the input x in the training set i and output y i ∈[-M,M], where M represents the real number boundary.

[0108] In the power system frequency security assessment unit, when the drop depth coefficient α and the disturbance total power time domain function ΔP L When the product of (t) is greater than 0.9 Hz, it is considered that the frequency safety constraint is not met.

[0109] Application Examples

[0110] The specific implementation process of the method of the present invention is as follows:

[0111] like Figure 3As shown in the figure, the equipment parameters and active disturbance power values ​​of a small number of synchronous machines in the power system are directly input into the common mode frequency analysis calculation link to obtain the estimated values ​​of the frequency low point and the falling speed (the new energy units in the system are considered to provide no frequency support service); the distribution position information (the total number and the variance, kurtosis and skewness of the total number of generator nodes) and the active disturbance power values ​​of the excitation link and the power system stabilizer PSS (using the same structure to facilitate the analysis of the law) not considered in the common mode frequency theory are mainly input into the kernel learning model (i.e., the fast kernel learning algorithm), and the measured values ​​of the frequency low point and the falling speed in the electromechanical transient simulation curve are used to train the kernel learning model parameters. Finally, the kernel learning model is directly used to estimate the frequency low point and the falling speed under different PSS distribution conditions and different power disturbance values, and compared with the two preset safety thresholds. If the threshold is not exceeded, the output result is 1, which means that the frequency safety and stability conditions of the power system are met; otherwise, the output is 0, which means that the power system setting does not meet the frequency safety and stability requirements.

[0112] In Matlab / Simulink software, the example uses the IEEE39 node system for simulation analysis. The rated frequency is 60Hz, including 10 synchronous generators, of which G1-9 are steam turbine generators and G10 is a hydro turbine generator. Table 1 gives the parameter settings of the simplified model of each generator, and the damping effect is not considered in the table. The load uses a constant impedance load for simplified analysis. In addition, G1 actually represents the external power grid, and the original capacity parameter is too large to affect the frequency dynamic process, so its capacity is set to the capacity of an ordinary generator.

[0113] Table 1 Simplified model parameter settings for each generator

[0114] dynamo G1 G2 G3 G4 G5 Rated capacity(MVA) 1000 700 800 800 600 Moment of inertia (pu) 10.0 8.66 8.95 7.15 8.67 dynamo G6 G7 G8 G9 G10 Rated capacity(MVA) 800 700 700 1000 1000 Moment of inertia (pu) 8.70 7.54 6.94 6.90 8.40

[0115] In addition, the general power system stabilizer PSS of PSS 2B model is adopted to reflect the impact of voltage control on frequency dynamics. By changing the installation location and quantity, training set data can be generated for kernel learning system parameter training.

[0116] The thresholds of the frequency low point and the falling speed are set to 49.1Hz and 0.9Hz / s respectively, the power disturbance (increased load power) is randomly distributed between 320-1000MW, and the number of PSS installations is randomly distributed between 0-9 (no PSS is set for the turbine). The IEEE39 bus system is simulated and analyzed using the Simulink simulation environment based on the Matlab platform, generating 150 sets of frequency low point and falling speed results as training set outputs; another 150 sets of data are used as test sets. According to Figure 3The method flow is as follows. The training set data is divided into two groups of data, 100 and 50 respectively. The first group is used to train the polynomial coefficients in kernel learning, and the second group is used to determine the optimal polynomial degree. The input is the number, variance, kurtosis and skewness of PSS, a total of 4 dimensions, and the obtained polynomial degree is 3.

[0117] The results in Table 2 verify the effectiveness of the method proposed in the present invention. Regardless of the test set or the training set, the accuracy of the kernel learning method of the present invention reaches more than 90%, indicating that the method proposed in the present invention does not have an overfitting problem. The positive judgment error rate in Table 2 represents that the output of 1 is mistakenly judged as 0, and the reverse judgment error rate is the opposite. In addition, whether in the training set or the test set, the positive judgment error rate is lower than the reverse judgment error rate. The main reason is that after considering the PSS configuration, the actual frequency low point is generally higher than the estimated value of the common mode frequency theory. The reason is that the voltage control dynamics will reduce the influence of the load power on the dynamics of the generator rotor. At the same time, since the kernel learning algorithm only uses one iteration, when the frequency low point in the data used is high, the training result is high, resulting in errors in the form of overestimation of the frequency low point.

[0118] Table 2 Performance of the method of the present invention

[0119]

[0120] Therefore, it can be seen from the above implementation that the fast kernel learning method of the present invention avoids the convergence problem caused by the iterative solution of the general machine learning algorithm, and does not reduce the generalization ability of the learning results.

[0121] The above embodiments are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications and changes made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A fast kernel learning method for evaluating the lowest point of power system frequency response, characterized in that: Includes steps: 1) The frequency response dynamics of the generator is approximated by the second-order motion equation, the common-mode component of the frequency response trajectory of the multi-machine power system is used as the source of the lowest point of the frequency response, and an analytical expression for estimating the lowest point of the frequency response by the second-order motion equation of the multi-machine is established; 2) taking the total number, mean, kurtosis and skewness of the power system stabilizer PSS as input, and using the trained fast kernel learning algorithm, outputting the frequency response minimum point estimation error caused by not considering the power system stabilizer PSS in step 1); 3) The analytical expression of the lowest frequency response point estimated in step 1) is superimposed with the estimated error of the lowest frequency response point obtained in step 2) to obtain the actual lowest frequency response point of the power system, and compare it with the safety threshold. If it is greater than the safety threshold, the output is 1, otherwise the output is 0, thereby realizing the frequency safety assessment of the power system.

2. The fast kernel learning method for evaluating the lowest point of power system frequency response according to claim 1, characterized in that: In step 1), the common mode component in the frequency response is used to simulate the frequency trajectory of the power system. When a step disturbance occurs in the load, the common mode frequency response of the power system is: Where Δω(s) represents the common mode frequency of the power system; s is the Laplace operator; ΔP L (s) represents the frequency domain function of the total load disturbance of the power system. The single-input and single-output transfer functions of the frequency-active power of the power generation equipment in the power system are represented by G i (s),i=1,…,n description, G i (s) Second order model representing only inertia, damping and governor effects: G i (s)=J i s+D i +s / K i , Among them, the parameter J i , D i and K i G i (s) Inertia, damping and governor action parameters corresponding to the second-order model; The common mode frequency response of the power system is expressed as a unified structural model: Among them, the parameter J u , D u and K u They are the sum of the corresponding parameters of the second-order model of the power generation equipment in the system.

3. The fast kernel learning method for evaluating the lowest point of the power system frequency response according to claim 2, characterized in that: In the step 1), the frequency minimum point analytical expression estimated by the multi-machine second-order motion equation is characterized by the drop depth coefficient α:

4. The fast kernel learning method for evaluating the lowest point of power system frequency response according to claim 1, characterized in that: In step 2), the fast kernel learning algorithm is divided into two parts. The first part is to determine the learning model after the polynomial degree s is given. The specific steps are as follows: a. Input test set data and the polynomial degree s∈N, N represents a set of non-negative integers, and m is the number of samples; b. Order Represents the number of center points selected from the test set and a set of independent input data represents the center point of the input data space, C is the number of combinations, and d represents the vector x i The dimension of c. Based on the center point Calculate the semi-positive definite kernel matrix K represents the kernel matrix, and the polynomial kernel function is expressed as: κ represents the kernel function. The function in the above formula can be used as a kernel function only when it satisfies the following semi-positive definite condition of the kernel matrix K: d. Let the output vector of the test set be y = (y1,...,y m ) T The undetermined parameters in the training model form a vector c = (c1,...,c n ) T , then c=pinv(A m,n )y, where pinv represents the generalized inverse operation; e. Output training results The algorithm terminates, where x represents a column vector consisting of input data; The specific steps of the second part are as follows: a. Test set data Divide into two, and obtain sub-dataset z1 with cardinality m1 and sub-dataset z2 with cardinality m2; b. z1 and Substitute them into the first part of the fast kernel learning algorithm for calculation to obtain a set of training model sequences c. Use the results in z2 and b and the following formula to get the final polynomial degree s′ and the corresponding training model: Among them, Ξ represents no more than The set of non-negative integers, represents the floor function, sgn represents the sign function, Represents the input x in the training set i and output y i ∈[-M,M], where M represents the real number boundary.

5. The fast kernel learning method for evaluating the lowest point of the frequency response of a power system according to claim 3, characterized in that: In step 3), when the drop depth coefficient α and the disturbance total power time domain function ΔP L When the product of (t) is greater than 0.9 Hz, it is considered that the frequency safety constraint is not met.

6. A fast kernel learning system for evaluating the lowest point of the frequency response of a power system, characterized in that: include: Frequency response lowest point analytical formula establishment unit: approximate the frequency response dynamics of the generator through the second-order motion equation, take the common mode component of the frequency response trajectory of the multi-machine power system as the source of the frequency response lowest point, and establish an analytical formula for estimating the frequency response lowest point by the multi-machine second-order motion equation; Fast kernel learning unit: takes the total number, mean, kurtosis and skewness of the power system stabilizer PSS as input, uses the trained fast kernel learning algorithm, and outputs the frequency response minimum point estimation error caused by not considering the power system stabilizer PSS; Power system frequency security assessment unit: superimpose the frequency response minimum point analytical expression obtained by the frequency response minimum point analytical expression establishment unit and the frequency response minimum point estimated error obtained by the estimated error output unit to obtain the actual power system frequency response minimum point, and compare it with the safety threshold. If it is greater than the safety threshold, the output is 1, otherwise the output is 0, thereby realizing the power system frequency security assessment.

7. The fast kernel learning system for evaluating the lowest point of the frequency response of a power system according to claim 6, characterized in that: In the frequency response lowest point analytical formula establishment unit, the common mode component in the frequency response is used to simulate the frequency trajectory of the power system. When a step disturbance occurs in the load, the common mode frequency response of the power system is: Where Δω(s) represents the common mode frequency of the power system; s is the Laplace operator; ΔP L (s) represents the frequency domain function of the total load disturbance of the power system. The single-input and single-output transfer functions of the frequency-active power of the power generation equipment in the power system are represented by G i (s),i=1,...,n description,G i (s) Second order model representing only inertia, damping and governor effects: G i (s)=J i s+D i +s / K i , Among them, the parameter J i , D i and K i G i (s) Inertia, damping and governor action parameters corresponding to the second-order model; The common mode frequency response of the power system is expressed as a unified structural model: Among them, the parameter J u , D u and K u They are the sum of the corresponding parameters of the second-order model of the power generation equipment in the system.

8. The fast kernel learning system for evaluating the lowest point of the power system frequency response according to claim 7, characterized in that: In the frequency response lowest point analytical formula establishing unit, the frequency lowest point analytical formula estimated by the multi-machine second-order motion equation is characterized by the drop depth coefficient α:

9. The fast kernel learning system for evaluating the lowest point of the frequency response of a power system according to claim 6, characterized in that: In the fast kernel learning unit, the fast kernel learning algorithm is divided into two parts. The first part is to determine the learning model after the polynomial degree s is given. The specific steps are as follows: a. Input test set data and the polynomial degree s∈N, N represents a set of non-negative integers, and m is the number of samples; b. Order Represents the number of center points selected from the test set and a set of independent input data represents the center point of the input data space, C is the number of combinations, and d represents the vector x i The dimension of c. Based on the center point Calculate the semi-positive definite kernel matrix K represents the kernel matrix, and the polynomial kernel function is expressed as: κ represents the kernel function. The function in the above formula can be used as a kernel function only when it satisfies the following semi-positive definite condition of the kernel matrix K: d. Let the output vector of the test set be y = (y1,...,y m ) T The undetermined parameters in the training model form a vector c = (c1,...,c n ) T , then c=pinv(A m,n )y, where pinv represents the generalized inverse operation; e. Output training results The algorithm terminates, where x represents a column vector consisting of input data; The specific steps of the second part are as follows: a. Test set data Divide into two, and obtain sub-dataset z1 with cardinality m1 and sub-dataset z2 with cardinality m2; b. z1 and Substitute them into the first part of the fast kernel learning algorithm for calculation to obtain a set of training model sequences c. Use the results in z2 and b and the following formula to get the final polynomial degree s′ and the corresponding training model: Among them, Ξ represents no more than The set of non-negative integers, represents the floor function, sgn represents the sign function, Represents the input x in the training set i and output y i ∈[-M,M], where M represents the real number boundary.

10. The fast kernel learning system for evaluating the lowest point of frequency response of a power system according to claim 8, characterized in that: In the power system frequency security assessment unit, when the drop depth coefficient α and the disturbance total power time domain function ΔP L When the product of (t) is greater than 0.9 Hz, it is considered that the frequency safety constraint is not met.

Images