A frequency modulation method based on light storage VSG parameter self-adaption and model prediction cooperation

Abstract

The application discloses a frequency modulation method of photovoltaic energy storage VSG parameter self-adaption and model prediction cooperation. The method firstly calculates virtual rotational inertia on line according to frequency deviation and its change rate by using RBF neural network, and calculates matched virtual damping coefficient based on constant damping ratio linkage formula; then, the inertia and damping parameters updated in real time are injected into the coefficient matrix of model prediction control as time-varying parameters to construct a linear time-varying prediction model; finally, the optimal active power compensation value is solved by using the model through rolling optimization, and the value is superimposed into the VSG power reference instruction. Through the deep cooperation of RBF parameter self-adaption and MPC active power compensation, the application solves the problems of traditional VSG response lag and model mismatch, and significantly improves the frequency stability and dynamic response speed of the photovoltaic energy storage system under complex disturbance.

Description

Technical Field

[0001] This invention relates to the field of microgrid control, and more specifically, to a frequency regulation method based on the coordinated adaptation of photovoltaic-storage VSG parameters and model prediction. Background Technology

[0002] With the continuous increase in the proportion of new energy power generation, the large-scale grid connection of power electronic devices, represented by photovoltaics, has led to the power system exhibiting characteristics of "low inertia and weak damping." Traditional inverters lack the inherent mechanical inertia of synchronous generators, making it difficult to effectively support grid frequency stability when facing load disturbances or power surges. Virtual synchronous generator (VSG) technology imparts external characteristics to inverters by simulating rotor motion equations, but traditional fixed-parameter VSGs have a contradiction between dynamic response speed and steady-state performance, making it difficult to meet the control requirements of different disturbance stages. Existing improvement schemes, such as parameter adaptive techniques based on Bang-Bang control, fuzzy logic, or reinforcement learning, while improving flexibility to some extent, often suffer from drawbacks such as large algorithm computation, reliance on expert experience, or neglecting damping linkage when only adjusting inertia, leading to system extreme point instability. More importantly, the above adaptive strategies are mostly passive parameter adjustments, exhibiting response lag and failing to quickly compensate for the active power deficit at the initial stage of a disturbance. While model predictive control (MPC) possesses excellent feedforward compensation capabilities, conventional MPC typically relies on fixed model predictions and cannot adapt to model mismatch issues caused by real-time changes in VSG parameters, thus limiting its application in adaptive systems. Therefore, there is an urgent need for a control strategy that can achieve deep synergy between adaptive parameter adjustment and active power compensation. Summary of the Invention

[0003] Technical problem to be solved: The present invention aims to solve the technical problems existing in the background art.

[0004] Technical Solution: To solve the above-mentioned technical problems, this invention proposes a frequency regulation method based on the coordinated adaptation of photovoltaic-storage VSG parameters and model prediction. The method is applied to a VSG system including a photovoltaic unit, an energy storage unit, and a grid-connected inverter. The method includes the following steps: Step S1: Real-time acquisition of grid voltage, grid current, and output angular frequency of the VSG system; Step S2: Calculation of the frequency deviation between the output angular frequency and the rated angular frequency, and the rate of change of the frequency deviation; Step S3: Normalization of the frequency deviation and the rate of change of the frequency deviation followed by inputting them into a radial basis function neural network, which calculates the current virtual moment of inertia value online; Step S4: Based on the virtual moment of inertia value, using a preset... The damping linkage formula is used to calculate the virtual damping coefficient value that matches the current virtual moment of inertia. Step S5: Construct the model predictive control model of the VSG system, and use the virtual moment of inertia value obtained in step S3 and the virtual damping coefficient value obtained in step S4 as time-varying parameters to update the system coefficient matrix of the model predictive control model in real time. Step S6: Using the model predictive control model with updated parameters, based on the current system state, solve for the optimal active power compensation value at future times through rolling optimization. Step S7: Superimpose the optimal active power compensation value into the active power reference command of the VSG, and input the virtual moment of inertia value and the virtual damping coefficient value into the VSG control loop to generate an inverter drive signal to achieve grid-connected frequency regulation.

[0005] Furthermore, in step S3, the calculation process of the RBF neural network specifically includes: constructing a three-layer neural network structure comprising an input layer, a hidden layer, and an output layer; the input layer receiving the normalized frequency deviation. and the normalized rate of change of the frequency deviation The hidden layer uses the Gaussian function to perform a nonlinear mapping on the input vector, and its output... Represented as:

[0006]

[0007] in, For the input vector, For the first The center vector of each hidden layer node The base width parameter is used; the output layer outputs the virtual moment of inertia through weighted summation and activation function processing. The activation function is the Sigmoid function, which is used to limit the output to a preset range of inertia values. Inside.

[0008] Furthermore, the RBF neural network also includes an online weight update step: defining a performance evaluation function. The square of the system frequency deviation: Gradient descent with momentum term is used to weight the output layer. Real-time correction is performed, and the correction formula is as follows:

[0009]

[0010] in, For learning rate, The inertia coefficient, This represents the weight adjustment amount from the previous time step, where k represents the time step.

[0011] Furthermore, the preset damping linkage formula in step S4 is established based on keeping the damping ratio of the second-order oscillation model of the VSG system constant, and the specific formula is as follows:

[0012]

[0013] in, This is the virtual damping coefficient. , This represents the change in active power. This represents the initial angular frequency.

[0014] Furthermore, in step S5, constructing the model predictive control model specifically includes: establishing a discretized state-space equation based on the VSG rotor motion equation:

[0015]

[0016] in, ; ; , Sampling time, This represents the damping coefficient, which is adaptively output by the neural network. Indicates the moment of inertia. This represents the damping coefficient, which is adaptively output by the neural network. The input power of the VSG. Output power for the virtual synchronizer. This represents the angular frequency deviation at time k+1.

[0017] Furthermore, in step S6, the rolling optimization solution specifically includes: establishing an objective function that includes a frequency deviation penalty term and a control quantity change penalty term:

[0018]

[0019] In the formula: ; ; ,Will Set as an intermediate variable, representing the deviation between the desired frequency trajectory and the system's natural response. The three-step model predictive control output at time k is defined; constraints are set, including the VSG output frequency constraint range and the energy storage unit charging and discharging power constraint range; the objective function is transformed into a quadratic programming problem, the optimal control sequence is obtained by solving the problem, and the first element in the sequence is taken as the optimal active power compensation value at the current time.

[0020] Furthermore, in step S7, the mathematical model of the VSG control loop is:

[0021]

[0022] in, This is the initial active power reference value. The optimal active power compensation value obtained in step S6. This refers to the actual output active power. The rated angular frequency, This is the actual output angular frequency.

[0023] Beneficial effects

[0024] Compared with existing technologies, this invention first utilizes the nonlinear mapping capability of the RBF neural network to perceive frequency disturbances in real time, dynamically adjusts the virtual inertia, and synchronously corrects the damping coefficient using a constant damping ratio linkage mechanism. This "inertia-damping" dual-parameter linkage ensures that the system's closed-loop poles are always constrained in the stable underdamped region, effectively solving the power oscillation problem caused by adjusting only the inertia. More importantly, this invention establishes a closed-loop collaborative path of "parameter-driven model update," dynamically injecting the real-time inertia and damping parameters output by the RBF into the state-space model of the MPC, transforming the MPC prediction model from a linear time-invariant model to a linear time-varying model that accurately reflects the current physical characteristics of the system. This mechanism eliminates model prediction mismatch errors caused by adaptive parameter adjustment, enabling the MPC to predict future frequency trajectories based on an accurate real-time model and calculate the optimal active power compensation amount to be directly superimposed on the power command. This organic combination of "MPC feedforward active energy compensation" and "RBF feedback parameter dynamic optimization" not only utilizes active power compensation to quickly suppress the rate of frequency change during disturbances, but also utilizes parameter optimization to improve the dynamic characteristics of the recovery process, thereby achieving rapid frequency response and strong robustness of the photovoltaic energy storage system under complex operating conditions. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a schematic diagram of the topology of the photovoltaic-storage VSG control system provided in an embodiment of the present invention;

[0027] Figure 2 This is a block diagram of the traditional VSG control principle in existing technology;

[0028] Figure 3 This is a schematic diagram of the RBF neural network structure used for adaptive parameter adjustment in an embodiment of the present invention;

[0029] Figure 4 This is a flowchart of the virtual inertia and damping adaptive control strategy based on RBF neural network provided in this embodiment of the invention;

[0030] Figure 5 This is the overall logic block diagram of the RBF adaptive parameter adjustment and MPC collaborative control method provided in the embodiments of the present invention;

[0031] Figure 6 This is a schematic diagram illustrating the specific principle of RBF and MPC collaborative control in an embodiment of the present invention;

[0032] Figure 7 The following are simulation results of the photovoltaic power output fluctuation scenario in the embodiment of the present invention. (a) is a waveform diagram of the output power of photovoltaic, energy storage and inverter, (b) is a waveform diagram of the change of energy storage state of charge (SOC), and (c) is a waveform diagram of grid-side three-phase current.

[0033] Figure 8 The following are simulation comparison results of the embodiments of the present invention and the comparison strategy under the load step in the islanded operation mode. Among them, (a) is the VSG frequency response comparison curve, and (b) is the compensation active power comparison curve.

[0034] Figure 9 These are simulation comparison results of the embodiments of the present invention and the comparison strategy under active power disturbance in grid-connected operation mode. Among them, (a) is a comparison diagram of the VSG output angular frequency response, and (b) is a comparison diagram of the VSG output active power response.

[0035] Figure 10 The above diagram shows the frequency response simulation results of the embodiments of the present invention and the comparison strategy under power grid frequency disturbances. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0037] Example 1:

[0038] 1. Photovoltaic-Storage VSG Control and Mathematical Modeling

[0039] 1.1 Photovoltaic Storage VSG System

[0040] The photovoltaic-storage virtual synchronous power generation system constructed in this paper is as follows: Figure 1 As shown, the system consists of photovoltaic (PV) units, energy storage modules, and a grid-side inverter. The PV front-end is equipped with a Boost DC-DC converter to achieve Maximum Power Point Tracking (MPPT). The energy storage system is connected to the DC bus via a bidirectional Buck-Boost converter, enabling flexible charge and discharge control while maintaining bus voltage stability through power regulation. The back-end of the system incorporates a voltage source inverter and a VSG control strategy to achieve dynamic regulation of grid voltage and frequency and grid-connected operation. Considering the volatility of PV power generation, the system fully utilizes the energy storage units to provide energy support on the DC side, achieving a dynamic balance between front-end power generation fluctuations and back-end output power. VSG control adjusts active power output based on system frequency deviation and power commands, thereby improving the system's dynamic response and operational stability. Figure 1 middle: and These are the filter resistor and the filter inductor, respectively. For filtering capacitors; and To distinguish between line inductance and line resistance; This refers to the inverter output current. and The output current and output voltage of the virtual synchronous generator.

[0041] 1.2 Traditional VSG Mathematical Model

[0042] The overall operation mode of VSG control simulates the mechanical motion equation and primary frequency regulation characteristics of a traditional synchronous generator. Based on this principle, the relationship between the inverter's output power and angular frequency is shown in equation (1).

[0043]

[0044] In the formula: J is the moment of inertia coefficient of VSG; and These are the actual angular velocity and rated angular velocity corresponding to the VSG virtual rotor, respectively; D is the virtual damping coefficient. and These are the inverter's mechanical power and actual output power, respectively.

[0045] The VSG control structure introduces an active power-frequency droop adjustment element, and its corresponding relationship can be expressed by equation (2):

[0046]

[0047] Reactive power-voltage regulation is accomplished by the excitation regulator in the VSG, and its corresponding reactive power-voltage droop characteristic can be described by the following equation:

[0048]

[0049] For ease of analysis, assume that the response of the voltage and current inner loop and the PWM stage is fast enough, and can be considered to be completed rapidly relative to frequency control. Then:

[0050]

[0051] Traditional VSG control block diagram as follows Figure 2 As shown, and These are reference values ​​for active and reactive power, respectively. This is the measured value of angular frequency; This is the measured value of reactive power; , These are the active power-frequency droop coefficient and the reactive power-voltage droop coefficient, respectively; E, U, These represent the single-phase output voltage of the inverter, the effective phase voltage of the grid, and the reference voltage, respectively; K is the integral coefficient of the excitation regulator.

[0052] according to Figure 2 The model shown yields the transfer function of the active power element in the VSG as follows:

[0053]

[0054] In the formula: Defined as the synchronization voltage coefficient. For outputting small signal quantities of active power, Small-signal disturbances to the active power reference value, where s is a complex frequency variable.

[0055] The natural oscillation angular frequency can be obtained through equation (5). Damping ratio The expressions for these two parameters are:

[0056]

[0057] In damping ratio At that time, with As the damping ratio increases, both the overshoot and dynamic adjustment time of the active power decrease, thereby reducing the power fluctuation impact on the energy storage device during the dynamic adjustment phase and improving the system's dynamic coordination capability and component lifespan. Therefore, the designed system is typically an underdamped system with a damping ratio range of (0,1).

[0058] In traditional VSG control, inertia and damping parameters are usually set to fixed values, making it difficult to effectively suppress frequency fluctuations under dynamic conditions such as load disturbances, thus affecting system response speed and stability. By introducing an online adaptive parameter mechanism, the virtual inertia and damping coefficient can be dynamically adjusted according to the system state, significantly improving frequency regulation performance and system robustness.

[0059] 1.3 Range of values ​​for J and D

[0060] According to equation (1), the motion equation of the VSG rotor can be derived as follows:

[0061]

[0062] According to equation (7), when there is a shortage of active power... Damping coefficient D, active droop coefficient and frequency deviation When constant, the virtual moment of inertia J and the rate of change of angular frequency of the VSG Inversely proportional to the rotational inertia, when the moment of inertia is small, the rate of change of the rotor angular frequency increases, and the system stability decreases. Conversely, excessive inertia causes slow changes in angular velocity, affecting dynamic response performance. The inertia adjustment rules proposed in the literature are shown in Table 1: when the angular frequency deviates significantly from the rated value, the inertia can be appropriately increased to suppress the frequency deviation; while when the angular frequency approaches the rated value, decreasing the inertia helps to accelerate the frequency recovery speed.

[0063] Table 1. Rules for selecting moment of inertia under different conditions

[0064]

[0065] Similarly, the damping coefficient D affects the angular frequency by the following: when D is too small, the system has a greater risk of overshoot. While increasing the damping can improve convergence speed, setting the damping parameter too large may cause a slow response. Therefore, when there is an active power deficit... Moment of inertia J, rate of change of angular frequency and active droop coefficient When D remains constant, the larger D is, The smaller.

[0066] The initial inertia and damping coefficient settings need to strike a balance between the system's dynamic performance and stability, with a focus on the stability margin of the active control element. Since the gain margin of a second-order system is always greater than zero, according to automatic control theory, to achieve a faster dynamic response and better system stability, the distribution of closed-loop poles needs to be reasonably constrained. Considering the above factors, the following set of parameter design relationships can be established:

[0067]

[0068] In this article, when At that time, one can obtain The closed-loop poles of the system are The open-loop transfer function for active power is:

[0069]

[0070] This represents the open-loop transfer function of the active power control loop. This represents the complex frequency domain characteristic of the open-loop transfer function of the active power control loop at the cutoff frequency. For complex frequency variables

[0071] Using the set cutoff frequency The range of values ​​for the damping coefficient D can be limited, which can be obtained from equation (9):

[0072]

[0073] Since the expression inside the radical in equation (10) is always greater than zero, we have:

[0074]

[0075] As shown in equation (11), there is an upper limit constraint on the cutoff frequency of active power. This upper limit is typically set to within 10% of twice the power frequency. Therefore, taking the maximum cutoff frequency as 10 Hz, the minimum damping value can be obtained. . This is the maximum cutoff frequency.

[0076] According to EN50438 standard, when distributed power sources are connected to the grid, their permissible frequency fluctuation range is 49~51Hz, and their active power output should vary between 40% and 100% of the rated power. With the inverter capacity set at 50 kW, we can obtain:

[0077]

[0078] in, The calculated range of J is [0.05, 0.5], and the range of D is [11.5, 25.3].

[0079] 2. RBF parameter adaptive control strategy

[0080] 2.1 Adaptive Adjustment of Virtual Inertia Based on RBF

[0081] Moment of inertia and damping coefficient are key physical parameters affecting the dynamic response of a system. Analyzing their control laws helps to gain a deeper understanding of the system's dynamic characteristics. Introducing an RBF neural network to achieve adaptive parameter adjustment can effectively simulate complex nonlinear systems, thereby improving model accuracy.

[0082] Figure 3 The diagram shows the structure of the RBF neural network in this paper. The three-layer structure consists of an input layer, an output layer, and a hidden layer. i, j, and l represent the input layer node, the hidden layer node, and the output layer node, respectively, which are 2, 5, and 1. The two inputs of the input layer are the angular frequency and the rate of change of the angular frequency of the VSG, and the output layer is the corresponding VSG moment of inertia J.

[0083] The input layer output, hidden layer input, and hidden layer output are respectively:

[0084]

[0085]

[0086]

[0087] In the formula: , To distinguish between different layers, the input layer, hidden layer, and output layer are represented by superscripts (1), (2), and (3), respectively. For hidden layer input, This is the output of the hidden layer.

[0088] The Gaussian function, the input to the output layer, and the output of the neural network are as follows:

[0089]

[0090]

[0091]

[0092] Initialize the network parameters, and set the activation function of the output layer as follows:

[0093]

[0094] Where u is the maximum value of the inertia coefficient, For input layer input, It is an exponential function.

[0095] The evaluation function for this neural network is:

[0096]

[0097] The weights are adjusted using gradient descent, and to improve convergence speed, an inertia term is added to the formula, i.e.:

[0098]

[0099] In the formula: For learning rate, The inertia coefficient.

[0100] 2.2 Inertia-driven damping adaptive linkage control

[0101] Existing adaptive inertia control methods neglect the crucial role of the damping coefficient in stabilizing active power output and suppressing frequency fluctuations. Properly configuring the damping ratio is key to ensuring the system's dynamic response performance. The overall control flow described in this chapter is as follows: Figure 4 As shown.

[0102] First, select a set of initial values ​​within the allowable range of the virtual inertia J, and then calculate the initial angular frequency using the rotor motion equations. ; through performance evaluation functions Determine whether the neural network weights need to be updated. Then, use the RBF neural network to adaptively adjust the virtual inertia J, construct the hidden layer output using a Gaussian kernel function, and feed it into the nonlinear sigmoid activation function after weighting, so that J is always within its value range. Then, according to equation (6), under the condition that the damping ratio is fixed, make it adaptively adjust with the inertia, and determine whether the constraint conditions are met, thereby determining the inertia and damping values ​​at the next moment; feed the obtained virtual inertia and damping coefficient back to the VSG rotor motion equation to form a closed-loop control mechanism and continuously optimize the dynamic performance of the system.

[0103] 3 Control Strategy Based on MPC-VSG

[0104] To achieve rapid frequency recovery and dynamic stability optimization of the virtual synchronous machine under different disturbance stages, this paper adopts a dual-time-scale coordination mechanism in the control structure. In the initial stage of a disturbance, the frequency deviation changes rapidly. During this stage, MPC control dominates, and the optimal power compensation is calculated through rolling optimization to achieve rapid suppression of the frequency deviation and power balance. As the system gradually enters the transition and steady-state stages, the frequency change slows down. At this point, the RBF network begins to adaptively adjust the virtual inertia and damping coefficient according to the frequency deviation and its rate of change, thereby further improving the dynamic characteristics and stability of the system. Figure 5 The flowchart of the proposed VSG system, which combines neural network adaptive parameter tuning with model prediction control, is presented.

[0105] 3.1 Establishment of a Prediction Model Combining Adaptive Parameter Adjustment and MPC

[0106] First, a mathematical model of the VSG control method combining adaptive parameter tuning and MPC needs to be constructed. Then, the moment of inertia obtained from the RBF adaptive method in Chapter 2 is used. and damping coefficient Introducing the rotor motion equations and combining them with the frequency dynamic characteristics of the VSG, the system's state equations can be derived as follows:

[0107]

[0108] Let t be the output variable of the VSG prediction system. The input power of the VSG, For the output power of the virtual synchronizer, The damping coefficient is the adaptive output of the neural network. This represents the inertia coefficient after adaptive output from the neural network. Since the object of study is a first-order model, it can be discretized using the forward Euler method, i.e.:

[0109]

[0110] In the formula: For the sampling time, discretizing equation (23) yields:

[0111]

[0112] in, ; ; .

[0113] To facilitate the direct acquisition of the value of the angular frequency change, this paper sets the MPC output equation as follows: To improve prediction accuracy while ensuring computational efficiency, this paper adopts a three-step prediction method, and the prediction equation is as follows:

[0114]

[0115] In the formula: This is the output of the three-step MPC control at time k; , , The coefficient matrix has the following expressions:

[0116]

[0117] 3.2 Construction of the objective function

[0118] When the system frequency gradually deviates from the rated value, in order to suppress the rate of frequency change, the frequency increment and the change in control quantity are selected as optimization targets, and the overall performance index is minimized by constructing their weighted sum objective function.

[0119]

[0120] and These are the weighting coefficients for the change in VSG output frequency and the change in input power, respectively. and Let $\mathbf{k}$ represent the angular frequency deviation and the active power deviation at time $k$, respectively. Equation (27) can be converted into matrix form as follows:

[0121]

[0122] In the formula: ; ; ,Will It is set as an intermediate variable, representing the deviation between the desired frequency trajectory and the system's natural response.

[0123]

[0124] By substituting equation (29) into equation (28), the above MPC control problem can be transformed into The standard quadratic programming problem takes the form of, where As its independent variable.

[0125]

[0126] Frequency fluctuations must be limited to an acceptable range, and the constructed MPC optimization problem should include constraints, described as follows:

[0127]

[0128] in, and The frequency constraint of the system is given. Equation (31) is solved by quadratic programming to obtain the optimal control matrix. Therefore, the reference input power of the VSG is:

[0129]

[0130] RBF and MPC collaborative control strategy, such as Figure 6 As shown, adaptive adjustment of virtual inertia and damping coefficient has good online adjustment capability, and can dynamically correct VSG parameters according to frequency state, thereby improving the system performance under various operating conditions. However, this method mainly relies on feedback mechanism, and the response has a certain lag, making it difficult to effectively cope with rapid disturbances. In contrast, model predictive control has a forward-looking characteristic, which can predict future trends based on the current system state and perform active power compensation in advance, thereby enhancing the initiative and accuracy of frequency regulation. The combination of the two forms a complementary control structure: adaptive control improves the dynamic performance of the system, and MPC provides rapid compensation in the early stage of disturbance, thereby significantly enhancing the robustness and response performance of frequency control.

[0131] 4. Simulation Results and Analysis

[0132] To verify the effectiveness and superiority of the proposed control strategy, a photovoltaic energy storage VSG simulation model was built on the MATLAB / Simulink platform. The key parameters are shown in Table 2.

[0133] Table 2 Main Simulation Parameters of the System

[0134]

[0135] 4.1 Functional Verification of Control Strategy under Photovoltaic Output Fluctuation Scenarios

[0136] Under the grid-connected conditions of photovoltaic-storage-VSG, the initial simulation settings are as follows: active power setpoint is 10 kW, ambient temperature is 25℃. The initial irradiance is 1000 W / m², decreasing to 600 W / m² at 0.4 s, and increasing to 1500 W / m² at 0.7 s. Simulation results show that in the interval from 0.2 s to 0.4 s, the photovoltaic output is approximately equal to the active power demand, and the energy storage unit does not participate in charging or discharging. From 0.4 s to 0.7 s, the reduced irradiance leads to insufficient photovoltaic output, failing to meet the set active power reference value. At this time, the energy storage system enters a discharging state, cooperating with the photovoltaic system to supply power to the grid, causing the SOC to continuously decrease. From 0.7 s to 1 s, the photovoltaic output exceeds the active power reference command, the energy storage battery begins charging, storing energy as a reserve, and the SOC continuously increases. During this process, the grid-side current is stable, and the total harmonic distortion rate is 0.98%, meeting the requirement that the grid-connected current harmonic distortion rate is less than 2%. The simulation results are as follows. Figure 7 As shown.

[0137] 4.2 Verification of Control Effect in Isolated Situations

[0138] To verify the proposed method's ability to regulate frequency during load increases and decreases in an islanded system, the initial active power reference value was set to 35 kW. At 1.5 s, the system load decreased by 5 kW, and the active power reference value jumped to 30 kW. At 3 s, the system load increased by 5 kW, and the active power jumped to 35 kW. To verify the effectiveness of this scheme, its control performance was compared and analyzed with three other control strategies. Control strategy 1 uses a traditional VSG with fixed inertia and damping parameters; control strategy 2 is a control method combining MPC and VSG.

[22] Control strategy 3 uses an RBF neural network to adaptively adjust inertia and damping coefficient.

[19] Strategy 4 is the control scheme proposed in this paper. A comparison of frequency response curves and compensation power during islanding is provided. Figure 8 As shown.

[0139] When the load is removed after 1.5 s, the traditional fixed-parameter VSG relies on its droop characteristic to regulate the frequency, but the system frequency rises rapidly by about 2.65 rad / s. Strategy 2 moderates the rate of frequency change, with a maximum increment of about 1.25 rad / s. Strategy 3 greatly reduces the change in angular frequency, with the increase reduced to 0.99 rad / s. The control strategy proposed in this paper has a more stable frequency waveform, with the maximum offset reduced to 0.77 rad / s, which is about 71% less than the change amplitude of the traditional VSG.

[0140] Depend on Figure 8As shown in (b), strategies 2 and 4 performed active power compensation, with maximum compensation powers of 1.98 kW and 1.65 kW, respectively. The compensation power curves showed an opposite trend to the frequency curves. Since strategies 1 and 3 did not have an MPC module, their power compensation was 0 kW.

[0141] 4.3 Verification of Control Effect During Grid Connection

[0142] The following simulations will simulate two operating conditions: active power disturbance and grid frequency disturbance. First, the active power disturbance scenario will be simulated. When the photovoltaic-storage VSG is connected to the grid, the initial reference value of the system's active power is 10 kW. At 0.4 s, the active power suddenly increases by 20 kW and continues until 1 s. The simulation results of the VSG output angular frequency curve and active power are as follows: Figure 9 As shown.

[0143] Figure 9 (a) The effects of different control strategies on the system angular frequency under the same operating conditions were compared. Power imbalance in the system causes frequency fluctuations. To more clearly and intuitively demonstrate the control effects, the control performance indicators for different control methods are shown in Table 3.

[0144] Table 3 Performance Comparison of Different Control Methods under Active Disturbance

[0145]

[0146] Depend on Figure 9 (b) It can be seen that the maximum active power deviations are 1.59, 1.03, 0.39, and 0.015 kW, respectively, with overshoots of 7.95%, 5.15%, 1.95%, and 0.072%, respectively. The comparison of the above control strategies shows that Strategy 2 can slightly reduce power overshoot, and Strategy 3 can effectively reduce the rate of power change, but both suffer from small power fluctuations. Strategy 4 provides stronger inertia support when the active power command changes abruptly, effectively suppressing grid frequency drops and avoiding frequency oscillations. Addressing the limitations of Strategy 2 and Strategy 3, the power waveform of Strategy 4 is smoother than the previous three strategies and reaches steady state in almost the same amount of time.

[0147] Next, we verify the operating conditions under grid frequency disturbance. Active and reactive power remain constant. From 0.4 s to 0.9 s, a 0.5 s frequency fluctuation occurs, during which the grid frequency rises from the rated power frequency to 50.1 Hz. The simulation frequency response results are as follows: Figure 10 As shown.

[0148] from Figure 10As can be seen, the maximum frequency deviations of the system are 50.111 Hz, 50.106 Hz, 50.102 Hz, and 50.101 Hz, with recovery times of 0.208 s, 0.372 s, 0.258 s, and 0.104 s, respectively. When using the conventional fixed-parameter VSG strategy, the overshoot is large and the frequency fluctuation is significant. Although strategies 2 and 3 can effectively reduce the overshoot, their response times are slightly longer and their reactions are not sensitive enough. Considering all factors, strategy 4 is superior to the other three control methods in terms of frequency overshoot and transient settling time.

[0149] 5. Conclusion

[0150] To address the power overshoot and frequency oscillation issues in photovoltaic-storage systems, this paper proposes a comprehensive control strategy combining adaptive parameters and MPC, yielding the following conclusions:

[0151] (1) Apply RBF to VSG to adaptively adjust the moment of inertia. At the same time, based on a fixed damping ratio, the damping coefficient is adaptively adjusted as the moment of inertia changes, and input to the MPC controller to correct the rated input power of VSG in real time, thereby enhancing the stability of the system during operation.

[0152] (2) When the system is in islanded operation, the strategy proposed in this paper significantly reduces the maximum angular frequency deviation. The maximum angular frequency deviation is reduced by 1.88 rad / s, 0.48 rad / s and 0.22 rad / s respectively compared with the other three strategies.

[0153] (3) When the system is running in grid connection, when the active power is disturbed, the proposed strategy does not have obvious overshoot or oscillation in power and frequency. When the difference in transient recovery time is not large, the frequency and power response curves are smoother than the other three methods. In the face of grid frequency fluctuations, the proposed strategy can effectively track the grid frequency, with a transient recovery time of only 0.104 s, which is smaller than the frequency deviation generated by the other three methods and the recovery time is also shorter.

[0154] In summary, the embodiments of the present invention provide a solution with fundamental advantages over existing technologies in predicting power load for special events by introducing a series of non-obvious combinations of multi-source knowledge bases, deep metric learning, and dynamic weighted prior construction.

Claims

1. A frequency modulation method based on the coordinated adaptation of photovoltaic-storage VSG parameters and model prediction, characterized in that, The method is applied to a VSG system comprising a photovoltaic unit, an energy storage unit, and a grid-connected inverter. The method includes the following steps: Step S1: Real-time acquisition of grid voltage, grid current, and output angular frequency of the VSG system; Step S2: Calculation of the frequency deviation between the output angular frequency and the rated angular frequency, and the rate of change of the frequency deviation; Step S3: Normalization of the frequency deviation and the rate of change of the frequency deviation, followed by inputting them into a radial basis function neural network, which calculates the current virtual moment of inertia value online; Step S4: Based on the virtual moment of inertia value, using a preset damping linkage formula, calculation of a virtual moment of inertia matching the current virtual moment of inertia. Step S5: Construct the model predictive control model of the VSG system, and use the virtual moment of inertia value obtained in step S3 and the virtual damping coefficient value obtained in step S4 as time-varying parameters to update the system coefficient matrix of the model predictive control model in real time; Step S6: Using the model predictive control model with updated parameters, based on the current system state, solve for the optimal active power compensation value at future times through rolling optimization; Step S7: Superimpose the optimal active power compensation value into the active power reference command of the VSG, and input the virtual moment of inertia value and virtual damping coefficient value into the VSG control loop to generate an inverter drive signal to achieve grid-connected frequency regulation.

2. The method according to claim 1, characterized in that, In step S3, the calculation process of the RBF neural network specifically includes: constructing a three-layer neural network structure comprising an input layer, a hidden layer, and an output layer; the input layer receiving the normalized frequency deviation. and the normalized rate of change of the frequency deviation The hidden layer uses the Gaussian function to perform a nonlinear mapping on the input vector, and its output... Represented as: ， in, For the input vector, For the first The center vector of each hidden layer node The base width parameter is used; the output layer outputs the virtual rotational inertia through weighted summation and activation function processing. The activation function is the Sigmoid function, which is used to limit the output to a preset range of inertia values. Inside.

3. The method according to claim 2, characterized in that, The RBF neural network also includes an online weight update step: defining a performance evaluation function. The square of the system frequency deviation: Gradient descent with momentum term is used to weight the output layer. Real-time correction is performed, and the correction formula is as follows: ， in, For learning rate, The inertia coefficient, This represents the weight adjustment amount from the previous time step, where k represents the time step.

4. The method according to claim 1, characterized in that, The preset damping linkage formula in step S4 is based on maintaining a constant damping ratio in the second-order oscillation model of the VSG system. The specific formula is as follows: ， in, This is the virtual damping coefficient. , This represents the change in active power. This represents the initial angular frequency.

5. The method according to claim 1, characterized in that, In step S5, constructing the model predictive control model specifically includes: establishing a discretized state-space equation based on the VSG rotor motion equation. ， in, ; ; , Sampling time, Represents the moment of inertia. This represents the damping coefficient, which is adaptively output by the neural network. The input power of the VSG. Output power for the virtual synchronizer. This represents the angular frequency deviation at time k+1.

6. The method according to claim 5, characterized in that, In step S6, the rolling optimization solution specifically includes: establishing an objective function that includes a frequency deviation penalty term and a control quantity change penalty term. ， In the formula: ; ; ,Will Set as an intermediate variable, representing the deviation between the desired frequency trajectory and the system's natural response. The three-step model predictive control output at time k is represented; constraints are set, including the VSG output frequency constraint range and the energy storage unit charging and discharging power constraint range; the objective function is transformed into a quadratic programming problem, the optimal control sequence is obtained by solving it, and the first element in the sequence is taken as the optimal active power compensation value at the current time.

7. The method according to claim 1, characterized in that, In step S7, the mathematical model of the VSG control loop is: ， in, This is the initial active power reference value. The optimal active power compensation value obtained in step S6. This refers to the actual output active power. The rated angular frequency, This is the actual output angular frequency.

Images