Virtual synchronous generator control method based on adaptive RBF-LADRC

Abstract

The invention provides a virtual synchronous generator control method based on adaptive RBF-LADRC, and the method comprises the steps: building a mathematical model of a virtual synchronous generator, introducing LADRC into a virtual speed regulator of the virtual synchronous generator as an outer loop frequency controller of an active control loop, and adjusting the active output of the virtual synchronous generator; adaptively adjusting a virtual coefficient in the mathematical model by using an RBF neural network; a voltage outer loop in voltage and current double-closed-loop control is designed based on linear active disturbance rejection control, a PI controller is replaced with an LED RC controller, a voltage and current double-closed-loop control system is obtained, and after reference voltage obtained through coupling of a virtual exciter and a virtual regulator is input into the voltage outer loop, double-closed-loop parameter adjustment control is carried out. Compared with traditional control, the method has the advantages that the system overshoot can be remarkably reduced, the active power tracking speed is increased, the power grid frequency stability is enhanced, and good anti-disturbance capacity is shown under the wind power plant short-circuit fault and low-frequency disturbance conditions.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to a virtual synchronous generator control method based on adaptive RBF-LADRC. Background Art

[0002] In order to further accelerate the realization of "carbon peak" and "carbon neutrality", my country's installed capacity of renewable energy, such as photovoltaics and wind power, has increased rapidly in recent years. Taking wind power generation as an example of new energy generation, with the continuous increase in wind power grid-connected capacity, large-scale wind power grid connection has brought new challenges to the stability of the power grid. On the one hand, the access of large-scale wind power has replaced some traditional generators, and the degree of electric energy cleanliness has been greatly deepened. However, on the other hand, due to the volatility of wind turbine output, the risk of low-frequency oscillation in the system has increased. The development of modern power electronics has provided many useful technologies for the static and transient stability of power systems. Among them, virtual synchronous generator technology has received widespread attention in recent years in both research and practical applications because it can provide reliable inertia and damping for the power grid.

[0003] Gao Benfeng et al. applied virtual synchronous control to address subsynchronous and supersynchronous oscillations and grid-connected stability issues that arise in the field of flexible HVDC transmission. They used harmonic linearization to establish a sequence impedance model for a modular multilevel converter controlled by a virtual synchronous machine and investigated the impact of various control parameters on its impedance characteristics. The results showed that the grid-connected MMC (modular multilevel converter) exhibited good weak grid stability, but faced the risk of subsynchronous and supersynchronous oscillations in strong grid conditions. Incorporating virtual impedance control into the control loop increased the converter's equivalent output impedance, thereby enhancing the stability of the MMC grid-connected system. Furthermore, reactive power droop and the voltage outer loop play a decisive role in the impedance characteristics of the grid-connected MMC. Appropriately increasing the reactive power droop coefficient and the voltage outer loop integral coefficient can help reduce the negative impedance damping range, thereby improving grid-connected system stability. Qu Keqing et al., using a grid-connected solar-storage integrated system as an example, proposed a model predictive dual-loop collaborative optimization control strategy based on virtual synchronous machines. In the power outer loop, model predictive control was used to modify the VSG reference power at different stages based on the synchronous generator rotor frequency characteristics. In the inner loop, a finite set three-vector model predictive current control is used to accurately track the outer loop output reference voltage. Miguel et al. designed a photovoltaic injection system using a virtual synchronous generator control strategy to provide voltage and frequency support for the grid. The maximum power point tracking algorithm is suitable for providing a DC voltage reference and injecting active power according to the droop frequency control. The control strategy was verified through simulation and key experimental device testing. The results show that it is possible to inject photovoltaic power and provide voltage and frequency support. D. Chen et al. proposed a concept for the intelligent and autonomous integration of DC microgrids into traditional AC grids. The concept uses a DC-AC converter as a universal interface between the AC grid and various distributed energy resources connected to the DC side based on virtual synchronous generators. Together with the DC microgrid, it responds to the grid's short-term and long-term frequency regulation needs, achieving autonomous power management of the AC grid and DC microgrid.

[0004] The invention patent with application number 202211566805.3 discloses a method for suppressing subsynchronous oscillations in a virtual synchronous doubly fed wind turbine grid-connected system via series compensation, including: obtaining an expression for the output impedance of the grid-side converter based on VSG; revealing the positive damping characteristics of the grid-side converter based on VSG for subsynchronous oscillations from the perspective of the equivalent impedance of the doubly fed wind turbine grid-connected system; analyzing the subsynchronous oscillation suppression mechanism of the virtual synchronous doubly fed wind turbine grid-connected system in combination with active disturbance rejection control; and constructing an improved active disturbance rejection controller to suppress subsynchronous oscillations. The subsynchronous oscillation analysis method adopted in the above invention has a certain degree of universality and can provide a theoretical analysis basis for the subsynchronous oscillation problem caused by the high proportion of power electronic devices participating in the new energy power generation system. At the same time, the suppression method proposed in the above invention is not only applicable to doubly fed generator sets, but also applicable to the suppression of subsynchronous oscillations extended to direct-drive wind turbines and weak AC systems. However, the above invention cannot adapt to complex situations with multiple working conditions. Summary of the Invention

[0005] In response to the technical problem that traditional virtual synchronous machine technology is difficult to meet actual operation needs under complex conditions of multiple working conditions, the present invention proposes a virtual synchronous generator control method based on adaptive RBF-LADRC: 1. LADRC is introduced as the outer loop frequency controller of active power control in the virtual speed regulator of the virtual synchronous generator to adjust the active power output of the virtual synchronous generator; 2. An RBF neural network is simultaneously constructed to adaptively adjust the damping and inertia of the virtual synchronous generator; 3. A voltage-current dual closed-loop control system is designed, and the PI controller of the voltage loop is replaced by the LADRC controller, so that the grid-connected inverter adopts a virtual synchronous generator control with better performance. Compared with traditional control, the control strategy based on RBF-LADRC can significantly reduce the system overshoot, improve the active power tracking speed, and enhance the grid frequency stability. This strategy exhibits good anti-disturbance capability under both short-circuit faults and low-frequency disturbances in wind farms.

[0006] In order to achieve the above object, the technical solution of the present invention is implemented as follows: a virtual synchronous generator control method based on adaptive RBF-LADRC, the steps of which are as follows:

[0007] Step 1: Establish a mathematical model of the virtual synchronous generator and introduce LADRC as the outer loop frequency controller of the active power control loop in the virtual speed regulator of the virtual synchronous generator to regulate the active power output of the virtual synchronous generator;

[0008] Step 2: Design an RBF neural network and use it to adaptively adjust the virtual coefficients in the mathematical model of the virtual synchronous generator;

[0009] Step 3: Design the voltage outer loop in the voltage-current dual closed-loop control based on linear active disturbance rejection control, replace the PI controller with the LEDRC controller, and obtain the voltage-current dual closed-loop control system. The reference voltage obtained by coupling the virtual exciter and the virtual regulator is input into the voltage outer loop for parameter adjustment control of the dual closed loop.

[0010] Preferably, the virtual synchronous generator is a rotor motion equation of the virtual synchronous generator introduced into the droop control active power control loop, and the mathematical model of the virtual synchronous generator is:

[0011]

[0012] Where J and D are the virtual inertia coefficient and virtual damping coefficient of the virtual synchronous generator, ω and ω0 are the rotor angular velocity and reference angular velocity of the virtual synchronous generator, respectively. m and P e are the mechanical power and electromagnetic power of the virtual synchronous generator, Tm is the mechanical torque of the virtual synchronous generator, T e is the electromagnetic torque of the virtual synchronous generator, θ represents the angle of the synchronous generator rotor simulated by the virtual synchronous generator relative to the grid synchronous rotation reference axis, that is, the rotor position angle;

[0013] The mathematical model of the virtual speed regulator is:

[0014]

[0015] Where K p is the active frequency droop coefficient of the virtual speed regulator, P ref is the reference power of the virtual synchronous machine, P is the actual active power of the virtual synchronous machine, D P represents the virtual damping coefficient;

[0016] The expression of the virtual excitation regulator of the virtual synchronous generator is:

[0017]

[0018] Where Q is the actual reactive power output of the virtual synchronous machine, Q ref is the reactive power reference value of the virtual synchronous machine, E ref is the terminal voltage reference value, D q is the reactive voltage droop coefficient, K is the fixed inertia coefficient in the virtual exciter; E represents the actual value of the terminal voltage, and U represents the actual output voltage;

[0019] The virtual speed regulator integrates the rotor angular velocity ω to obtain the rotor position angle θ. The virtual excitation regulator obtains the output actual voltage U and the rotor position angle θ to generate the reference voltage U in the dq coordinate system. dref and U qref , reference voltage U dref and U qref Transmitted to the input end of the voltage outer loop control.

[0020] Preferably, the method of introducing LADRC as an outer-loop frequency controller of the active power control loop in the virtual speed regulator of the virtual synchronous generator to adjust the active power output of the virtual synchronous generator is as follows: the active power control loop of the virtual synchronous generator is designed using linear active disturbance rejection control, and the LADRC controller is designed as an additional controller in the active power control loop of the VSG. By utilizing the anti-disturbance capability of the linear active disturbance rejection control itself, the active power fluctuation is quickly responded to and controlled, and reliable active power compensation is provided to the system in real time.

[0021] After eliminating the non-target amount in the angular velocity deviation Δω=ω-ω0 through bandpass filtering, it is input into the LADRC controller. The compensation power P obtained in the active control link LThe reference power P of the virtual synchronous machine ref , the actual active power P of the virtual synchronous machine is simultaneously input into the virtual speed regulator.

[0022] Preferably, the expression of the LADRC controller is:

[0023]

[0024] Where z1, z2, z3 are the three outputs of the linear extended state observer, e1 and e2 are the inputs of the linear state error feedback control law, and k p 、k d is the control parameter in the linear state error feedback control law, u1 is the output of the linear state error feedback control law, b0 is the model parameter, P L is the compensation power of the additional active control link, β represents the linear parameter of the linear extended state observer gain, ω1 is the natural frequency of the linear state error feedback control law, and ω2 is the bandwidth of the linear extended state observer.

[0025] Preferably, the RBF neural network has the ability to fit any nonlinear function, and the virtual inertia coefficient J of the virtual synchronous generator is adjusted online, and the virtual damping coefficient D is adaptively adjusted by utilizing the damping ratio relationship between the virtual damping coefficient D and the virtual inertia coefficient J.

[0026] Preferably, a VSG small signal model is established to reasonably set the virtual damping coefficient D and the virtual inertia coefficient J of the virtual synchronous generator. The closed-loop transfer function of the active power loop of the VSG small signal model is:

[0027]

[0028] Where S E is the active reference value P ref and reactive reference value Q ref The associated value, P e (s) is the model output power, P ref (s) is the model reference power;

[0029] Natural oscillation angular frequency ω of the VSG small signal model n and the damping ratio ξ are:

[0030]

[0031] The allowed range of the damping ratio ξ of the virtual synchronous generator is: 0.8-0.85; according to the active response time t s <500ms, and combined with the time it takes for the system to reach steady state, we can get: The maximum cutoff frequency is set to 10 Hz, and the range of values of the virtual inertia coefficient J and the virtual damping coefficient D of the system stability is obtained through simulation.

[0032] Preferably, the method of adaptively adjusting the virtual coefficients in the mathematical model of the virtual synchronous generator using an RBF neural network is as follows: designing an RBF neural network to fit the nonlinear functional relationship between the virtual inertia coefficient J and the angular velocity ω, and adjusting the virtual damping coefficient D by fixing the damping ratio ξ; and the implementation method of the RBF neural network is as follows:

[0033] The input variables of the RBF neural network are the angular velocity increment Δω and the angular velocity differential dω / dt. The number of input layer nodes is 2. The output of the RBF neural network is the virtual inertia coefficient J. The number of output layer nodes is 1. The number of hidden layer nodes of the RBF neural network is 5. The hidden layer function of the RBF neural network is a Gaussian function, and the Gaussian function is used to determine the neuron output of the hidden layer.

[0034] Taking into account the range of the virtual inertia coefficient obtained by simulation, the virtual inertia coefficient J = f(w i );w i is the value of the neuron in the i-th hidden layer, i=1-5.

[0035] Preferably, the weights of the RBF neural network are adjusted by the gradient descent method: by selecting a suitable performance function The parameter adjustment formula of the RBF neural network is obtained as follows:

[0036]

[0037] Where η∈(0,1) is the learning rate, α is the momentum coefficient, and w i (t), w i (t+1) represents the weight of the i-th hidden layer in the t-th and t+1-th iterations, c i (t), c i (t+1) represents the center vector of the node in the tth and t+1th iterations of the i-th hidden layer, respectively, and b i (t), b i (t+1) represents the basis width parameter of the i-th hidden layer in the t-th and t+1-th iterations, respectively. ω0(k) and ω(k) are the rated angular velocity and actual angular velocity of the system.

[0038] Preferably, the voltage and current dual closed-loop control system in step 3 adopts linear active disturbance rejection control instead of the original PI controller to achieve zero-error regulation. Combined with the voltage outer loop control, the state equation under the dq coordinate can be obtained as follows:

[0039]

[0040] Among them, C f 、L f They are the filter capacitor in the voltage outer loop and the filter inductor in the current loop; i Ld 、i Lq They are the dq components of the filter inductor current, R l is the line resistance, u od 、u oq They are the dq components of the output voltage control quantity of the three-phase AC voltage, are the dq components of the AC voltage amplitude target, i od 、i oq is the dq component of the desired current value, which is used to control the active / reactive power, and ω is the rotor angular velocity.

[0041] Preferably, the transformation value of the three-phase AC voltage, that is, the dq component u of the output voltage control amount is obtained. od and u oq The second-order differential of is:

[0042]

[0043] Considering the duality of the state equation under dq coordinates, the d-axis is analyzed and the linear active disturbance rejection control object model of the grid-side inverter is obtained. The total disturbance f is: Where b is the actual control gain, b0 is the expected control gain, and u represents the control input;

[0044] The total disturbance f is estimated in real time through a linear extended state observer, and the output voltage control variable is compensated in the linear state error feedback control law.

[0045] Compared with the prior art, the present invention has the following beneficial effects: 1. Designing an RBF neural network and applying it to a virtual synchronous generator, adaptively updating the virtual inertia J of a traditional virtual synchronous generator, making corresponding changes as the system operating state changes, optimizing the inertia and damping parameters through the RBF neural network, and improving the dynamic response capability of the system; 2. Simultaneously, using the linear active disturbance rejection control technology to perform additional damping control on the active link of the virtual synchronous generator, further improving the system operation stability; 3. Applying the linear active disturbance rejection control technology to the voltage outer loop control, changing the PI control link to the linear active disturbance rejection control technology. Compared with PI control, the linear active disturbance rejection control technology has superior response speed and tracking effect, further improving the system anti-disturbance capability.

[0046] Starting from the perspective of renewable energy grid connection, this paper focuses on the power system stability issues faced by the development of new technologies and proposes an improved virtual synchronous generator control strategy based on adaptive RBF neural network and linear active disturbance rejection control (LADRC) to verify its feasibility in suppressing low-frequency oscillations caused by large-scale wind power grid connection. First, the paper analyzes the principle of virtual synchronous generator, establishes a mathematical model of virtual synchronous generator, and introduces LADRC as the outer loop frequency controller of active power control in the virtual synchronous generator virtual speed regulator to regulate the active power output of the virtual synchronous generator. At the same time, an RBF neural network is constructed to adaptively adjust the damping and inertia of the virtual synchronous generator. Secondly, to ensure a more stable output voltage of the virtual synchronous generator, a voltage-current dual closed-loop control system is designed. The PI controller of the voltage loop is replaced by the LADRC controller, so that the grid-connected inverter adopts the virtual synchronous generator control with better performance. The performance of different working modes is compared through simulation experiments, verifying that the LADRC controller has better performance in the voltage loop. Finally, the effectiveness of the virtual synchronous generator control strategy based on adaptive RBF-LADRC is verified by simulation and experiment on the Matlab / Simulink platform.

[0047] The virtual synchronous generator (VSG) improves power system stability by simulating the inertia and damping characteristics of synchronous generators. Considering that traditional VSG control strategies struggle to adapt to complex operating conditions, this paper introduces LADRC (Layered Detection Control) as the outer-loop frequency controller for active power control in the VSG's virtual speed regulator to optimize active power output. Furthermore, a voltage-current dual closed-loop control system is designed, in which the outer-loop voltage controller is replaced with a LADRC controller instead of a traditional PI controller, improving voltage regulation performance. Furthermore, a RBF neural network is used to adaptively adjust the virtual inertia coefficient J of the VSG. The inertia and damping parameters are optimized using the RBF neural network, and the weights are adjusted using the gradient descent method to enhance the system's dynamic response. Finally, the effectiveness of the proposed control strategy is verified through Matlab / Simulink simulations. Compared with traditional control, the RBF-LADRC-based control strategy significantly reduces system overshoot, improves active power tracking speed, and enhances grid frequency stability. This strategy demonstrates excellent disturbance immunity under wind farm short-circuit faults and low-frequency disturbances. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 It is a control flow chart of the present invention.

[0050] Figure 2 This is the control diagram of the virtual synchronous generator.

[0051] Figure 3 This is the model diagram of the virtual speed regulator.

[0052] Figure 4 This is the model diagram of the virtual excitation regulator.

[0053] Figure 5 This is the model diagram of the voltage and current dual closed-loop controller.

[0054] Figure 6 This is the control model diagram of the virtual synchronous machine.

[0055] Figure 7 This is a model diagram of the VSG small signal model of the present invention.

[0056] Figure 8 This is a curve diagram of the value range of virtual inertia and virtual damping of the present invention.

[0057] Figure 9 This is a flow chart of virtual inertia adjustment based on RBF neural network of the present invention.

[0058] Figure 10 This is the control principle diagram of the improved voltage-current dual closed-loop control system of the present invention.

[0059] Figure 11 This is a control principle diagram of the present invention with additional active power control.

[0060] Figure 12 This is a comparison diagram of the voltage outer loop control effect based on LADRC control of the present invention.

[0061] Figure 13 This is a comparison diagram of the control effect of the VSG active power control link based on LADRC control of the present invention.

[0062] Figure 14 This is a comparison diagram of the overall effect of the present invention, where (a) is the active power of the inverter and (b) is the frequency. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without creative work are within the scope of protection of the present invention.

[0064] like Figure 1 As shown in FIG, a virtual synchronous generator control method based on adaptive RBF-LADRC is implemented as follows:

[0065] Step 1: Establish a mathematical model of the virtual synchronous generator and introduce LADRC as the outer loop frequency controller of active power control in the virtual speed regulator of the virtual synchronous generator to adjust the active power output of the virtual synchronous generator.

[0066] The essence of virtual synchronous generator control is the improvement of droop control. Its characteristic is that the rotor motion equation of the virtual synchronous generator is introduced into the active control link of droop control, as shown in formula (1).

[0067]

[0068] Where J and D are the virtual inertia coefficient and virtual damping coefficient of the virtual synchronous generator, ω and ω0 are the generator rotor angular velocity and the grid rated angular velocity, respectively. m and P e are the mechanical power and electromagnetic power of the virtual synchronous generator, T m is the mechanical torque of the virtual synchronous generator, T e is the electromagnetic torque of the virtual synchronous generator. θ represents the angle of the synchronous generator rotor simulated by the virtual synchronous generator relative to the grid synchronous rotation reference axis, that is, the rotor position angle.

[0069] In the power system, inertia mainly comes from the rotor of the traditional synchronous generator. From the virtual inertia, we can see that the design concept of VSG technology comes from the synchronous generator. Therefore, the modeling of the virtual synchronous generator can be carried out on the basis of the synchronous generator theory. That is, the virtual synchronous generator technology introduces the inertia moment and damping coefficient parameters of the synchronous generator into the inverter control. By introducing virtual inertia and virtual damping, the sensitivity of the controller output frequency to load fluctuations is reduced. The schematic diagram of the virtual synchronous generator is shown in the figure below. Figure 2 shown.

[0070] Depend on Figure 2 It can be seen that the virtual synchronous generator control includes a virtual speed regulator, a virtual excitation controller, and a virtual synchronous machine regulation equation. Voltage and current dual closed-loop control is used to control and stabilize the voltage. The virtual synchronous machine regulation equation is shown in Equation (1), which simulates the operating characteristics of a traditional synchronous generator.

[0071] The function of a virtual speed regulator is similar to that of a traditional synchronous generator participating in the primary frequency regulation of the power grid. It plays a vital role in stabilizing the frequency of the power grid. Similar to the traditional speed regulator's workflow, the virtual speed regulator measures the system's operating angular velocity and compares it with the grid's rated angular velocity. When deviations occur, the generator speed is adjusted promptly. The virtual speed regulator is modeled, and its mathematical expression is shown in Equation (2).

[0072]

[0073] Where ω is the virtual angular velocity of the virtual rotor, ω0 is the reference angular velocity, and K p is the active frequency droop coefficient of the virtual speed regulator, P ref is the reference power of the virtual synchronous machine. When the virtual speed regulator is discussed separately, P is the actual mechanical power of the virtual synchronous machine. P Represents the virtual damping coefficient. The simulation model of the virtual speed regulator can be obtained from the principle as follows: Figure 3 As shown in Figure 1, J is the virtual inertia coefficient, and 1 / s is the integral operator used in the control system to convert the time domain signal into a complex frequency domain signal.

[0074] For the design of virtual exciter, different from traditional synchronous generator, the virtual synchronous generator model does not directly introduce excitation current, but adjusts the reactive power and terminal voltage of VSG output by directly adjusting the no-load electromotive force. The simulation model of virtual excitation regulator is as follows: Figure 4 As shown. The expression obtained through the model diagram of the virtual excitation regulator is:

[0075]

[0076] Where Q is the actual reactive power output of the virtual synchronous machine, Q ref is the reactive power reference value of the virtual synchronous machine, E ref is the terminal voltage reference value, D q is the reactive voltage droop coefficient, K is the fixed inertia coefficient in the virtual exciter, E represents the actual value of the terminal voltage, and U represents the actual output voltage.

[0077] The purpose of voltage and current dual closed-loop control is to make the output voltage more stable. The voltage loop is used as the outer loop control, and the current loop is used as the inner loop control, which can improve the system response speed. The voltage and current dual closed-loop control expression is:

[0078]

[0079] Where K cp , K ci They are the proportional coefficient and integral coefficient of the PI controller in the current loop; K vp , K viare the proportional coefficient and integral coefficient of the PI controller in the voltage loop; C f 、L f They are the filter capacitor in the voltage loop and the filter inductor in the current loop. d and u q Is the output voltage control quantity. I d * and I q * The expected current values of the d-axis and q-axis calculated by the voltage error are used to control the active / reactive power. d 、V q The three-phase AC voltage is converted to the dq coordinate system through Park transformation, which represents the components of the voltage in the dq coordinate system. d * and V q * is the DC bus voltage or AC voltage amplitude target. Ld 、i Lq are the dq components of the filter inductor current respectively. The voltage and current dual closed-loop controller model constructed based on the expression is as follows: Figure 5 shown.

[0080] By analyzing and modeling each part of the virtual synchronous generator, the final virtual synchronous machine control model is as follows: Figure 6 As shown, the combined generation module generates a reference voltage U according to the output actual voltage U and θ dref and U qref , reference voltage U dref and U qref The data is transmitted to the input of the voltage outer loop control. This invention improves upon the control of virtual synchronous generators by, on the one hand, replacing and optimizing the voltage and current dual closed-loop control with active disturbance rejection control, and on the other hand, optimizing the active power control of the virtual synchronous generator. This multi-optimization strategy aims to enhance the voltage regulation and active power control capabilities of the virtual synchronous generator, thereby strengthening its role and application in renewable energy generation.

[0081] The active power control link of the virtual synchronous generator is designed using linear active disturbance rejection control. The LADRC controller is designed as an additional controller in the active power control loop of the VSG. Its purpose is to quickly respond to and control the active power fluctuation through the anti-disturbance capability of the linear active disturbance rejection control technology itself, and provide reliable active power compensation for the system in real time. Figure 11 As shown, combined with the linear active disturbance rejection control principle, we can get:

[0082]

[0083] Where z1, z2, z3 are the three outputs of the linear extended state observer, e1 and e2 are the inputs of the linear state error feedback control law, and k p 、k d is the control parameter in the linear state error feedback control law, u1 is the output of the linear state error feedback control law, b0 is an effective substitute for the accurate high-frequency gain value b and is widely used in the design of controller model parameters. L is the compensation power of the additional active control link, β represents the linear parameter of the LESO gain, ω1 is the natural frequency of the LESF (Linear State Error Feedback) control law, and ω2 is the bandwidth of the LESO (Linear Extended State Observer). The LESO design endows the system with strong parameter adaptability. Combined with the structural characteristics and control effectiveness of linear active disturbance rejection control, the system can respond more quickly to sudden disturbances or events, thereby achieving better control results.

[0084] The active power control link of the additional anti-disturbance control designed by the present invention adjusts the output active power of the virtual synchronous generator by introducing linear anti-disturbance control. The goal of the additional active power control is to enhance the system's dynamic response capability to power disturbances, suppress frequency fluctuations, introduce compensation terms outside the original control loop, and dynamically adjust the actual mechanical power P of the virtual synchronous machine. The virtual speed regulator adjusts the actual active power P. Specific implementation: detect power or frequency oscillation signals, and generate damping torque through phase compensation. The angular velocity deviation Δω=ω-ω0 is used as the input signal of the additional control. The bandpass filter of the designed filtering link is used to extract the components of the low-frequency oscillation mode. The bandpass filter is constructed by designing low-pass filtering and high-pass filtering to eliminate the interference of non-target quantities in the angular velocity deviation Δω on the control strategy.

[0085] Step 2: Design an RBF neural network and use it to adaptively adjust the virtual inertia in the mathematical model of the virtual synchronous generator. Adaptively adjust the virtual damping by using the damping ratio relationship between the virtual damping and the virtual inertia.

[0086] The present invention combines linear active disturbance rejection control with RBF neural network and applies it to the control of virtual synchronous generator. The virtual inertia J is quickly learned and adaptively adjusted by RBF neural network, while linear active disturbance rejection control is used to improve and optimize the structure of voltage outer loop control and virtual synchronous generator active link. The overall solution of virtual synchronous generator is as follows: Figure 1 As shown. Figure 1In the improvements shown, the RBF neural network is used for adaptive virtual inertia adjustment; the linear active disturbance rejection control technology applied in the voltage and current dual closed loop plays a role in zero-difference regulation and power stabilization, increasing the stability of virtual synchronous control; the linear active disturbance rejection control applied in the active power control link of the virtual synchronous generator serves as an outer loop controller to regulate the output active power.

[0087] The design of the present invention adopts RBF neural network to improve the active power control link of the virtual synchronous generator. The ability of RBF neural network to fit any nonlinear function is used to adjust the virtual inertia coefficient J online, and the virtual damping coefficient D is further adaptively adjusted by utilizing the damping ratio relationship between virtual damping and virtual inertia.

[0088] By establishing a VSG small signal model, the damping parameters (virtual damping coefficient D) and inertia parameters (virtual inertia coefficient J) of the virtual synchronous generator can be reasonably set. The VSG small signal model is as follows: Figure 7 As shown. The closed-loop transfer function of the active control link or active power loop of the VSG small signal model is:

[0089]

[0090] Where S E With active reference value P ref and reactive reference value Q ref When certain, S E is a constant. E is the active reference value P ref and reactive reference value Q ref The associated value, P e (s) is the model output power, P ref (s) is the model reference power.

[0091] The natural oscillation angular frequency ω of the VSG small signal model can be obtained n and the damping ratio ξ are:

[0092]

[0093] Regarding the setting of the virtual inertia coefficient J and virtual damping coefficient D, the optimal damping ratio for virtual synchronous generators permitted by the national standard GB / T 38983.1-2020, Part 1 of Virtual Synchronous Machines (hereinafter referred to as the national standard) is as follows: 1: 0.707 ≤ ξ < 1. In adaptive damping control, a fixed damping ratio is set to further adjust the damping coefficient as the inertia coefficient changes. Considering that the damage rate of power electronic equipment is often related to power overshoot, the fixed damping ratio should be selected as large as possible within the specified range. However, considering that a larger damping ratio slows down the response speed, a damping ratio ξ between 0.8 and 0.85 is ultimately the most suitable.

[0094] According to the national standard, the active response time t s <500ms, and combined with the time it takes for the system to reach steady state, we can get condition 2:

[0095]

[0096] In order to reduce the influence of the power loop on the voltage loop, the maximum cutoff frequency is set to 10Hz. Therefore, combined with the damping ratio ξ between 0.8-0.85 in condition 1, condition 2, and the cutoff frequency setting, the range of the virtual inertia and virtual damping values of the system obtained by the simulation program is as follows: Figure 8 As shown in FIG, what is obtained is the value range of the virtual inertia coefficient J and the virtual damping coefficient.

[0097] In this design phase, the nonlinear functional relationship between the virtual inertia coefficient J and the instantaneous angular velocity ω is fitted by designing and applying the RBF neural network, and the virtual damping coefficient D is further adjusted by fixing the damping ratio ξ. The number of nodes in the input and output layers of the designed RBF neural network is Figure 9 As shown, it can be seen that the input variables are 2, namely the angular velocity increment Δω and the angular velocity differential dω / dt.

[0098] like Figure 9 As shown in the figure, the input of the RBF neural network is the angular velocity increment Δω and the angular velocity differential dω / dt, the output is the virtual inertia coefficient J, and a 2-5-1 structure is selected, with 5 hidden layer nodes. The hidden layer function of the RBF neural network is a Gaussian function, which is used as the output of the neuron and the activation function of the hidden layer of the RBF neural network. With its local response characteristics, mathematical smoothness, parameter flexibility and anti-noise ability, it has shown significant advantages in nonlinear modeling, pattern recognition and dynamic system control. The Gaussian function is

[0099]

[0100] Among them, x is the input of RBF neural network, c ijrepresents the mean of the membership function of the jth fuzzy set of the RBF neural network in the i-th input variable, b j Represents the basis width vector.

[0101] The output of the hidden layer of the RBF neural network is:

[0102]

[0103] Among them, w li are the weights of the nodes in the i-th hidden layer and the l-th output layer, w1, w2, w5 are the weights of each hidden layer, and g1, g2, g5 represent the neuron outputs of the hidden layer.

[0104] Finally, considering the Figure 8 The range of the virtual inertia coefficient is set as shown in the figure. The final output virtual inertia coefficient J of the RBF neural network is obtained through the activation function f(x) of the output layer:

[0105]

[0106] Where n is the upper limit of inertia, usually 1 or 2.

[0107] The method used in the present invention to adjust the weights of the RBF neural network is the gradient descent method. By selecting an appropriate performance function E(t), the parameter adjustment formula of the RBF neural network is obtained as follows:

[0108]

[0109] In the formula, η∈(0,1) is the learning rate, which is used to accelerate the convergence of the network. In the actual design, the learning rate is 0.5. α is the momentum coefficient, which controls the retention ratio of historical parameter values. i (t), w i (t+1) represents the weight of the i-th hidden layer in the t-th and t+1-th iterations, c i (t), c i (t+1) represents the center vector of the node in the tth and t+1th iterations of the i-th hidden layer, respectively, and b i (t), b i (t+1) represents the base width parameter of the i-th hidden layer in the t-th and t+1-th iterations respectively. The performance function E(t) is calculated based on the formula of angular velocity ω. ω0(k) and ω(k) represent the rated angular velocity and actual angular velocity of the system respectively. ω0(k) and ω(k) are used to explain the selection of E(t) function.

[0110] The virtual inertia mentioned in formula (7) is related to the virtual damping. When the damping ratio remains unchanged, the changes in J and D are closely related.

[0111] Step 3: Design the voltage outer loop based on linear active disturbance rejection control, replace the PI controller with the LEDRC controller, and obtain a voltage and current dual closed-loop control system. The reference voltage U is obtained by coupling the virtual exciter and the virtual regulator. dqref After inputting the voltage outer loop, double closed loop parameter adjustment control is performed.

[0112] The voltage outer loop is designed with linear active disturbance rejection control to overcome the shortcomings of conventional voltage and current dual closed-loop control, which has a long dynamic response time and is insensitive to overall disturbance changes. The design principle diagram of the dual closed-loop controller with linear active disturbance rejection control is shown in the figure below. Figure 10 shown.

[0113] Figure 10 A new control strategy for the voltage loop is presented in this paper. Linear active disturbance rejection control technology is used to replace the original PI controller to achieve zero-error regulation. The designed controller is compared with the original PI controller to verify that it has a stronger control effect. Combining the new outer loop control with the system structure, the state equation in dq coordinates is:

[0114]

[0115] Among them, i Ld 、i Lq is the dq component of the filter inductor current. l is the line resistance, u od 、u oq is the output voltage control quantity, is the DC bus voltage or AC voltage amplitude target, i od 、i oq The dq axis expected current value calculated by the voltage error is used to control the active / reactive power. Figure 10 It is obtained by performing differential changes based on the formula.

[0116] Combining formula (13) and formula (14), we can get the transformation value u of the three-phase AC voltage: od and u oq The second-order differential of is:

[0117]

[0118] Considering the duality of the state equation in dq coordinates, the d-axis is analyzed, and finally the linear active disturbance rejection control object model of the grid-side inverter can be obtained. The total disturbance f is defined as:

[0119]

[0120] Where b is the actual control gain, and b0 is the desired control gain, obtained experimentally. u represents the control input, which is the actual control variable output by the controller. The core of linear active disturbance rejection control is to estimate the total disturbance f in real time using a linear extended state observer and compensate for it in the control law.

[0121] In order to verify the feasibility and effectiveness of the research strategy proposed in this paper, the following Figure 6 The simulation model of a single-machine grid-connected virtual synchronous generator system with wind power is shown. Based on the experimental results under different operating conditions, the effectiveness of the proposed control strategy is analyzed and verified.

[0122] First, in order to verify the feasibility and effectiveness of the voltage control strategy based on linear active disturbance rejection, the active power of the virtual synchronous generator was experimentally adjusted, and the active power of the virtual synchronous generator was increased by 50×10 3 W, the simulation time is 1s, and the control strategy results are compared as shown in the figure below. Figure 12 shown.

[0123] like Figure 12 The inverter's active power output characteristics are shown in the figure. By comparing the control effects of a PI controller and a linear active disturbance rejection controller, we find that the traditional PI controller exhibits large output power fluctuations and a long stabilization time. In contrast, the adoption of the linear active disturbance rejection control strategy significantly reduces system power fluctuations, reduces system overshoot from 13% to 7.1%, accelerates tracking and stabilization speeds, and significantly improves overall control performance.

[0124] Furthermore, in order to verify the effectiveness of the proposed active power control strategy based on linear active disturbance rejection control, a corresponding simulation model was built for experimental verification. The scenario was set as follows: a three-phase short circuit fault occurred at the grid connection point of the wind farm under constant wind speed, with the starting time being 0.2s and the duration being set to 30ms. The output frequency of the virtual synchronous generator control link is as follows: Figure 13 The simulation results show that the system was operating in a stable state 0.2s before the three-phase short-circuit fault occurred. In the model without the additional control link, the output frequency oscillation amplitude of the virtual synchronous generator control link was large and converged slowly. In contrast, the simulation results obtained after the additional control link was added show that the proposed control strategy can effectively suppress the oscillation of the inverter output power, reducing the overshoot while significantly improving the overall response speed of the system.

[0125] Table 1 Basic system parameters

[0126]

[0127] To verify that the proposed adaptive RBF-LADRC-based virtual synchronous generator control strategy has good ability to cope with system dynamic disturbances in the power system with wind power grid connection, the modeling was applied to the simulation system and corresponding simulation tests were conducted. The basic parameters of the designed system are shown in Table 1. In addition, since the wind turbine generator set is always set to a constant wind speed operation state, a 690V DC power supply is used instead to further simplify the simulation model.

[0128] In order to verify the overall effectiveness of the proposed control strategy, the reference power P is set in Simulink at 0.5s. r ef A low-frequency continuous disturbance with an amplitude of 20kW and a disturbance frequency of 1Hz occurs at the location and lasts for 1.5s. The dynamic characteristics of the active power and system frequency are as follows: Figure 14 As shown. Figure 14 As shown, when the reference power P ref When subjected to a 1Hz disturbance, the inverter's active power output also fluctuates. The disturbance begins at t = 0.5s. The blue line in Figure (a) represents the conventional virtual synchronous generator technology without the RBF-LADRC control strategy, while the red line represents the VSG response after adopting this control strategy. As can be seen from the figure, it is difficult to effectively suppress the fluctuation of active power when relying solely on traditional virtual inertia and virtual damping. In the time period of 0.5s to 0.8s, the active power output corresponding to the blue line deviates significantly from the rated value, while the red line shows a faster recovery ability and quickly approaches the rated power. From 0.8s to 2.0s, both control strategies gradually approach the rated output under their own regulation. However, after the disturbance ends, the RBF-LADRC-based VSG relies on the RBF neural network to adaptively adjust the virtual inertia coefficient J. Combined with the linear active disturbance rejection control to accurately compensate for the active control link and optimize the voltage outer loop control, the system can suppress the oscillation of frequency and active power more quickly when subjected to disturbances, showing a better control effect. Figure 14 The simulation results in (b) also show the same Figure 14 The same control effect as (a) is achieved, and effective control capabilities are demonstrated within the corresponding time period. Figure 14 Figure (a) and Figure (b) are different parameter performances of the same experiment, and the overall performance is the same.

[0129] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A virtual synchronous generator control method based on adaptive RBF-LADRC, characterized in that: The steps are as follows: Step 1: Establish a mathematical model of the virtual synchronous generator and introduce LADRC as the outer loop frequency controller of the active power control loop in the virtual speed regulator of the virtual synchronous generator to regulate the active power output of the virtual synchronous generator; Step 2: Design an RBF neural network and use it to adaptively adjust the virtual coefficients in the mathematical model of the virtual synchronous generator; Step 3: Design the voltage outer loop in the voltage-current dual closed-loop control based on linear active disturbance rejection control, replace the PI controller with the LEDRC controller, and obtain the voltage-current dual closed-loop control system. The reference voltage obtained by coupling the virtual exciter and the virtual regulator is input into the voltage outer loop for parameter adjustment control of the dual closed loop.

2. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 1 is characterized in that: The virtual synchronous generator is introduced into the rotor motion equation of the virtual synchronous generator in the droop control active power control loop. The mathematical model of the virtual synchronous generator is: Where J and D are the virtual inertia coefficient and virtual damping coefficient of the virtual synchronous generator, ω and ω0 are the rotor angular velocity and reference angular velocity of the virtual synchronous generator, respectively. m and P e are the mechanical power and electromagnetic power of the virtual synchronous generator, T m is the mechanical torque of the virtual synchronous generator, T e is the electromagnetic torque of the virtual synchronous generator, θ represents the angle of the synchronous generator rotor simulated by the virtual synchronous generator relative to the grid synchronous rotation reference axis, that is, the rotor position angle; The mathematical model of the virtual speed regulator is: Where K p is the active frequency droop coefficient of the virtual speed regulator, P ref is the reference power of the virtual synchronous machine, P is the actual active power of the virtual synchronous machine, D P represents the virtual damping coefficient; The expression of the virtual excitation regulator of the virtual synchronous generator is: Where Q is the actual reactive power output of the virtual synchronous machine, Q ref is the reactive power reference value of the virtual synchronous machine, E ref is the terminal voltage reference value, D q is the reactive voltage droop coefficient, K is the fixed inertia coefficient in the virtual exciter; E represents the actual value of the terminal voltage, and U represents the actual output voltage; The virtual speed regulator integrates the rotor angular velocity ω to obtain the rotor position angle θ. The virtual excitation regulator obtains the output actual voltage U and the rotor position angle θ to generate the reference voltage U in the dq coordinate system. dref and U qref , reference voltage U dref and U qref Transmitted to the input end of the voltage outer loop control.

3. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 1 or 2, characterized in that: The method for introducing LADRC as an outer-loop frequency controller of an active power control loop in a virtual speed regulator of a virtual synchronous generator to adjust the active power output of the virtual synchronous generator is as follows: the active power control loop of the virtual synchronous generator is designed using linear active disturbance rejection control, and the LADRC controller is designed as an additional controller in the active power control loop of the VSG. By leveraging the anti-disturbance capability of the linear active disturbance rejection control, the active power fluctuation is quickly responded to and controlled, and reliable active power compensation is provided to the system in real time. After eliminating the non-target amount in the angular velocity deviation Δω=ω-ω0 through bandpass filtering, it is input into the LADRC controller. The compensation power P obtained in the active control link L The reference power P of the virtual synchronous machine ref , the actual active power P of the virtual synchronous machine is simultaneously input into the virtual speed regulator.

4. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 3 is characterized in that: The expression of the LADRC controller is: Where z1, z2, z3 are the three outputs of the linear extended state observer, e1 and e2 are the inputs of the linear state error feedback control law, and k p 、k d is the control parameter in the linear state error feedback control law, u1 is the output of the linear state error feedback control law, b0 is the model parameter, P L is the compensation power of the additional active control link, β represents the linear parameter of the linear extended state observer gain, ω1 is the natural frequency of the linear state error feedback control law, and ω2 is the bandwidth of the linear extended state observer.

5. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 2 or 3, characterized in that: By utilizing the ability of RBF neural network to fit any nonlinear function, the virtual inertia coefficient J of the virtual synchronous generator is adjusted online, and the virtual damping coefficient D is adaptively adjusted by utilizing the damping ratio relationship between the virtual damping coefficient D and the virtual inertia coefficient J.

6. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 5, characterized in that: A VSG small signal model is established to reasonably set the virtual damping coefficient D and virtual inertia coefficient J of the virtual synchronous generator. The closed-loop transfer function of the active power loop of the VSG small signal model is: Where S E is the active reference value P ref and reactive reference value Q ref The associated value, P e (s) is the model output power, P ref (s) is the model reference power; Natural oscillation angular frequency ω of the VSG small signal model n and the damping ratio ξ are: The allowed range of the damping ratio ξ of the virtual synchronous generator is: 0.8-0.85; according to the active response time t s <500ms, and combined with the time it takes for the system to reach steady state, we can get: The maximum cutoff frequency is set to 10 Hz, and the range of values of the virtual inertia coefficient J and the virtual damping coefficient D of the system stability is obtained through simulation.

7. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 6, characterized in that: The method for adaptively adjusting the virtual coefficients in the mathematical model of the virtual synchronous generator using the RBF neural network is as follows: the RBF neural network is designed to fit the nonlinear functional relationship between the virtual inertia coefficient J and the angular velocity ω, and the virtual damping coefficient D is adjusted by fixing the damping ratio ξ; the implementation method of the RBF neural network is as follows: The input variables of the RBF neural network are the angular velocity increment Δω and the angular velocity differential dω / dt. The number of input layer nodes is 2. The output of the RBF neural network is the virtual inertia coefficient J. The number of output layer nodes is 1. The number of hidden layer nodes of the RBF neural network is 5. The hidden layer function of the RBF neural network is a Gaussian function, and the Gaussian function is used to determine the neuron output of the hidden layer. Taking into account the range of the virtual inertia coefficient obtained by simulation, the virtual inertia coefficient J = f(w i );w i is the value of the neuron in the i-th hidden layer, i=1-5.

8. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 7, characterized in that: Adjust the weights of the RBF neural network by gradient descent method: by selecting the appropriate performance function The parameter adjustment formula of the RBF neural network is obtained as follows: Where η∈(0,1) is the learning rate, α is the momentum coefficient, and w i (t), w i (t+1) represents the weight of the i-th hidden layer in the t-th and t+1-th iterations, c i (t), c i (t+1) represents the center vector of the node in the tth and t+1th iterations of the i-th hidden layer, respectively, and b i (t), b i (t+1) represents the basis width parameter of the i-th hidden layer in the t-th and t+1-th iterations, respectively. ω0(k) and ω(k) are the rated angular velocity and actual angular velocity of the system.

9. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 4, characterized in that: The voltage and current dual closed-loop control system in step 3 adopts linear active disturbance rejection control to replace the original PI controller to achieve zero-error regulation. Combined with the voltage outer loop control, the state equation under the dq coordinate can be obtained as follows: Among them, C f 、L f They are the filter capacitor in the voltage outer loop and the filter inductor in the current loop; i Ld 、i Lq They are the dq components of the filter inductor current, R l is the line resistance, u od 、u oq They are the dq components of the output voltage control quantity of the three-phase AC voltage, are the dq components of the AC voltage amplitude target, i od 、i oq is the dq component of the desired current value, which is used to control the active / reactive power, and ω is the rotor angular velocity.

10. The virtual synchronous generator control method based on adaptive RBF-LADRC according to claim 9, characterized in that: The transformation value of the three-phase AC voltage is obtained, that is, the dq component u of the output voltage control quantity od and u oq The second-order differential of is: Considering the duality of the state equation under dq coordinates, the d-axis is analyzed and the linear active disturbance rejection control object model of the grid-side inverter is obtained. The total disturbance f is: Where b is the actual control gain, b0 is the expected control gain, and u represents the control input; The total disturbance f is estimated in real time through a linear extended state observer, and the output voltage control variable is compensated in the linear state error feedback control law.

Images