Power system frequency regulation control method

Abstract

The invention discloses a power system frequency regulation control method, which is based on a control strategy that a bipolar photovoltaic grid-connected inverter and a doubly-fed wind turbine generator participate in system frequency regulation, and aims to overcome the defect that a traditional photovoltaic inverter lacks active participation in frequency regulation and the defect of a conventional frequency modulation control mode of the doubly-fed wind turbine generator. By analyzing the control algorithm of the inverter and the requirement for frequency regulation of the power system and combining a neural network algorithm and an existing control strategy, the capability of the photovoltaic inverter and the doubly-fed wind turbine generator to participate in frequency regulation is enhanced, and the frequency stability and the frequency supporting capability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the field of power system control, and in particular to a frequency regulation control method for a new energy power system including a wind power generating set and a photovoltaic generating set. Background Art

[0002] It is an irresistible trend to develop renewable energy and build a new power system with new energy as the main body. Both developed and developing countries have taken the development of renewable energy such as wind energy and solar energy as an important means to cope with the dual challenges of energy and climate change in the new century. However, new energy such as wind energy and solar energy are connected to the power grid through converters. The low inertia and weak damping characteristics of the new power system are becoming increasingly prominent, and the inertial response ability of the power system is greatly weakened. At present, most new energy power stations generally have problems such as large power fluctuations and poor stability, which seriously threaten the safe and stable operation of the power grid and cause great difficulties in the control and operation of the power grid. Since the operation mode of traditional synchronous generators is very different from that of new energy units, the integration of a high proportion of new energy into the power grid will have a great impact on the stability of the power system. First of all, the traditional synchronous generator has a rotor that can store kinetic energy. When the power system is disturbed, it can spontaneously release the rotor kinetic energy to provide a certain degree of inertia support for the system. At the same time, it can provide primary frequency regulation for the system and has good frequency regulation capabilities. However, new energy units are mainly connected to the power grid through power electronic devices, which do not have inertia capacity. Since a large number of power electronic devices have replaced traditional synchronous generators, the overall inertia level of the system has been weakened. When the system is disturbed, the frequency fluctuation speed can be aggravated and the system stability can be reduced. Secondly, the traditional synchronous generator has an excitation regulation system. When the voltage drops, the excitation system controls the excitation current to adjust the output of reactive power, thereby maintaining the stability of the output voltage. However, new energy units generally only control the port voltage through a simple reactive droop link. When the system has a large voltage drop, the voltage support capacity provided by the reactive / voltage linear regulation of new energy units is limited. Therefore, the access of large-scale new energy units to the power grid is likely to weaken the overall frequency and voltage regulation capabilities of the system, weaken the system robustness, and bring greater risks to the system operation. Therefore, it is necessary to develop a frequency regulation control method suitable for new energy power systems including wind turbines and photovoltaic generators. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a method for frequency regulation and control of a power system, comprising the following steps:

[0004] S1, collecting the operation data of the wind turbine generator set and the corresponding output value of the wind turbine generator set from the power system operation history data, and preprocessing the operation data of the wind turbine generator set and the output value of the wind turbine generator set, wherein the preprocessing includes deleting null values, deleting missing values, deleting duplicate values ​​and normalizing;

[0005] S2, collecting the operation data of the photovoltaic generator set and the output value of the corresponding wind generator set from the power system operation history data, and preprocessing the operation data of the photovoltaic generator set and the output value of the photovoltaic generator set, wherein the preprocessing includes deleting null values, deleting missing values, deleting duplicate values ​​and normalizing;

[0006] S3, using the preprocessed operation data of the wind turbine generator set as the input value of the neural network, using the corresponding preprocessed output value of the wind turbine generator set as the output value of the neural network, and training the first neural network to obtain a trained first neural network;

[0007] S4, using the preprocessed operation data of the photovoltaic power generation group as the input value of the neural network, using the corresponding preprocessed output value of the photovoltaic power generation group as the output value of the neural network, and training the neural network to obtain a trained second neural network;

[0008] S5, collecting real-time operating data of wind turbine generator sets in the power system that needs frequency regulation, and normalizing the real-time operating data of the wind turbine generator sets;

[0009] S6, inputting the normalized real-time operation data of the wind turbine generator set into the trained first neural network, and the output value of the first neural network is the normalized predicted power of the wind turbine generator set;

[0010] S7, collecting real-time operating data of the photovoltaic generator set in the power system that needs frequency regulation, and normalizing the real-time operating data of the photovoltaic generator set;

[0011] S8, inputting the normalized real-time operation data of the photovoltaic power generation group into the trained second neural network, and the output value of the second neural network is the normalized predicted power of the photovoltaic power generation group;

[0012] S9, calculating an adjustment value of the power system power according to a target value of the power control of the power system, a predicted power of the wind turbine generator set, and a predicted power of the photovoltaic generator set;

[0013] S10, distributing the adjustment value of the power of the electric power system to obtain a power adjustment plan for the wind turbine generator set and the photovoltaic generator set;

[0014] S11, the wind turbine generator set and the photovoltaic generator set adjust their operating states according to the power adjustment plan, thereby realizing the frequency regulation control of the power system.

[0015] Furthermore, the wind turbine generator set is one or more double-fed wind turbine generator sets.

[0016] Furthermore, the photovoltaic power generation group is one or more bipolar photovoltaic grid-connected inverters.

[0017] Furthermore, the operating data of the wind turbine generator set includes blade radius, wind speed, pitch angle and rotor speed.

[0018] Furthermore, the operating data of the photovoltaic power generation group includes light intensity, illumination, external temperature and battery temperature.

[0019] Furthermore, the method for setting the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, the learning rate of the model and the number of iterations of the model of the neural network model comprises the following steps:

[0020] S601, generate a population consisting of m 4-dimensional vector particles, and randomly generate an initial population particle, the initial position of the particle is x0(h 1,0 ,h 2,0 ,ε i,0 ,n i,0 ), where h 1,0 is the initial number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,0 is the initial number of neurons in the second hidden layer of the medium gated recurrent unit neural network model, ε0 is the initial learning rate of the medium gated recurrent unit neural network model, and n0 is the initial number of iterations of the medium gated recurrent unit neural network model;

[0021] S602, the particles in the population search for the best position. The specific search method is:

[0022] p t =x t-1 +v t

[0023] v t Find the speed of the particles in the population moving in space for the t-th position, p t is the position found by the particle in the t-th position search, x t-1 The position found by the particle in the population in the t-1th position search;

[0024] S603, determine whether the position found by the particle in the population is the best position. If the fitness of the position found by the particle in the t-th position search is higher than the position found by the particle in the population in the t-1th position search, then update the position of the particle in the population. The updating method is:

[0025] x t =p t

[0026] If the fitness of the particle found in the t-th position search is not higher than the position found by the particle in the population in the t-1-th position search, the particle position remains unchanged, that is,

[0027] x t =x t-1

[0028] S604, determine whether the number of times the particle in the population searches for the best position has reached the set number. If not, return to step S502. If it has reached the set number, the search for the best position of the particle in the population is completed. At this time, the position of the particle is x t (h 1,t ,h 2,t ,ε t ,n t ) is the best position, where h 1,t is the number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,t is the number of neurons in the second hidden layer of the gated recurrent unit neural network model, ε t is the learning rate of the gated recurrent unit neural network model, n t is the number of iterations of the gated recurrent unit neural network model;

[0029] Furthermore, the method for calculating the fitness of the particle in finding the position in the t-th position search in S603 comprises the following steps:

[0030] S60301, extract the component elements of the position found by the particle in the t-th position search as the parameters of the medium gated recurrent unit neural network model, where the position found by the particle in the t-th position search is p t (h 1,t ,h 2,t ,ε t ,n t ), h 1,t is the number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,t is the number of neurons in the second hidden layer of the gated recurrent unit neural network model, ε t is the learning rate of the gated recurrent unit neural network model, n t is the number of iterations of the gated recurrent unit neural network model;

[0031] S60302, using the data samples in the test set as input values, inputting them into the generator set power prediction model composed of the medium gated cyclic unit neural network model parameters obtained by the calculation method described in S60301, and the output value obtained is the generator set power prediction data.

[0032] S60303, calculate the fitness of the particle in the t-th position search. The specific method is:

[0033]

[0034] Where M is the fitness, n is the number of samples in the test set, and y is the true value of the generator power. is the predicted value of the generator set power.

[0035] Furthermore, the velocity v of the particles in the population moving in space in the t-th position search in S602 is t The calculation method is

[0036] v t =v t-1 +c1×rand()×(pb t -x t )+c2×rand()×(gb t -x t )

[0037] Among them, v t-1 Find the speed of the particles in the population moving in space for the t-th position, pb t is the position with the best fitness among all the positions experienced by the particle individual, gb t is the position with the best fitness searched by all particles in the population, c1 is the first learning factor, c2 is the second learning factor, rand() is a random number between (0,1), and x t The position found by the particle in the population for the t-1th position search.

[0038] Furthermore, the method for normalizing the data is a z-score method.

[0039] Further, the method of calculating the adjustment value of the power system power according to the target value of the power control of the power system, the predicted power of the wind turbine generator set and the predicted power of the photovoltaic generator set in S9 is:

[0040] ΔP=P TAR -P1-P2

[0041] Where ΔP is the adjustment value of the power system, P TAR is the target value of power control of the power system, P1 is the predicted power of the wind turbine generator set, and P2 is the predicted power of the photovoltaic generator set.

[0042] The power system frequency regulation control method of the present invention is based on the control strategy of bipolar photovoltaic grid-connected inverter and doubly-fed wind turbine participating in system frequency regulation. In view of the defect of traditional photovoltaic inverter lacking active participation in frequency regulation and the insufficiency of conventional frequency regulation control mode of doubly-fed wind turbine, the control algorithm of the inverter and the demand of power system frequency regulation are analyzed, and the neural network algorithm and the existing control strategy are combined to enhance the ability of photovoltaic inverter and doubly-fed wind turbine to participate in frequency regulation, and improve the frequency stability and frequency support capability of the system. In addition, the present invention optimizes the neural network model by using the particle swarm algorithm, and applies it to the power prediction of wind and photovoltaic generator sets. Through the particle swarm algorithm, the hidden layer and the number of neurons of the gated recurrent unit neural network are improved, so that the characteristics of the generator set operation can be fully obtained, and the accuracy of the model for predicting the power of the generator set is improved. At the same time, the advantages of the simple structure of the neural network model are retained, and the efficiency of predicting the power of the generator set is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings in the specification, which constitute a part of this application, are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0044] Figure 1 The present invention is a flowchart of a method for controlling frequency regulation of a power system. DETAILED DESCRIPTION

[0045] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to embodiments and drawings.

[0046] like Figure 1 As shown, the power system frequency regulation control method in this embodiment includes the following steps:

[0047] S1, collecting the operation data of the wind turbine generator set and the corresponding output value of the wind turbine generator set from the power system operation history data, and preprocessing the operation data of the wind turbine generator set and the output value of the wind turbine generator set, wherein the preprocessing includes deleting null values, deleting missing values, deleting duplicate values ​​and normalizing;

[0048] S2, collecting the operation data of the photovoltaic generator set and the output value of the corresponding wind generator set from the power system operation history data, and preprocessing the operation data of the photovoltaic generator set and the output value of the photovoltaic generator set, wherein the preprocessing includes deleting null values, deleting missing values, deleting duplicate values ​​and normalizing;

[0049] S3, using the preprocessed operation data of the wind turbine generator set as the input value of the neural network, using the corresponding preprocessed output value of the wind turbine generator set as the output value of the neural network, and training the first neural network to obtain a trained first neural network;

[0050] S4, using the preprocessed operation data of the photovoltaic power generation group as the input value of the neural network, using the corresponding preprocessed output value of the photovoltaic power generation group as the output value of the neural network, and training the neural network to obtain a trained second neural network;

[0051] S5, collecting real-time operating data of wind turbine generator sets in the power system that needs frequency regulation, and normalizing the real-time operating data of the wind turbine generator sets;

[0052] S6, inputting the normalized real-time operation data of the wind turbine generator set into the trained first neural network, and the output value of the first neural network is the normalized predicted power of the wind turbine generator set;

[0053] S7, collecting real-time operating data of the photovoltaic generator set in the power system that needs frequency regulation, and normalizing the real-time operating data of the photovoltaic generator set;

[0054] S8, inputting the normalized real-time operation data of the photovoltaic power generation group into the trained second neural network, and the output value of the second neural network is the normalized predicted power of the photovoltaic power generation group;

[0055] S9, calculating an adjustment value of the power system power according to a target value of the power control of the power system, a predicted power of the wind turbine generator set, and a predicted power of the photovoltaic generator set;

[0056] S10, distributing the adjustment value of the power of the electric power system to obtain a power adjustment plan for the wind turbine generator set and the photovoltaic generator set;

[0057] S11, the wind turbine generator set and the photovoltaic generator set adjust their operating states according to the power adjustment plan, thereby realizing the frequency regulation control of the power system.

[0058] The wind turbine generator set is one or more double-fed wind turbine generator sets.

[0059] The photovoltaic power generation group is one or more bipolar photovoltaic grid-connected inverters.

[0060] The operating data of the wind turbine generator set includes blade radius, wind speed, pitch angle and rotor speed.

[0061] The operating data of the photovoltaic power generation group includes light intensity, illumination, external temperature and battery temperature.

[0062] The method for setting the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, the learning rate of the model and the number of iterations of the model of the neural network model comprises the following steps:

[0063] S601, generate a population consisting of m 4-dimensional vector particles, and randomly generate an initial population particle, the initial position of the particle is x0(h 1,0 ,h 2,0 ,ε i,0 ,n i,0 ), where h 1,0 is the initial number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,0 is the initial number of neurons in the second hidden layer of the medium gated recurrent unit neural network model, ε0 is the initial learning rate of the medium gated recurrent unit neural network model, and n0 is the initial number of iterations of the medium gated recurrent unit neural network model;

[0064] S602, the particles in the population search for the best position. The specific search method is:

[0065] p t =x t-1 +v t

[0066] v t Find the speed of the particles in the population moving in space for the t-th position, p t is the position found by the particle in the t-th position search, x t-1 The position found by the particle in the population in the t-1th position search;

[0067] S603, determine whether the position found by the particle in the population is the best position. If the fitness of the position found by the particle in the t-th position search is higher than the position found by the particle in the population in the t-1th position search, then update the position of the particle in the population. The updating method is:

[0068] x t =p t

[0069] If the fitness of the particle found in the t-th position search is not higher than the position found by the particle in the population in the t-1-th position search, the particle position remains unchanged, that is,

[0070] x t =x t-1

[0071] S604, determine whether the number of times the particle in the population searches for the best position has reached the set number. If not, return to step S502. If it has reached the set number, the search for the best position of the particle in the population is completed. At this time, the position of the particle is x t (h 1,t ,h 2,t ,ε t ,n t ) is the best position, where h 1,t is the number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,t is the number of neurons in the second hidden layer of the gated recurrent unit neural network model, ε t is the learning rate of the gated recurrent unit neural network model, n t is the number of iterations of the gated recurrent unit neural network model;

[0072] The method for calculating the fitness of the particle in finding the position in the t-th position search in S603 includes the following steps:

[0073] S60301, extract the component elements of the position found by the particle in the t-th position search as the parameters of the medium gated recurrent unit neural network model, where the position found by the particle in the t-th position search is p t (h 1,t ,j 2,t ,ε t ,n t ), h 1,t is the number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,t is the number of neurons in the second hidden layer of the gated recurrent unit neural network model, ε t is the learning rate of the gated recurrent unit neural network model, n t is the number of iterations of the gated recurrent unit neural network model;

[0074] S60302, using the data samples in the test set as input values, inputting them into the generator set power prediction model composed of the medium gated cyclic unit neural network model parameters obtained by the calculation method described in S60301, and the output value obtained is the generator set power prediction data.

[0075] S60303, calculate the fitness of the particle in the t-th position search. The specific method is:

[0076]

[0077] Where M is the fitness, n is the number of samples in the test set, and y is the true value of the generator power. is the predicted value of the generator set power.

[0078] The speed v of the particles in the population moving in space in the t-th position search in S602 t The calculation method is

[0079] v t =v t-1 +c1×rand()×(pb t -x t )+c2×rand()×(gb t -x t )

[0080] Among them, v t-1 Find the speed of the particles in the population moving in space for the t-th position, pb t is the position with the best fitness among all the positions experienced by the particle individual, gb t is the position with the best fitness searched by all particles in the population, c1 is the first learning factor, c2 is the second learning factor, rand() is a random number between (0,1), and x t The position found by the particle in the population for the t-1th position search.

[0081] The method for normalizing the data is the z-score method.

[0082] The method of calculating the adjustment value of the power system power according to the target value of the power control of the power system, the predicted power of the wind turbine generator set and the predicted power of the photovoltaic generator set in S9 is as follows:

[0083] ΔP=P TAR -P1-P2

[0084] Where ΔP is the adjustment value of the power system, P TAR is the target value of power control of the power system, P1 is the predicted power of the wind turbine generator set, and P2 is the predicted power of the photovoltaic generator set.

[0085] The frequency regulation control method of the power system provided in this embodiment is applied to perform frequency regulation control on the simulation model. The rated output power of wind power generation is 1000MW, the wind speed is 7m / s, the rated output power of the synchronous generator set is 1300MW, the active power consumed by the local load is 500MW, and the DC transmission power is 1000MW. When the load increases by 150MW, if wind power generation does not participate in the frequency regulation of the power system, the minimum frequency of the system reaches 49.4Hz at this time, and the stability and safety of the power system frequency is threatened. When the wind turbine participates in the frequency regulation of the power system, the wind turbine increases the active power of 50MW to support the stability of the power system frequency under the control method of this embodiment. When the wind turbine ends the frequency regulation, the system enters a new steady state, which is consistent with the steady-state value of the frequency when the wind turbine does not participate in the frequency regulation. That is, the virtual inertia control can better support the stability of the power grid frequency, increase the inertia of the system, and reduce the fluctuation amplitude of the system frequency under load fluctuations. This embodiment is applied to control the frequency regulation of the wind and solar power generation power system.

[0086] In addition, photovoltaic power generation is verified using a single-machine equivalent model. The rated output power of photovoltaic power generation is 400MW, the rated output power of synchronous generators is 1300MW, the active power consumed by local loads is 600MW, the DC transmission power is 1000MW, and the voltage level is 330kV. When the load increases by 100MW, if photovoltaic power generation does not participate in the frequency regulation of the power system, the minimum frequency of the system reaches 49.3Hz, which brings certain hidden dangers to the safe and stable operation of the power system. When photovoltaic power generation participates in the frequency regulation of the power system, the generator adds 80MW of active power to support the stability of the power system frequency. When photovoltaic power generation ends the frequency regulation, the system enters a new steady state, which is consistent with the steady-state value of the frequency when the photovoltaic generator does not participate in the frequency regulation. It has good support for the system frequency, improves the inertia of the system, and reduces the fluctuation amplitude of the system frequency under load fluctuations.

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for controlling the frequency regulation of a power system, characterized in that: The following steps are involved: S1, collecting the operation data of the wind turbine generator set and the corresponding output value of the wind turbine generator set from the power system operation history data, and preprocessing the operation data of the wind turbine generator set and the output value of the wind turbine generator set, wherein the preprocessing includes deleting null values, deleting missing values, deleting duplicate values ​​and normalizing; S2, collecting the operation data of the photovoltaic generator set and the output value of the corresponding wind generator set from the power system operation history data, and preprocessing the operation data of the photovoltaic generator set and the output value of the photovoltaic generator set, wherein the preprocessing includes deleting null values, deleting missing values, deleting duplicate values ​​and normalizing; S3, using the preprocessed operation data of the wind turbine generator set as the input value of the neural network, using the corresponding preprocessed output value of the wind turbine generator set as the output value of the neural network, and training the first neural network to obtain a trained first neural network; S4, using the preprocessed operation data of the photovoltaic power generation group as the input value of the neural network, using the corresponding preprocessed output value of the photovoltaic power generation group as the output value of the neural network, and training the neural network to obtain a trained second neural network; S5, collecting real-time operating data of wind turbine generator sets in the power system that needs frequency regulation, and normalizing the real-time operating data of the wind turbine generator sets; S6, inputting the normalized real-time operation data of the wind turbine generator set into the trained first neural network, and the output value of the first neural network is the normalized predicted power of the wind turbine generator set; S7, collecting real-time operating data of the photovoltaic generator set in the power system that needs frequency regulation, and normalizing the real-time operating data of the photovoltaic generator set; S8, inputting the normalized real-time operation data of the photovoltaic power generation group into the trained second neural network, and the output value of the second neural network is the normalized predicted power of the photovoltaic power generation group; S9, calculating an adjustment value of the power system power according to a target value of the power control of the power system, a predicted power of the wind turbine generator set, and a predicted power of the photovoltaic generator set; S10, distributing the adjustment value of the power of the electric power system to obtain a power adjustment plan for the wind turbine generator set and the photovoltaic generator set; S11, the wind turbine generator set and the photovoltaic generator set adjust their operating states according to the power adjustment plan, thereby realizing the frequency regulation control of the power system.

2. The method for controlling the frequency regulation of a power system according to claim 1, characterized in that: The wind turbine generator set is one or more double-fed wind turbine generator sets.

3. The power system frequency regulation control method according to claim 1, characterized in that: The photovoltaic power generation group is one or more bipolar photovoltaic grid-connected inverters.

4. The method for controlling the frequency regulation of a power system according to claim 1, characterized in that: The operating data of the wind turbine generator set includes blade radius, wind speed, pitch angle and rotor speed.

5. The method for controlling the frequency regulation of a power system according to claim 1, characterized in that: The operating data of the photovoltaic power generation group includes light intensity, illumination, external temperature and battery temperature.

6. The method for controlling the frequency regulation of a power system according to claim 1, characterized in that: The method for setting the number of neurons in the first hidden layer, the number of neurons in the second hidden layer, the learning rate of the model and the number of iterations of the model of the neural network model comprises the following steps: S601, generate a population consisting of m 4-dimensional vector particles, and randomly generate an initial population particle, the initial position of the particle is x0(h 1，0 ,h 2，0 , ε i，0 , n i，0 ), where h 1，0 is the initial number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2，0 is the initial number of neurons in the second hidden layer of the medium gated recurrent unit neural network model, ε0 is the initial learning rate of the medium gated recurrent unit neural network model, and n0 is the initial number of iterations of the medium gated recurrent unit neural network model; S602, the particles in the population search for the best position. The specific search method is: p t =x t-1 +v t v t Find the speed of the particles in the population moving in space for the t-th position, p t is the position found by the particle in the t-th position search, x t-1 The position found by the particle in the population in the t-1th position search; S603, determine whether the position found by the particle in the population is the best position. If the fitness of the position found by the particle in the t-th position search is higher than the position found by the particle in the population in the t-1th position search, then update the position of the particle in the population. The updating method is: x t ＝ p t If the fitness of the particle found in the t-th position search is not higher than the position found by the particle in the population in the t-1-th position search, the particle position remains unchanged, that is, x t =x t-1 S604, determine whether the number of times the particle in the population searches for the best position has reached the set number. If not, return to step S502. If it has reached the set number, the search for the best position of the particle in the population is completed. At this time, the position of the particle is x t (h 1，t ,h 2,t , ε t ,n t ) is the best position, where h 1,t is the number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,t is the number of neurons in the second hidden layer of the gated recurrent unit neural network model, ε t is the learning rate of the gated recurrent unit neural network model, n t is the number of iterations of the gated recurrent unit neural network model.

7. The method for controlling the frequency regulation of a power system according to claim 6, characterized in that: The method for calculating the fitness of the particle in finding the position in the t-th position search in S603 includes the following steps: S60301, extract the component elements of the position found by the particle in the t-th position search as the parameters of the medium gated recurrent unit neural network model, where the position found by the particle in the t-th position search is p t (h 1，t ,h 2,t , ε t , n t ), h 1,t is the number of neurons in the first hidden layer of the gated recurrent unit neural network model, h 2,t is the number of neurons in the second hidden layer of the gated recurrent unit neural network model, ε t is the learning rate of the gated recurrent unit neural network model, n t is the number of iterations of the gated recurrent unit neural network model; S60302, using the data samples in the test set as input values, inputting them into the generator set power prediction model composed of the medium gated cyclic unit neural network model parameters obtained by the calculation method described in S60301, and the output value obtained is the generator set power prediction data. S60303, calculate the fitness of the particle in the t-th position search. The specific method is: Where M is the fitness, n is the number of samples in the test set, and y is the true value of the generator power. is the predicted value of the generator set power.

8. The method for controlling the frequency regulation of a power system according to claim 6, characterized in that: The speed v of the particles in the population moving in space in the t-th position search in S602 t The calculation method is v t =v t-1 +c1×rand()×(pb t -x t )+c2×rand()×(gb t -x t ) Among them, v t-1 Find the speed of the particles in the population moving in space for the t-th position, pb t is the position with the best fitness among all the positions experienced by the particle individual, gb t is the position with the best fitness searched by all particles in the population, c1 is the first learning factor, c2 is the second learning factor, rand() is a random number between (0, 1), and x t The position found by the particle in the population for the t-1th position search.

9. The method for controlling the frequency regulation of a power system according to claim 1, characterized in that: The method for normalizing the data is the z-score method.

10. The power system frequency regulation control method according to claim 1, characterized in that: The method of calculating the adjustment value of the power system power according to the target value of the power control of the power system, the predicted power of the wind turbine generator set and the predicted power of the photovoltaic generator set in S9 is ΔP=P TAR -P1-P2 Where ΔP is the adjustment value of the power system, P TAR is the target value of power control of the power system, P1 is the predicted power of the wind turbine generator set, and P2 is the predicted power of the photovoltaic generator set.

Images