Modular multilevel energy storage system virtual synchronous generator robust control method and system

Abstract

This invention provides a robust control method and system for a virtual synchronous generator in a modular multilevel energy storage system. The robust control method includes the following steps: (1) establishing a VSG control system for energy storage; (2) constructing a mathematical model of the modular multilevel energy storage system; (3) establishing a passive control current inner loop architecture and a sliding mode control current inner loop structure, and designing a PBC-SMC current inner loop control structure with differential mode voltage as the control quantity by combining the two structures; (4) dynamically adjusting and optimizing the network parameters through a BP neural network, and finally outputting the optimized convergence rate coefficient to achieve online optimization of the coefficient to improve the current inner loop control performance of the converter. This invention effectively improves the response speed, transient response performance, steady-state control accuracy, and robust performance of the current inner loop, thereby improving the converter's ability to quickly and actively support the power, frequency, and voltage of the grid connected to the wind and solar new energy generator units.

Description

Technical Field

[0001] This invention relates to the field of robust generator control technology, and in particular to a robust control method and system for a virtual synchronous generator in a modular multilevel energy storage system. Background Technology

[0002] The rapid integration of new energy power plants, such as wind, solar, energy storage, flexible DC, and DC power, into the power grid has led to stability issues and performance degradation problems in new power systems with a high proportion of new energy, including active power fluctuations, reactive power deficits, frequency disturbances, and voltage exceeding limits. This has increased the risk to the safe and stable operation of new power systems. Energy storage devices have emerged as an effective technical means to address these issues, and their practical applications in photovoltaic power plants, offshore wind power platforms, and urban power grids have been promoted. They can suppress random fluctuations in the output power of wind and solar generators and regulate the frequency and voltage stability of regional power grids. By optimizing the charging and discharging sequence of power plants, the power revenue of the grid, the utilization rate of green new energy, and the operating efficiency of the power plants themselves can be improved, making them of significant research and application value.

[0003] Control strategies are crucial for the performance of energy storage power equipment, directly impacting the stability of the equipment itself, the stability of power, frequency, and voltage at the point of common coupling (PCC), and the curtailment rate of wind and solar power units. Therefore, it is essential to research control methods for energy storage systems to improve their power, frequency, and voltage regulation performance, optimize the external characteristics of the converter's operation control, and meet the real-time operation and regulation requirements of the PCC. Energy storage power stations at 6.6kV and 35kV voltage levels often employ a modular multi-level converter (MMC) topology. Existing power stations primarily use constant power control strategies and AC voltage-reactive power control strategies to achieve stable grid-connected operation of the energy storage system converter, deliver constant power to the grid, and maintain stable AC bus voltage. However, both strategies lack the ability to rapidly and actively support the power, frequency, and voltage fluctuations caused by changes in the output of wind and solar power units. In response, researchers and technicians at home and abroad have proposed droop control, virtual synchronous generator (VSG) control and their improvement strategies. Such strategies can simulate the control operation characteristics of traditional synchronous generators, such as primary frequency regulation, damped oscillation, and reactive power support, effectively improving the regulation performance of grid-connected converters in energy storage systems. This enables new power systems to connect more wind and solar new energy units and achieve safe, reliable and stable operation.

[0004] Regarding the control technology of modular multilevel energy storage power systems, existing technologies do not employ robust control theory, nor do they study robust control methods and disturbance and fault conditions. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a robust control method and system for a virtual synchronous generator in a modular multilevel energy storage system. Considering the poor robustness of traditional current inner-loop control strategies under load disturbances and fault disturbances, a VSG control system for energy storage is established. A passive-based control (PBC)-sliding mode control (SMC) current inner-loop control structure is invented, improving the parameter disturbance rejection performance of the current inner loop. An optimized PBC-SMC current inner-loop structure is constructed, which can improve the power, frequency, and voltage regulation performance and robustness of the energy storage system.

[0006] The technical solution provided by this invention: a robust control method for a virtual synchronous generator in a modular multilevel energy storage system, comprising the following steps:

[0007] (1) Establish an energy storage VSG control system;

[0008] (2) Construct a mathematical model for a modular multilevel energy storage system;

[0009] (3) Establish a passive control current inner loop architecture, taking into account the injection damping parameters and control parameter disturbances, establish a sliding mode control current inner loop structure considering the two disturbances, and combine the two structures to design a PBC-SMC current inner loop control structure with differential mode voltage as the control quantity.

[0010] (4) Input the output current deviation vector and its derivative into the BP neural network, and perform dynamic adjustment and network parameter optimization through the BP neural network. Finally, output the optimized convergence coefficient to realize online optimization of the coefficient to improve the converter current inner loop control performance.

[0011] Furthermore, step (1) includes: designing the topology of the MMC distributed energy storage system as the hardware circuit of the energy storage system; firstly, deploying distributed energy storage batteries as the basic architecture of the energy storage system; the energy storage batteries as the core components of the sub-modules; adding DC / DC conversion modules to realize the forward and reverse transmission of electrical energy; the battery body adopts lithium batteries; through the charging / discharging of lithium batteries, the control target of the interaction between DC-side power and AC-side power is realized; the transmission relationship between AC-side power, DC-side power, and lithium battery power is shown in equation (1).

[0012] (1)

[0013] In the formula, PDC The value is zero because the DC voltage is zero, and the power transferred from the DC side to the lithium battery is zero; P AC This represents the power transmitted by the energy storage system to the AC side, i.e., the grid side; P Li, bat This indicates the output power of the lithium battery.

[0014] Furthermore, step (2) includes:

[0015] The inductor voltage equation of the bridge arm was established, as shown in equation (2).

[0016] (2)

[0017] In the formula, For bridge arm inductance; This refers to the voltage of the upper bridge arm; for; For the upper bridge arm voltage of each phase; For the bridge arm resistance; For the upper arm current of each phase; This represents the lower bridge arm current for each phase; This refers to the voltage of the lower bridge arm; This represents the lower bridge arm voltage for each phase;

[0018] Subtracting the second formula from the first formula in equation (2) gives equation (3).

[0019] (3)

[0020] Observing equation (2), we know that i ku -i kl Corresponding phase current i k , and v p +v n Approximately zero, let the differential mode voltage v kdifm as follows:

[0021] (4)

[0022] Analyze equation (3), list the equations for the three phases ABC, transform them, and substitute them into equation (4) to obtain equation (5).

[0023] (5)

[0024] In the formula, , , It is a three-phase current; , , It is a three-phase voltage; , , This is the three-phase differential mode voltage;

[0025] Based on the Park transformation formula, equation (5) is transformed from the three-phase stationary coordinate system to the two-phase synchronous rotating coordinate system, resulting in equation (6).

[0026] (6)

[0027] In the formula, i d Indicates the d-axis quantity of the converter output current; i q This represents the q-axis quantity of the converter output current; v d This represents the d-axis quantity of the converter output voltage; v q This represents the q-axis quantity of the converter output voltage; v ddifm This represents the d-axis quantity of the differential-mode voltage; v qdifm This represents the q-axis quantity of the differential-mode voltage; The angular frequency of the power grid;

[0028] By transforming equation (6) using the Laplace transform principle, we obtain the transfer function formula that accurately describes the output current, as shown in equation (7).

[0029] (7)

[0030] In the formula, For the Laplace operator; For the Laplace transform of the d-axis current; Laplace transform of the q-axis current; Laplace transform of the d-axis voltage; Laplace transform of the q-axis voltage; Laplace transform of the d-axis differential-mode voltage; This is the Laplace transform of the q-axis differential-mode voltage.

[0031] Furthermore, the establishment of the passive control current inner loop architecture in step (3) includes:

[0032] An Euler-Lagrange mathematical model describing the dynamic characteristics of the inner current loop voltage is proposed, as shown in equation (8).

[0033] (8)

[0034] For the inner-loop current control structure, the converter output current d and q-axis quantities i are selected. d i q Let x be the state variable vector, and define the converter output current d and the q-axis reference value i. dref i qref The reference vector for the state variables is x. ref And proposes the converter output current deviation vector x eThe mathematical description method is shown in equation (9).

[0035] (9)

[0036] To achieve the goal of PBC, control damping is injected into the mathematical model shown in equation (9), and the positive definite matrix composed of the injected control damping is... ,in , To control the positive definite coefficients of the damping, the injection control damping dissipation term is established as follows:

[0037] (10)

[0038] In the formula, R d The matrix representing the injection control damping dissipation term.

[0039] By combining equations (9) and (8) and substituting them into equation (10), the formula is transformed into a scalar equation system using matrices and vectors. This constructs a mathematical model of the PBC current inner loop structure with differential-mode voltage as the control variable, as shown in equation (11).

[0040] , (11).

[0041] Furthermore, the sliding mode control current inner loop structure in step (3) includes:

[0042] For the design of the SMC current inner loop structure, the sliding surface of the SMC is first designed, as shown in equation (12).

[0043] (12)

[0044] In the formula, For the d-axis sliding surface, It is the q-axis sliding surface;

[0045] By combining the sign function sgn(s) and the linear reaching law, the chattering that exists during the convergence of the sliding surface is effectively reduced. The designed reaching rate is shown in Equation (13).

[0046] (13)

[0047] In the formula, λ1, μ1, λ2, and μ2 represent the approach rate coefficients that are greater than zero, and their values ​​determine the convergence speed of the sliding surface and the magnitude of chattering.

[0048] By combining equations (7), (12), and (13), and transforming the formula into a scalar equation system, a mathematical model of the SMC current inner loop structure with differential mode voltage as the control quantity is constructed, as shown in equation (14).

[0049] (14)

[0050] Furthermore, in step (3)

[0051] Combining equations (11) and (14), a mathematical model for the converter output current deviation vector is constructed, as shown in equation (15).

[0052] (15)

[0053] By combining equations (11) and (15), a mathematical model of the PBC-SMC current inner loop structure with differential mode voltage as the control quantity is constructed, as shown in equation (16).

[0054] , (16).

[0055] Furthermore, in step (4), the output current deviation vector and its derivative are input into the BP neural network, and the network outputs the convergence coefficients μ1 and μ2. By optimizing μ1 and μ2 online, the speed and robustness of the converter's inner current tracking response are improved, and chatter is reduced.

[0056] x1 and x2 are network inputs, w 11 ~w 23 The weights from the input layer to the intermediate layer correspond to the output of the intermediate layer, as shown in Equation (17).

[0057] (17)

[0058] In the formula, The output of the j-th neuron in the hidden layer.

[0059] The m-function uses a sigmoid function to perform the nonlinear amplification task, transforming the data from −∞ to +∞ into the range of 0 to 1. The m-function is shown in equation (18).

[0060] (18)

[0061] w'1~w'3 are the weights from the intermediate layer to the output layer, corresponding to the output layer output, as shown in equation (19).

[0062] (19)

[0063] The n function uses a piecewise function to transform the data from −∞ to +∞ into the range −1 to 1. The n function is shown in equation (20).

[0064] (20)

[0065] Propose the expected value μ of the convergence coefficientref The mathematical description of the deviation from the online output μ is shown in Equation (21).

[0066] , (twenty one)

[0067] Analyzing equation (21), the total deviation is constructed as shown in equation (22).

[0068] , (twenty two)

[0069] The steepest descent method is used to optimize the adjustment of the weights. First, the gradient of the total deviation e(k) with respect to the weights w'1~w'3 is calculated, as shown in equation (23).

[0070] (twenty three)

[0071] In the formula, This represents the derivative of the function n.

[0072] Observing equation (20), the derivative of the score segment function n(x) is calculated to be 1. The learning factor η is set to 0.05. By transforming equation (23), the dynamic adjustment amount Δw' of the weights from the intermediate layer to the output layer shown in equation (24) is obtained. j ,

[0073] , (twenty four)

[0074] Similarly, the dynamic adjustment amount Δw of the weights from the input layer to the intermediate layer ij As shown in equation (25),

[0075] (25)

[0076] In the formula, ε i The local gradient from the input layer to the intermediate layer is represented by equation (26).

[0077] (26)

[0078] In the formula, This represents the derivative of the m-function.

[0079] Another technical solution provided by this invention: a robust control system for a modular multilevel energy storage system virtual synchronous generator, comprising:

[0080] Framework modules are used to construct the energy storage VSG control system;

[0081] Module construction: Based on the VSG energy storage control system, establish a mathematical model of a modular multilevel energy storage system;

[0082] The current inner loop control module constructs a PBC-SMC current inner loop control structure with differential mode voltage as the control quantity.

[0083] The BP neural network online optimization coefficient module dynamically adjusts and optimizes network parameters based on the BP neural network, and outputs the optimized convergence coefficient.

[0084] This invention designs an overall control architecture for a VSG energy storage system and further invents a PBC-SMC current inner loop control structure to solve the problems of slow response speed, low steady-state accuracy, and poor robustness of traditional PI control. It discloses a robust control method and system for a virtual synchronous generator suitable for modular multilevel energy storage systems, which effectively improves the robustness of the energy storage system itself and its ability to quickly and actively support the power, frequency, and voltage of the power grid.

[0085] This invention completes the establishment of energy storage system topology, construction of converter mathematical model, and embedding of VSG control. It also invented a new PBC-SMC current inner loop control architecture and BP neural network online optimization coefficient module, and provides a robust control method for energy storage systems to improve the current inner loop tracking response performance. A calculation model for the bridge arm inductor voltage, an output current transfer function model, and an Euler-Lagrange mathematical model for the inner current loop of the converter were established. A PBC-SMC current inner loop control structure with differential mode voltage as the control variable was designed. A process for dynamic adjustment of BP neural network weights and optimization of network parameters and convergence coefficients were provided. This resulted in a robust control method and system for a modular multilevel energy storage system using a virtual synchronous generator. This method and system represent a novel robust control approach for energy storage systems, effectively improving the response speed, transient response performance, steady-state control accuracy, and robustness of the inner current loop. Consequently, it enhances the converter's ability to rapidly and actively support the power, frequency, and voltage of the grid connected to wind and solar power units, playing a role similar to that of a traditional synchronous machine inertia support, damped oscillation, and frequency and voltage stabilization. This provides a technical solution for improving the overall performance of the inner current loop by offering a robust control architecture and system design for a virtual synchronous generator in an energy storage system. Attached Figure Description

[0086] Figure 1 This is a schematic diagram of a single-phase topology for a modular multilevel energy storage system.

[0087] Figure 2 A schematic diagram for optimizing μ1 and μ2 in a BP neural network;

[0088] Figure 3 This is a diagram of the overall control architecture of a modular multilevel energy storage system. Detailed Implementation

[0089] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0090] A robust control method for a virtual synchronous generator in a modular multilevel energy storage system will be developed through specific steps including energy storage system topology research, model construction, and invention of a PBC-SMC current inner loop control structure, as detailed below:

[0091] First, this embodiment of the invention designs the topology of an MMC distributed energy storage system as the hardware circuit of the energy storage system. Distributed energy storage batteries are deployed as the basic architecture of the energy storage system. The energy storage batteries are the core components of the submodules (SMs), and DC / DC conversion modules are added to realize the forward and reverse transmission of electrical energy. The battery body uses lithium batteries, which have advantages such as fast charging and discharging speed and high power density. A single-phase topology diagram of the modular multilevel energy storage system is shown below. Figure 1 As shown. When a large number of submodules are added, the sinusoidal nature of the converter's output voltage and current will be higher, and the grid-side L-type / LC-type / LCL-type filter may not need to be installed; when a small number of submodules are added, the waveform of the converter's output voltage and current will be a sinusoidal stepped wave, and the hardware cost of the converter will be reduced. Figure 1 In the middle, v dc Indicates DC voltage; v p Indicates positive voltage; v n Represents the negative terminal voltage; R represents the bridge arm resistance; L represents the bridge arm inductance; R l L represents the equivalent circuit resistance. l R represents the equivalent line inductance. g Indicates the equivalent resistance on the grid side; L g Indicates the equivalent inductance on the grid side; v sa This represents the voltage of phase A of the power supply; v oa Indicates the A-phase output voltage on the grid side; i oa This indicates the A-phase output current on the grid side; v a Indicates the A-phase output voltage of the converter; i a This represents the A-phase output current of the converter; v au Indicates the A-phase voltage of the upper bridge arm; i au This represents the A-phase current of the upper bridge arm; v al Indicates the A-phase voltage of the lower bridge arm; i al This represents the A-phase current of the lower bridge arm.

[0092] This topology achieves the control objective of DC-side power and AC-side power interaction through lithium battery charging / discharging. The transmission relationship between AC-side power, DC-side power and lithium battery power is shown in Equation (1).

[0093] (1)

[0094] In the formula, P DC The value is zero because the DC voltage is zero, and the power transferred from the DC side to the lithium battery is zero; P AC This represents the power transmitted by the energy storage system to the AC side, i.e., the grid side; P Li, bat This indicates the output power of the lithium battery. Figure 1 The energy storage power equipment shown is mainly installed near photovoltaic projects, inside offshore wind power converter station platforms, inside urban power grid charging stations, and near urban power grid data service centers. It absorbs excess output from wind and solar new energy units, suppresses fluctuations in power, frequency, and voltage connected to the PCC, improves the safety, reliability, and stability of the new power system, promotes the absorption rate of new energy units, ensures uninterrupted power supply for important loads, and guarantees the quality of power supply.

[0095] Subsequently, a mathematical model of the modular multilevel energy storage system was constructed. The topology of the modular multilevel energy storage system was analyzed, and the inductor voltage equation of the bridge arm was established, as shown in equation (2).

[0096] (2)

[0097] In the formula, For bridge arm inductance; This refers to the voltage of the upper bridge arm; for; For the upper bridge arm voltage of each phase; For the bridge arm resistance; For the upper arm current of each phase; This represents the lower bridge arm current for each phase; This refers to the voltage of the lower bridge arm; This represents the lower bridge arm voltage for each phase;

[0098] Subtracting the second formula from the first formula in equation (2) gives equation (3).

[0099] (3)

[0100] Observing equation (2), we know that i ku -i kl Corresponding phase current i k , and v p +v n Approximately zero, let the differential mode voltage v kdifm as follows:

[0101] (4)

[0102] Analyze equation (3), list the equations for the three phases ABC, transform them, and substitute them into equation (4) to obtain equation (5).

[0103] (5)

[0104] In the formula, , , It is a three-phase current; , , It is a three-phase voltage; , , This is the three-phase differential mode voltage;

[0105] Based on the Park transformation formula, equation (5) is transformed from the three-phase stationary coordinate system to the two-phase synchronous rotating coordinate system to obtain equation (6).

[0106] (6)

[0107] In the formula, i d Indicates the d-axis quantity of the converter output current; i q This represents the q-axis quantity of the converter output current; v d This represents the d-axis quantity of the converter output voltage; v q This represents the q-axis quantity of the converter output voltage; v ddifm This represents the d-axis quantity of the differential-mode voltage; v qdifm This represents the q-axis quantity of the differential-mode voltage; This is the angular frequency of the power grid.

[0108] By transforming equation (6) using the Laplace transform principle, we obtain the transfer function formula that accurately describes the output current, as shown in equation (7).

[0109] (7)

[0110] In the formula, For the Laplace operator; For the Laplace transform of the d-axis current; Laplace transform of the q-axis current; Laplace transform of the d-axis voltage; Laplace transform of the q-axis voltage; Laplace transform of the d-axis differential-mode voltage; This is the Laplace transform of the q-axis differential-mode voltage.

[0111] Next, the PBC current inner loop control structure was invented. First, a current inner loop structure that injects damping into the current inner loop was proposed to regulate the system's energy supply, stabilize the system, and achieve the current tracking target. Simultaneously, considering the disturbances of the injected damping parameters and control parameters, a SMC current inner loop structure that considers these two disturbances was further proposed to improve control robustness. Then, the above two structures were combined into a new current inner loop control structure, and the design was completed.

[0112] Therefore, this invention proposes an Euler-Lagrange mathematical model to describe the dynamic characteristics of the inner loop voltage, as shown in equation (8).

[0113] (8)

[0114] For the inner-loop current control structure, the converter output current d and q-axis quantities i are selected. d i q Let x be the state variable vector. In this invention, the converter output current d and the q-axis reference value i are defined. dref i qref The reference vector for the state variables is x. ref And proposes the converter output current deviation vector x e The mathematical description method is shown in Equation (9).

[0115] (9)

[0116] To achieve the goal of PBC, control damping is injected into the mathematical model shown in equation (9), and the positive definite matrix composed of the injected control damping is... ,in , To control the positive definite coefficients of the damping, the injection control damping dissipation term is established as follows:

[0117] (10)

[0118] In the formula, R d The matrix represents the dissipation term of the injection control damping.

[0119] By combining equations (9) and (8) and substituting them into equation (10), the formula is transformed into a set of scalar equations using matrices and vectors. This results in a mathematical model of the PBC current inner loop structure with differential mode voltage as the control quantity, as shown in equation (10).

[0120] (11)

[0121] Similarly, for the design of the SMC current inner loop structure, the sliding surface of the SMC is designed first, as shown in Equation (12).

[0122] (12)

[0123] In the formula, For the d-axis sliding surface, It is the q-axis sliding surface;

[0124] The method of combining the sign function sgn(s) and the linear reaching law can effectively reduce chattering during the convergence of the sliding surface. The designed reaching rate is shown in Equation (13).

[0125] (13)

[0126] In the formula, λ1, μ1, λ2, and μ2 represent the convergence rate coefficients that are greater than zero, and their values ​​determine the convergence speed of the sliding surface and the magnitude of chattering.

[0127] By combining equations (7), (12), and (13), and transforming the equations into a scalar equation system, a mathematical model of the SMC current inner loop structure with differential mode voltage as the control quantity is constructed, as shown in equation (14).

[0128] (14)

[0129] Next, by combining equations (11) and (14), a mathematical model for the converter output current deviation vector is constructed, as shown in equation (15).

[0130] (15)

[0131] By combining equations (11) and (15), a mathematical model of the PBC-SMC current inner loop structure with differential mode voltage as the control quantity is constructed, as shown in equation (16).

[0132] (16)

[0133] Generally, the convergence coefficients μ1 and μ2 are set to larger values ​​to accelerate the convergence speed of the sliding surface and improve the transient response performance and speed of the system. However, if their values ​​are too large, the output current deviation vector will cross back and forth near the equilibrium point of the sliding surface, causing the system to chatter, which reduces the system stability and steady-state performance. At the same time, if the values ​​are too small, although the chattering is reduced, the overall response speed becomes slower and the settling time increases. To address this, this invention proposes an online optimization method for the convergence coefficients μ1 and μ2 using a back-propagation (BP) neural network to achieve faster response and reduced chatter, thereby improving the overall response performance of the system.

[0134] This invention inputs the output current deviation vector and its derivative into a BP neural network. The network outputs convergence coefficients μ1 and μ2. By optimizing μ1 and μ2 online, the converter's inner-loop current tracking response speed and robustness are improved, and chatter is reduced. A schematic diagram of the optimization of μ1 and μ2 is shown below. Figure 2 As shown.

[0135] analyze Figure 3 x1 and x2 are network inputs, w 11 ~w 23 The weights from the input layer to the intermediate layer correspond to the output of the intermediate layer, as shown in Equation (17).

[0136] (17)

[0137] In the formula, This represents the output of the j-th neuron in the hidden layer.

[0138] The m-function uses an S-shaped function to complete the nonlinear amplification task, transforming the data from −∞ to +∞ into the range of 0 to 1. The m-function is shown in equation (18).

[0139] (18)

[0140] w'1~w'3 are the weights from the intermediate layer to the output layer, corresponding to the output layer output, as shown in equation (19).

[0141] (19)

[0142] The n function uses a piecewise function to transform the data from -∞ to +∞ into the range of -1 to 1. The n function is shown in equation (20).

[0143] (20)

[0144] Propose the expected value μ of the convergence coefficient ref The mathematical description of the deviation from the online output μ is shown in Equation (21).

[0145] (twenty one)

[0146] Analyzing equation (21), the total deviation is constructed as shown in equation (22).

[0147] (twenty two)

[0148] The steepest descent method is used to optimize the adjustment of weights. To do this, the gradient of the total deviation e(k) with respect to the weights w'1~w'3 is first calculated, as shown in equation (23).

[0149] (twenty three)

[0150] In the formula, This represents the derivative of the function n.

[0151] Observing equation (20), the derivative of the score segment function n(x) is calculated to be 1. The learning factor η is set to 0.05. By transforming equation (23), the dynamic adjustment amount Δw' of the weights from the intermediate layer to the output layer shown in equation (24) is obtained. j .

[0152] (twenty four)

[0153] Similarly, the dynamic adjustment amount Δw of the weights from the input layer to the intermediate layer ij As shown in equation (25).

[0154] (25)

[0155] In the formula, ε i The local gradient from the input layer to the intermediate layer is represented by Equation (26).

[0156] (26)

[0157] In the formula, This represents the derivative of the m-function.

[0158] The dynamic adjustment of the weights from the input layer to the intermediate layer and from the intermediate layer to the output layer of the BP neural network and the optimization of the network parameters are completed by equations (17) to (26). Finally, the optimized convergence coefficient is output to realize online optimization of the coefficients to improve the performance of the converter current inner loop control.

[0159] Figure 3 This is a diagram of the overall control architecture of a modular multilevel energy storage system. Figure 3In the process, the signal measurement and coordinate transformation module first performs voltage and current signal acquisition and Park transformation, and the signal is sent to the power calculation module to output the real-time active power and reactive power signals of the converter. Then, the VSG control module, which simulates the external characteristics of the traditional synchronous machine, receives the power signal and outputs the electromotive force and phase angle signals through the internal active-frequency control module and reactive-voltage control module. After passing through the voltage vector synthesis module and Park transformation module, the voltage reference signal is output. The virtual impedance control module outputs the current reference signal. As shown in Equation (16), the PBC-SMC current inner loop control module receives the current reference signal and optimizes the approach rate coefficient μ online through the BP neural network optimization module, and outputs the differential mode voltage signal. In addition, the circulating current suppression control module will also output the compensation voltage signal. After the two voltage signals are superimposed, the trigger pulse is output through the nearest level approximation module, so that the converter can play the role of rapid active support of power, frequency and voltage in the grid, realize the target of robust control of the current inner loop, and complete the regulation task of suppressing grid power fluctuations.

[0160] This invention provides a robust control system for a virtual synchronous generator in a modular multilevel energy storage system, comprising:

[0161] Framework modules are used to construct the energy storage VSG control system;

[0162] Module construction: Based on the VSG energy storage control system, establish a mathematical model of a modular multilevel energy storage system;

[0163] The current inner loop control module constructs a PBC-SMC current inner loop control structure with differential mode voltage as the control quantity.

[0164] The BP neural network online optimization coefficient module dynamically adjusts and optimizes network parameters based on the BP neural network, and outputs the optimized convergence coefficient.

[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention without departing from the spirit and scope of the present invention.

Claims

1. A robust control method for a virtual synchronous generator in a modular multilevel energy storage system, characterized in that, Includes the following steps: (1) Establish an energy storage VSG control system; (2) Construct a mathematical model for a modular multilevel energy storage system; (3) Establish a passive control current inner loop architecture, taking into account the injection damping parameters and control parameter disturbances, establish a sliding mode control current inner loop structure considering the two disturbances, and combine the two structures to design a PBC-SMC current inner loop control structure with differential mode voltage as the control quantity. (4) Input the output current deviation vector and its derivative into the BP neural network, and perform dynamic adjustment and network parameter optimization through the BP neural network. Finally, output the optimized convergence coefficient to realize online optimization of the coefficient to improve the converter current inner loop control performance.

2. The robust control method for a virtual synchronous generator in a modular multilevel energy storage system according to claim 1, characterized in that, Step (1) includes: designing the topology of the MMC distributed energy storage system as the hardware circuit of the energy storage system; firstly, deploying distributed energy storage batteries as the basic architecture of the energy storage system; the energy storage batteries as the core components of the sub-modules; adding DC / DC conversion modules to realize the forward and reverse transmission of electrical energy; the battery body adopts lithium batteries; through the charging / discharging of lithium batteries, the control target of the interaction between DC side power and AC side power is realized; the transmission relationship between AC side power, DC side power, and lithium battery power is shown in equation (1). ， (1) In the formula, P DC The value is zero because the DC voltage is zero, and the power transferred from the DC side to the lithium battery is zero; P AC This represents the power transmitted by the energy storage system to the AC side, i.e., the grid side; P Li, bat This indicates the output power of the lithium battery.

3. The robust control method for a virtual synchronous generator in a modular multilevel energy storage system according to claim 1, characterized in that, Step (2) includes: The inductor voltage equation of the bridge arm was established, as shown in equation (2). ， (2) In the formula, For bridge arm inductance; This refers to the voltage of the upper bridge arm; for; For the upper bridge arm voltage of each phase; For the bridge arm resistance; For the upper arm current of each phase; This represents the lower bridge arm current for each phase; This refers to the voltage of the lower bridge arm; This represents the lower bridge arm voltage for each phase; Subtracting the second formula from the first formula in equation (2) gives equation (3). ， (3) Observing equation (2), we know that i ku -i kl Corresponding phase current i k , and v p +v n Approximately zero, let the differential mode voltage v kdifm as follows: ， (4) Analyze equation (3), list the equations for the three phases ABC, transform them, and substitute them into equation (4) to obtain equation (5). ， (5) In the formula, , , It is a three-phase current; , , It is a three-phase voltage; , , This is the three-phase differential mode voltage; Based on the Park transformation formula, equation (5) is transformed from the three-phase stationary coordinate system to the two-phase synchronous rotating coordinate system, resulting in equation (6). ， (6) In the formula, i d Indicates the d-axis quantity of the converter output current; i q This represents the q-axis quantity of the converter output current; v d This represents the d-axis quantity of the converter output voltage; v q This represents the q-axis quantity of the converter output voltage; v ddifm This represents the d-axis quantity of the differential-mode voltage; v qdifm This represents the q-axis quantity of the differential-mode voltage; The angular frequency of the power grid; By transforming equation (6) using the Laplace transform principle, we obtain the transfer function formula that accurately describes the output current, as shown in equation (7). ， (7) In the formula, For the Laplace operator; For the Laplace transform of the d-axis current; Laplace transform of the q-axis current; Laplace transform of the d-axis voltage; Laplace transform of the q-axis voltage; Laplace transform of the d-axis differential-mode voltage; This is the Laplace transform of the q-axis differential-mode voltage.

4. The robust control method for a virtual synchronous generator in a modular multilevel energy storage system according to claim 1, characterized in that, The establishment of the passive control current inner loop architecture in step (3) includes: An Euler-Lagrange mathematical model describing the dynamic characteristics of the inner current loop voltage is proposed, as shown in equation (8). (8) For the current inner loop control structure, the converter output current d and q-axis quantity i are selected. d i q Let x be the state variable vector, and define the converter output current d and the q-axis reference value i. dref i qref The reference vector for the state variables is x. ref And proposes the converter output current deviation vector x e The mathematical description method is shown in equation (9). ， (9) To achieve the goal of PBC, control damping is injected into the mathematical model shown in equation (9), and the positive definite matrix composed of the injected control damping is... ,in , To control the positive definite coefficients of the damping, the injection control damping dissipation term is established as follows: ， (10) In the formula, R d The matrix representing the injection control damping dissipation term. By combining equations (9) and (8) and substituting them into equation (10), the formula is transformed into a scalar equation system using matrices and vectors. This constructs a mathematical model of the PBC current inner loop structure with differential-mode voltage as the control variable, as shown in equation (11). ， (11)。 5. The robust control method for a virtual synchronous generator in a modular multilevel energy storage system according to claim 4, characterized in that, The sliding mode control current inner loop structure in step (3) includes: For the design of the SMC current inner loop structure, the sliding surface of the SMC is first designed, as shown in equation (12). ， (12) In the formula, For the d-axis sliding surface, For the q-axis sliding surface, By combining the sign function sgn(s) and the linear reaching law, the chattering that occurs during the convergence of the sliding surface is effectively reduced. The designed reaching rate is shown in equation (13). (13) In the formula, λ1, μ1, λ2, and μ2 represent the approach rate coefficients that are greater than zero, and their values ​​determine the convergence speed of the sliding surface and the magnitude of chattering. By combining equations (7), (12), and (13), and transforming the formula into a scalar equation system, a mathematical model of the SMC current inner loop structure with differential mode voltage as the control quantity is constructed, as shown in equation (14). ， (14)。 6. The robust control method for a virtual synchronous generator in a modular multilevel energy storage system according to claim 5, characterized in that, In step (3) Combining equations (11) and (14), a mathematical model for the converter output current deviation vector is constructed, as shown in equation (15). ， (15) By combining equations (11) and (15), a mathematical model of the PBC-SMC current inner loop structure with differential mode voltage as the control quantity is constructed, as shown in equation (16). ， （16）。 7. The robust control method for a virtual synchronous generator in a modular multilevel energy storage system according to claim 1, characterized in that, In step (4), the output current deviation vector and its derivative are input into the BP neural network. The network outputs the convergence coefficients μ1 and μ2. By optimizing μ1 and μ2 online, the speed and robustness of the converter's inner current tracking response are improved, and chatter is reduced. x1 and x2 are network inputs, w 11 ~w 23 The weights from the input layer to the intermediate layer correspond to the output of the intermediate layer, as shown in equation (17). ， (17) In the formula, This represents the output of the j-th neuron in the hidden layer. The m-function uses a sigmoid function to perform the nonlinear amplification task, transforming the data from −∞ to +∞ into the range of 0 to 1. The m-function is shown in equation (18). ， (18) w'1~w'3 are the weights from the intermediate layer to the output layer, corresponding to the output layer output, as shown in equation (19). ， (19) The n function uses a piecewise function to transform the data from −∞ to +∞ into the range −1 to 1. The n function is shown in equation (20). ， (20) Propose the expected value μ of the convergence coefficient ref The mathematical description of the deviation from the online output μ is shown in Equation (21). ， (21) Analyzing equation (21), the total deviation is constructed as shown in equation (22). ， (22) The steepest descent method is used to optimize the adjustment of the weights. First, the gradient of the total deviation e(k) with respect to the weights w'1~w'3 is calculated, as shown in equation (23). ， （23） In the formula, This represents the derivative of the function n. Observing equation (20), the derivative of the score segment function n(x) is calculated to be 1. The learning factor η is set to 0.

05. By transforming equation (23), the dynamic adjustment amount Δw' of the weights from the intermediate layer to the output layer shown in equation (24) is obtained. j , ， (24) Similarly, the dynamic adjustment amount Δw of the weights from the input layer to the intermediate layer ij As shown in equation (25), ， (25) In the formula, ε i The local gradient from the input layer to the intermediate layer is represented by equation (26). ， (26) In the formula, This represents the derivative of the m-function.

8. A robust control system for a virtual synchronous generator in a modular multilevel energy storage system, applied to the control method described in any one of claims 1-7, characterized in that, include: Framework modules are used to construct the energy storage VSG control system; Modules are constructed based on the VSG energy storage control system to establish a mathematical model of a modular multilevel energy storage system. The current inner loop control module constructs a PBC-SMC current inner loop control structure with differential mode voltage as the control quantity. The BP neural network online optimization coefficient module dynamically adjusts and optimizes network parameters based on the BP neural network, and outputs the optimized convergence coefficient.

Images